New Multi-Threaded Pi Program - Faster than SuperPi and PiFast

Printable View

Show 100 post(s) from this thread on one page

01-02-2011, 08:40 PM
alpha754293

Quote:

Originally Posted by poke349

Thanks! Last time I was here was 13 years ago. So it's almost completely new to me. :ROTF:

Ordering food. If you ask me, that's pretty important. :rofl: It's hard to get around without food. :yepp:

My Cantonese is native, but not fluent. :ROTF: (It's my first language, but having been in the states all my life, English is obviously my strongest language.)

EDIT:
I wanna learn enough Japanese to go to Japan to meet Shigeru Kondo. From what I've seen/heard, few people in Japan outside of the tourist areas speak English.
Right now, all the Japanese I know is from Anime - which I know is not enough (nor appropriate) to use on the streets. :rofl:

My mom is kinda afraid of the whole middle east... But I also think it's a little over the top.

The last time that I was in Hong Kong was I think 1994/95. My mom keeps talking about wanting to take me there. *shrug* who knows. I personally, have little to no desire to go.

Cantonese is also technically my native tongue as well, but I grew up in Canada, so, like yourself, English is definitely my strongest language (to the point that there are quite a number of people that think I'm born here).

I ran one of the bigger runs but it came back to 3 errors that were fixed with ECC. So, I don't know. I could try to run them again and see what happens with it.
01-03-2011, 03:36 AM
poke349

Quote:

Originally Posted by alpha754293

The last time that I was in Hong Kong was I think 1994/95. My mom keeps talking about wanting to take me there. *shrug* who knows. I personally, have little to no desire to go.

Cantonese is also technically my native tongue as well, but I grew up in Canada, so, like yourself, English is definitely my strongest language (to the point that there are quite a number of people that think I'm born here).

Most people have no idea I can speak anything else... until I get a call from my mom or something... :rofl::rofl::rofl:
Then I make ears bleed with a horrific mix of English and Canto... :ROTF:

Quote:

I ran one of the bigger runs but it came back to 3 errors that were fixed with ECC. So, I don't know. I could try to run them again and see what happens with it.

Interesting, I rarely see cases where a run will still complete after multiple corrected errors. 1 is common. 2 - I've heard of but not seen myself. 3 - first time.
It usually BSODs or hits an uncorrectable error with that many errors. :rofl:
01-03-2011, 07:02 AM
alpha754293

Quote:

Originally Posted by poke349

Most people have no idea I can speak anything else... until I get a call from my mom or something... :rofl::rofl::rofl:
Then I make ears bleed with a horrific mix of English and Canto... :ROTF:

Interesting, I rarely see cases where a run will still complete after multiple corrected errors. 1 is common. 2 - I've heard of but not seen myself. 3 - first time.
It usually BSODs or hits an uncorrectable error with that many errors. :rofl:

Same thing. When asked if I can speak Chinese, my official answer is always "no". But if you hang around me long enough, you will find that that's NOT entirely true. I can, but at a very minimalistic level. Like...kindergarten level. And even then, most of the time, I would respond in English anyways. (It takes me a while to translate it since I think in English). Talking to my grandparents...OMG....GREAT FUN. LOL. :rofl:

Yayyyyy Chinglish!!! But it's not that surprising for people from Hong Kong though.

I'm re-running it now. Yea, I got a nice little message saying that it ran successfully, but it had 3 errors that it needed to fix itself. I don't have that logfile anymore (because I'm rerunning it), so hopefully I'll have the new results soon. (Maybe like...45 minutes to an hour or so). I want to see if it is going to do it again.

*edit*
25B passed on the second go-around, without error.

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2010 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): AMD Opteron(tm) Processor 6174 Logical Cores: 48 Physical Memory: 136,765,112,320 bytes ( 128 GB ) CPU Frequency: 2,200,048,722 Hz Program Version: 0.5.4 Build 9148 (fix 1) (x64 SSE3 - Windows ~ Kasumi) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 25,000,000,000 Hexadecimal Digits: 20,762,050,594 Threading Mode: 64 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 112 GB Start Date: Mon Jan 03 08:09:10 2011 End Date: Mon Jan 03 12:34:19 2011 Computation Time: 14,069.180 seconds Total Time: 15,908.900 seconds CPU Utilization: 3972.97 % Multi-core Efficiency: 82.77 % Last Digits: 2448547079 5329693979 7145627081 9204187454 9483487803 : 24,999,999,950 1309759846 5364560010 7388984278 8403481193 9913806533 : 25,000,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 1c42b250a604b3070246bf06edf0d17bebd53bd6a1781fdd11ed4c47abd19822
01-08-2011, 07:10 AM
poke349

Alright, I'm back in the states - jetlagged like hell. :rofl: I'll get these updated in a bit.
01-09-2011, 03:40 PM
poke349

Charts updated. :D

Just making sure before I place my order. The following are compatible right?

ASUS P8P67 PRO LGA 1155 Intel P67 SATA 6Gb/s USB 3.0 Intel Motherboard
http://www.newegg.com/Product/Produc...82E16813131682

Intel Core i7-2600K Sandy Bridge 3.4GHz (3.8GHz Turbo Boost) LGA 1155 95W Quad-Core Desktop Processor BX80623I72600K
http://www.newegg.com/Product/Produc...82E16819115070

G.SKILL Ripjaws Series 16GB (4 x 4GB) 240-Pin DDR3 SDRAM DDR3 1333 (PC3 10666) Desktop Memory Model F3-10666CL9Q-16GBRL
http://www.newegg.com/Product/Produc...82E16820231312

CORSAIR CWCH50-1 High Performance CPU Cooler
http://www.newegg.com/Product/Produc...82E16835181010

I'll be throwing in my spare GTS 250 and my spare 1000W Corsair PS. (It's probably total overkill, so I might get a lower-end PS and keep the 1000W spare for open-air testing of all my other junk.)

Nothing spectacular. Just enough to OC and play with AVX. I won't be doing any "serious" builds for a while.
Other than that, I'll be using it mainly as a gigabit/esata/usb3 fileserver.
01-13-2011, 12:50 PM
skycrane

hey Poke, its almost christmas in january time :) what i need to know is what do you need from me?

im working on making a beauwolf cluster out of my tyans with a 10gbit inifiniband interconnect :)

so im going to send you my arima quad socket if thats ok.... its got 2 8220 dual core optis in it, i can get 2 more if you need all 4. i dont think i have any ram for it, ill have to find a vid card for it, they are a bit finickey on the vid.
do you have an extra psu? 500w should be more than enough, the WCG guys are running it with 4 8347 quad core, so 4 duals should pull less then that.

last thing, do you need a OS in it ive got win server 03 or umbutu if you want it.

let me know what you need for it and ill get it together for ya.
01-13-2011, 07:18 PM
poke349

Quote:

Originally Posted by skycrane

hey Poke, its almost christmas in january time :) what i need to know is what do you need from me?

im working on making a beauwolf cluster out of my tyans with a 10gbit inifiniband interconnect :)

so im going to send you my arima quad socket if thats ok.... its got 2 8220 dual core optis in it, i can get 2 more if you need all 4. i dont think i have any ram for it, ill have to find a vid card for it, they are a bit finickey on the vid.
do you have an extra psu? 500w should be more than enough, the WCG guys are running it with 4 8347 quad core, so 4 duals should pull less then that.

last thing, do you need a OS in it ive got win server 03 or umbutu if you want it.

let me know what you need for it and ill get it together for ya.

Awesome. :D I think all I need is the mobo and any 4 identical CPUs. I'll take care of everything else. (I shouldn't be asking for more anyway. :rofl:)

The extra stuff I have lying around are:
CM 932 HAF - Currently empty. My Harpertown rig used to be in it before I moved it back to California. So it's furniture right now. :rofl: The quad is going in this case.
GTX 9800+ - My Harpertown rig's old card. If this is compatible, I'll throw it in the quad.
GTS 250 - This is gonna go in my SB rig.
Corsair 1000W - Unused, currently serving as my parts tester.

I'll be stopping at Frys on my way back to Urbana to pickup anything else I might need. They pretty much have everything that isn't server-oriented.
I'll be getting a small case and a weaker PSU for my SB rig. (I won't be running SLI/Xfire nor will be clocking above 5 GHz... so the 1000W Corsair is overkill.) I can easily pickup another PSU for the quad.

I'm not sure if Frys has socket F heatsinks. And I bet they don't have DDR2 ECC ram. I'll probably grab those off the egg or any old servers that whomever might be throwing out.

So I'm guessing I'll need 8 sticks to get it running at full speed? (4 sockets x dual-channel)
I'm probably gonna need at least 16GB of ram to be able to do any serious NUMA/MPI programming on it. (Man, I wish they were compatible with my Harpertown rig! That thing is loaded. :ROTF:)
But if you happen to have any extra lying around that you no longer want... :D

I'm also good with the OS. Is your Win Server already "activated" or "locked" to the rig?
01-16-2011, 08:53 PM
poke349

Back on campus + two long days of unpacking/organizing/building/debugging...

I spent most of today wrestling with my SB rig... Sometimes it posts 16GB, sometimes only 8GB... Everything is at stock. :shrug:

I'm guessing it probably has something to do with the auto-tuning settings...
Disabling the auto-tuner and manually setting everything to stock did improve the reliability a bit.
It's still too early... probably gonna need to wait for a BIOS update.

My Harpertown rig with a new mobo... (click to enlarge)
http://www.numberworld.org/screenie_...0776_small.jpg http://www.numberworld.org/screenie_...0777_small.jpg

The SB rig... (click to enlarge)
http://www.numberworld.org/screenie_...0781_small.jpg http://www.numberworld.org/screenie_...0787_small.jpg

Anyways... Whenever it does manage post and boot with 16GB, its rock-stable. So I'm sure it's just a BIOS issue and not a faulty mobo. :shakes:

Everything at stock for now... I'll OC it later once I'm done fiddling with the stupid memory issue... (click to enlarge)
http://www.numberworld.org/screenie_...2011_small.png

http://www.numberworld.org/screenie_...2011_small.png

Running 10b right now...

EDIT: Done. Too tired to stay up, so I let it run overnight.
http://www.numberworld.org/screenie_...2011_small.png

Anyways... It'll probably take me a few weeks to make the AVX code-paths. I've been gone for too long and I have a lot of school/research-work to catch up on.
01-18-2011, 10:46 AM
alpha754293

poke349
I just noticed this, but on the charts, you have the quad AMD Opteron 6174 listed as a single AMD Opteron 6174 and only 64 GB. It's 4x AMD Opteron 6174 (2.2 GHz, 12-core) and it's 128 GB of DDR3-1333 RAM.

Just a minor correction. Thanks.
01-18-2011, 11:43 AM
poke349

Quote:

Originally Posted by alpha754293

poke349
I just noticed this, but on the charts, you have the quad AMD Opteron 6174 listed as a single AMD Opteron 6174 and only 64 GB. It's 4x AMD Opteron 6174 (2.2 GHz, 12-core) and it's 128 GB of DDR3-1333 RAM.

Just a minor correction. Thanks.

I fixed the 4x. The memory is correct. You didn't run 128GB when you ran those benches - except for the 25b.
01-19-2011, 07:06 AM
alpha754293

Quote:

Originally Posted by poke349

I fixed the 4x. The memory is correct. You didn't run 128GB when you ran those benches - except for the 25b.

Oh. My bad. haha...maybe that's when we pulled half the RAM out thinking that it was tripping the PSU's overcurrent protection.

Hmmm, I'll re-run it with the full 128 GB then.

Sorry about that.

*edit*

Where, here's some of the results back. I had to stop it from the 2.5 B mark, cuz I needed the system to do some FEAs for me.

*edit*
I re-ran them all again, upto 10 B.

*edit*
Huhhh....the new runs are slower than the old ones. I think that I'm going to reboot the system, to clear the system caches and stuff from doing FEAs and re-run it yet again. Stay-tuned for the results.

*edit*
Verrrry interesting. With 128 GB of RAM, some of the runs were faster, some were slower. Hmmm....

25M:

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2010 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): AMD Opteron(tm) Processor 6174 Logical Cores: 48 Physical Memory: 136,765,112,320 bytes ( 128 GB ) CPU Frequency: 2,200,046,076 Hz Program Version: 0.5.4 Build 9148 (fix 1) (x64 SSE3 - Windows ~ Kasumi) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 25,000,000 Hexadecimal Digits: Disabled Threading Mode: 64 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 221 MB Start Date: Fri Jan 21 08:34:39 2011 End Date: Fri Jan 21 08:34:50 2011 Computation Time: 8.882 seconds Total Time: 10.203 seconds CPU Utilization: 1321.75 % Multi-core Efficiency: 27.53 % Last Digits: 3803750790 9491563108 2381689226 7224175329 0045253446 : 24,999,950 0786411592 4597806944 2455112852 2554677483 6191884322 : 25,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 264dd02129d2100b1b350442f1ee309d6611473c187724dd18cfeeca25a0d5c4

50M:

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2010 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): AMD Opteron(tm) Processor 6174 Logical Cores: 48 Physical Memory: 136,765,112,320 bytes ( 128 GB ) CPU Frequency: 2,200,040,226 Hz Program Version: 0.5.4 Build 9148 (fix 1) (x64 SSE3 - Windows ~ Kasumi) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 50,000,000 Hexadecimal Digits: Disabled Threading Mode: 64 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 331 MB Start Date: Fri Jan 21 08:34:50 2011 End Date: Fri Jan 21 08:35:11 2011 Computation Time: 18.560 seconds Total Time: 20.704 seconds CPU Utilization: 1820.55 % Multi-core Efficiency: 37.92 % Last Digits: 4127897300 0153683630 8346732220 0943329365 1632962502 : 49,999,950 5130045796 0464561703 2424263071 4554183801 7945652654 : 50,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 77fcb30338665cc71c510a5f0fb1b312a29ef6287b997a661cdb4dc02987f9e3

100M:

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2010 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): AMD Opteron(tm) Processor 6174 Logical Cores: 48 Physical Memory: 136,765,112,320 bytes ( 128 GB ) CPU Frequency: 2,200,034,201 Hz Program Version: 0.5.4 Build 9148 (fix 1) (x64 SSE3 - Windows ~ Kasumi) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 100,000,000 Hexadecimal Digits: Disabled Threading Mode: 64 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 550 MB Start Date: Fri Jan 21 08:35:11 2011 End Date: Fri Jan 21 08:35:56 2011 Computation Time: 40.932 seconds Total Time: 45.280 seconds CPU Utilization: 2307.99 % Multi-core Efficiency: 48.08 % Last Digits: 9948682556 3967530560 3352869667 7734610718 4471868529 : 99,999,950 7572203175 2074898161 1683139375 1497058112 0187751592 : 100,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 46e074e75637fe340a42b6669709bb924ca66d8fbd9931f81fcb11ee14ae6317

250M:

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2010 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): AMD Opteron(tm) Processor 6174 Logical Cores: 48 Physical Memory: 136,765,112,320 bytes ( 128 GB ) CPU Frequency: 2,200,048,178 Hz Program Version: 0.5.4 Build 9148 (fix 1) (x64 SSE3 - Windows ~ Kasumi) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 250,000,000 Hexadecimal Digits: Disabled Threading Mode: 64 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 1.27 GB Start Date: Fri Jan 21 08:35:56 2011 End Date: Fri Jan 21 08:38:12 2011 Computation Time: 125.443 seconds Total Time: 135.587 seconds CPU Utilization: 3197.15 % Multi-core Efficiency: 66.60 % Last Digits: 3673748634 2742427296 0219667627 3141599893 4569474921 : 249,999,950 9958866734 1705167068 8515785208 0067520395 3452027780 : 250,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: c3626149b4dfe67701c72a7159fad2ec70103148572c5a300a52546dba3c7208

500M:

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2010 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): AMD Opteron(tm) Processor 6174 Logical Cores: 48 Physical Memory: 136,765,112,320 bytes ( 128 GB ) CPU Frequency: 2,200,048,908 Hz Program Version: 0.5.4 Build 9148 (fix 1) (x64 SSE3 - Windows ~ Kasumi) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 500,000,000 Hexadecimal Digits: Disabled Threading Mode: 64 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 2.43 GB Start Date: Fri Jan 21 08:38:12 2011 End Date: Fri Jan 21 08:43:21 2011 Computation Time: 288.644 seconds Total Time: 308.968 seconds CPU Utilization: 3576.81 % Multi-core Efficiency: 74.51 % Last Digits: 3896531789 0364496761 5664275325 5483742003 7847987772 : 499,999,950 5002477883 0364214864 5906800532 7052368734 3293261427 : 500,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 938fb14688423d861d98a3aa2d4c1f23f16e42f72cb3432f1d08b242b18886e6

1B:

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2010 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): AMD Opteron(tm) Processor 6174 Logical Cores: 48 Physical Memory: 136,765,112,320 bytes ( 128 GB ) CPU Frequency: 2,200,048,604 Hz Program Version: 0.5.4 Build 9148 (fix 1) (x64 SSE3 - Windows ~ Kasumi) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 1,000,000,000 Hexadecimal Digits: Disabled Threading Mode: 64 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 4.77 GB Start Date: Fri Jan 21 08:43:21 2011 End Date: Fri Jan 21 08:53:36 2011 Computation Time: 574.231 seconds Total Time: 614.573 seconds CPU Utilization: 3735.20 % Multi-core Efficiency: 77.81 % Last Digits: 6434543524 2766553567 4357021939 6394581990 5483278746 : 999,999,950 7139868209 3196353628 2046127557 1517139511 5275045519 : 1,000,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 1562561e2e61dfae00333138f191f0f9343a4cc65b0509ab071b7675a666f600

2.5B:

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2010 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): AMD Opteron(tm) Processor 6174 Logical Cores: 48 Physical Memory: 136,765,112,320 bytes ( 128 GB ) CPU Frequency: 2,200,043,100 Hz Program Version: 0.5.4 Build 9148 (fix 1) (x64 SSE3 - Windows ~ Kasumi) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 2,500,000,000 Hexadecimal Digits: Disabled Threading Mode: 64 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 11.3 GB Start Date: Fri Jan 21 08:53:36 2011 End Date: Fri Jan 21 09:23:18 2011 Computation Time: 1,660.550 seconds Total Time: 1,781.429 seconds CPU Utilization: 3981.26 % Multi-core Efficiency: 82.94 % Last Digits: 0917027898 3554136437 7123165188 3528593128 0032489094 : 2,499,999,950 9228502005 4677489552 2459688725 5242233502 7255998083 : 2,500,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 19a547b8f537e6f018fc2f9ffc6168a17c49b8f7942a5e05171764aece3756f7

5B:

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2010 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): AMD Opteron(tm) Processor 6174 Logical Cores: 48 Physical Memory: 136,765,112,320 bytes ( 128 GB ) CPU Frequency: 2,200,049,740 Hz Program Version: 0.5.4 Build 9148 (fix 1) (x64 SSE3 - Windows ~ Kasumi) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 5,000,000,000 Hexadecimal Digits: Disabled Threading Mode: 64 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 22.5 GB Start Date: Fri Jan 21 09:23:19 2011 End Date: Fri Jan 21 10:10:49 2011 Computation Time: 2,626.111 seconds Total Time: 2,850.030 seconds CPU Utilization: 4059.12 % Multi-core Efficiency: 84.56 % Last Digits: 4384678622 1397184596 0181195416 0748430457 5386741865 : 4,999,999,950 0914971996 1298184401 9216126684 9425834935 5440797257 : 5,000,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: e4ff0dea9b2eef8e04cc430fcc56aabfc1b94e0edfeeaf9607a530b4c8d6175e

10B:

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2010 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): AMD Opteron(tm) Processor 6174 Logical Cores: 48 Physical Memory: 136,765,112,320 bytes ( 128 GB ) CPU Frequency: 2,200,595,152 Hz Program Version: 0.5.4 Build 9148 (fix 1) (x64 SSE3 - Windows ~ Kasumi) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 10,000,000,000 Hexadecimal Digits: Disabled Threading Mode: 64 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 45.1 GB Start Date: Fri Jan 21 10:10:51 2011 End Date: Fri Jan 21 11:43:42 2011 Computation Time: 5,040.317 seconds Total Time: 5,571.566 seconds CPU Utilization: 4003.85 % Multi-core Efficiency: 83.41 % Last Digits: 9763261541 1423749758 2083180752 2573977719 9605119144 : 9,999,999,950 9403994581 8580686529 2375008092 3106244131 4758821220 : 10,000,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 0a154ee18dfa32711d292b9203e1adb220604626a7ef440218e5d76bb7d72c1b
01-21-2011, 12:30 PM
poke349

Nice! I'll update these over the weekend. I also have a bunch of 5 GHz SB results from Shigeru Kondo which I'll add then as well.

One of my larger runs. (click to enlarge)
http://www.numberworld.org/y-crunche...2011_small.png

VCore was 1.344 under load. The CPUz value is wrong since the rig wasn't under load the moment I took the screenie.
4.6 GHz is 3 hours prime stable at the same voltage.

I was able to get 4.9 GHz benchable and 5.0 GHz screenie'able.
Unlike my original i7 rig, this one isn't heat limited. But rather, voltage limited. I don't wanna to feed more than 1.45v into the chip... :(

Definitely not the best chip... :(:(:(
01-23-2011, 04:59 PM
poke349

Charts Updated. :D

I'll hold off on my own results until get the AVX version running...
01-24-2011, 04:14 AM
st0ned

Just saw this today, and I like this kind of projects very much. So if I can do something more to help tell me :)

Also, hopping for AVX support ASAP :cool:

5.0Ghz, 2600k with 8Gb DDR3 800Mhz - 9 - 9 -9 -24

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2010 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz Logical Cores: 8 Physical Memory: 8,569,073,664 bytes ( 8.00 GB ) CPU Frequency: 3,400,047,072 Hz Program Version: 0.5.4 Build 9148 (fix 1) (x64 SSE4.1 - Windows ~ Ushio) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 1,000,000,000 Hexadecimal Digits: Disabled Threading Mode: 8 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 4.75 GB Start Date: Mon Jan 24 12:03:05 2011 End Date: Mon Jan 24 12:08:50 2011 Computation Time: 324.820 seconds Total Time: 345.375 seconds CPU Utilization: 739.36 % Multi-core Efficiency: 92.42 % Last Digits: 6434543524 2766553567 4357021939 6394581990 5483278746 : 999,999,950 7139868209 3196353628 2046127557 1517139511 5275045519 : 1,000,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 45d8a0622c232d741a0507c864da8a3438b163d682cfc0f10acf75fa26d47f09
01-25-2011, 09:55 AM
poke349

Quote:

Originally Posted by st0ned

Just saw this today, and I like this kind of projects very much. So if I can do something more to help tell me :)

Also, hopping for AVX support ASAP :cool:

5.0Ghz, 2600k with 8Gb DDR3 800Mhz - 9 - 9 -9 -24

Nice! :D I'll update it later.

So there's one very annoying thing about AVX now... And that would be the LACK OF LAYMAN DOCUMENTATION on the new instructions... :mad::rofl::yepp::ROTF::shocked::shakes:

I'm having to use trial and error to figure out exactly wtf instructions like "VPERM2F128" (C-intrinsic: _mm256_permute2f128_pd) do...

:rofl::rofl::rofl:
Not funny actually...

The stuff is too new... I'm used to reading everything off of MSDN or forums... but I'm basically working blind right now. lol

EDIT: Oh goodie... :) Visual Studio 2010 is miscompiling some of the AVX instructions... :D:eek:;):(:p::shocked::shakes:
That's gonna be a bit of a problem... :mad: The Intel Compiler is fine, but it's a lot slower at compiling than Visual Studio...
01-28-2011, 04:09 PM
poke349

Thought I'd give everyone a little update on the AVX version:

So I've ported all the easily vectorizable floating-point code to AVX, but the results are actually pretty disappointing. :(

The AVX throughput isn't double the SSE. This is in contrast to what I initially thought (since a number of Intel Employees on XS mentioned that Sandy Bridge will have 2 x 256-bit execution unit - same as the 128-bit units on the Core 2's and the Nehalems.) My guess is that Intel didn't put too much effort into the AVX unit yet since there will probably be very few early adopters. This will be the same situation when Bulldozer arrives. It will not have a full 256-bit module for every core.

Furthermore, (and I've stated this in the past):
Starting from v0.5.2, y-cruncher is no longer a floating-point bound application.
Indeed the application uses a lot of floating-point, but thanks to technologies like SSE and AVX (as well as better skills on my part), this floating-point has been reduced to almost nothing. :yepp::D;):(
AVX does not support 256-bit integer vector operations - and actually, none of y-cruncher's integer code is vectorized... yet :hrhr:

Do not expect the 2x performance that you'd see from other applications that are 100% floating-point.

So when I release v0.5.5, you'll be seeing the effect of AVX on an application that is not very floating-point heavy.
With AVX, y-cruncher is almost entirely integer and memory bound.

(click to enlarge)
http://www.numberworld.org/y-crunche...view_small.png

One more thing:

The Intel Compiler v11.1 does not support compiling AVX programs for non-Intel processors. To top it off, Visual Studio 2010 miscompiles AVX.

So I will not be able to provide an AVX binary for Bulldozer until I get a newer version of either compiler that fixes their problem. (Or if I can find a work-around for the bug in Visual Studio 2010...)
01-29-2011, 09:04 AM
st0ned

5.2Ghz mems were really relaxed ( hard to run high freq and low timings at this MHz :( )

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2010 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz Logical Cores: 8 Physical Memory: 8,571,334,656 bytes ( 8.00 GB ) CPU Frequency: 3,400,051,711 Hz Program Version: 0.5.4 Build 9148 (fix 1) (x64 SSE4.1 - Windows ~ Ushio) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 1,000,000,000 Hexadecimal Digits: Disabled Threading Mode: 8 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 4.75 GB Start Date: Sat Jan 29 16:49:52 2011 End Date: Sat Jan 29 16:55:27 2011 Computation Time: 314.985 seconds Total Time: 335.500 seconds CPU Utilization: 734.85 % Multi-core Efficiency: 91.85 % Last Digits: 6434543524 2766553567 4357021939 6394581990 5483278746 : 999,999,950 7139868209 3196353628 2046127557 1517139511 5275045519 : 1,000,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: f09b28c057e7ca6f0007d33626109d18dd6f6853199d49db012de9b2630e690f
01-29-2011, 06:30 PM
Alpha

New Lappy:

stock speed of 2ghz, turbos to 2.5 at load.

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2010 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): Intel(R) Core(TM) i7-2630QM CPU @ 2.00GHz Logical Cores: 8 Physical Memory: 8,569,573,376 bytes ( 8.00 GB ) CPU Frequency: 1,995,525,983 Hz Program Version: 0.5.4 Build 9148 (fix 1) (x64 SSE4.1 - Windows ~ Ushio) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 500,000,000 Hexadecimal Digits: Disabled Threading Mode: 8 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 2.42 GB Start Date: Sun Jan 30 21:20:20 2011 End Date: Sun Jan 30 21:25:25 2011 Computation Time: 287.077 seconds Total Time: 305.364 seconds CPU Utilization: 758.11 % Multi-core Efficiency: 94.76 % Last Digits: 3896531789 0364496761 5664275325 5483742003 7847987772 : 499,999,950 5002477883 0364214864 5906800532 7052368734 3293261427 : 500,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 788c9880a228b40d0b054777b6990edebc1a3fd62bb58e5e0ea563308f91d071
01-29-2011, 09:29 PM
poke349

Quote:

Originally Posted by st0ned

5.2Ghz mems were really relaxed ( hard to run high freq and low timings at this MHz :( )

Nice! I'll update these sometime tomorrow or Monday.

Quote:

Originally Posted by Alpha

New Lappy:

stock speed of 2ghz, turbos to 2.5 at load.

Sandy Bridge laptops already?!?!?!?! :shocked::shocked::shocked:
Dang... now I want one. :rofl:

EDIT:

Holy $#!+... :eek: 4.7 GHz LinX AVX stable for more than 2 hours... (with and without HT)
I drop the multiplier to 46x keeping all other settings the same...

And it fails y-cruncher v0.5.5 AVX about 8 hours into a 100b Advanced Swap run @ 4.6 GHz... :eek::shocked:

I've measured it, v0.5.5 with AVX doesn't run any hotter than v0.5.4 with SSE4.1... :rofl: But I guess temperatures don't mean that much anyway. :ROTF:

p.s. v0.5.5 will be out in a few days if I don't find any major bugs. ;)

EDIT 2:
Charts updated. ;)
v0.5.5 appears to be about 1% faster than v0.5.4 even without AVX. Somewhat surprising since I didn't really do any optimizations. I guess I'll have to credit that to the newer version of the Intel Compiler.

I'll have SB benchmarks in a few days. I hope I can get 4.8 GHz stable enough to run 2.5b.
But I've been hearing from my beta-testers that v0.5.5 is indeed more intensive than v0.5.4 because of the AVX. I can barely hold 4.9GHz stable for more than a few min. under LinX (no AVX). I'll see about that.
02-01-2011, 12:19 PM
poke349

Alright. I think it's time to let this thing loose:

Version v0.5.5 is released!!! Now with support for AVX instructions!

Download: http://www.numberworld.org/y-cruncher/

Note that the AVX version will NOT run on Bulldozer.
This is because the Intel Compiler v11.1 cannot be configured to compile AVX for non-Intel processors. Visual Studio 2010 can compile AVX, but it miscompiles it. (I have yet to find a work-around.)
So currently, I have no way to build an AVX binary for Windows that will run on Bulldozer.

I'm hoping Visual Studio 2010 SP1 will fix this issue, or an update to the Intel Compiler will let me compile for non-Intel processors. Bulldozer isn't out yet, so I still have time.

Here's a few of my benchmarks:

4.9 GHz with AVX. SSE4.1 would run at 5.0 GHz, but not with AVX... Could only get 4.9 GHz benchable.
Shigeru Kondo is reporting similar results. AVX seems knock about 1 or 2 multipliers off of any v0.5.4 stable overclock.

(click to enlarge)

http://www.numberworld.org/y-crunche...11_2_small.png

And a 100 billion digit run at 4.6 GHz with AVX - 36.486 hours. I had to put a TON of volts into it to get it to pass. :mad:
My first try was with 1.39v which was > 2 LinX (with AVX) stable @ 4.7 GHz. But it BSOD'ed 8 hours in at only 4.6 GHz. :shocked:

I raised the vcore to 1.41v and it held:

(click to enlarge)
http://www.numberworld.org/y-crunche...rend_small.png

http://www.numberworld.org/y-crunche..._end_small.png

EDIT:
I surprisingly didn't need to fix anything to get it to compile for Linux. None of the new code from the past few months broke the Linux version in anyway. :)

And woah... It looks to be about a percent faster than the Windows version. :yepp: Though I doubt that will still be the case for the swap modes.

http://www.numberworld.org/y-crunche...2011_small.png

Gimme a few days to test the Linux builds. Even though all the code compiles and seems to run fine, I still need to test a few things.
02-03-2011, 11:50 AM
alpha754293

Might I suggest that for those who are looking to run with AVX that they need to be aware that of course, their hardware has to support AVX, along with the OS. (In case people are trying to run it and they're noticing that very little has changed.)

*edit*
P.S. There are other compilers out there that should compile your program without Intel compiler and/or Visual Studios.
02-03-2011, 12:17 PM
poke349

Quote:

Originally Posted by alpha754293

Might I suggest that for those who are looking to run with AVX that they need to be aware that of course, their hardware has to support AVX, along with the OS. (In case people are trying to run it and they're noticing that very little has changed.)

*edit*
P.S. There are other compilers out there that should compile your program without Intel compiler and/or Visual Studios.

In Windows:
If the hardware supports AVX, but the OS doesn't, it will tell you outright before continuing. (similar message if you were to run a 32-bit OS on 64-bit hardware)
Similarly, if you were to run this version on Bulldozer, it will tell you that it can't use AVX yet.

In Linux:
I haven't made a dispatcher for it yet, so the user has to pick from one of the 4 binaries. I assume Linux users are more technically inclined. So if they run the AVX version and it crashes, it might be a good hint that the hardware doesn't support it. (Linux has had support for AVX for quite a while now, so I shouldn't have to worry about that.)

If the hardware doesn't support AVX in the first place, then it won't tell you anything about not being able to use AVX.

As for the compilers, GCC is fine. So the Linux version of the AVX binary will run on Bulldozer.
For Windows, I'll wait until closer to the release date of Bulldozer. There's no point in spending time on getting/learning/trying another compiler if the problem will go away by itself.
If neither the Intel Compiler nor Visual Studio has fixed their problem by Bulldozer's release, then I'll start looking for another compiler - MinGW is first on my list.

EDIT:
Here's what it looks like if you try running v0.5.5 on SB without OS support for AVX.

I actually have 3 OS's installed on this machine at the moment.
Windows 7: My default, has all my games and stuff installed.
Windows 7 SP1 Beta: Sole purpose -> AVX programming
Ubuntu 10.10: Linux programming.

The Windows 7 SP1 Beta install is just temporary. I put it on an extra laptop drive I had sitting around. I'm gonna get rid of it when MS officially releases SP1.

(click to enlarge)
http://www.numberworld.org/y-crunche..._AVX_small.png
02-07-2011, 12:22 PM
alpha754293

Quote:

Originally Posted by poke349

In Windows:
If the hardware supports AVX, but the OS doesn't, it will tell you outright before continuing. (similar message if you were to run a 32-bit OS on 64-bit hardware)
Similarly, if you were to run this version on Bulldozer, it will tell you that it can't use AVX yet.

In Linux:
I haven't made a dispatcher for it yet, so the user has to pick from one of the 4 binaries. I assume Linux users are more technically inclined. So if they run the AVX version and it crashes, it might be a good hint that the hardware doesn't support it. (Linux has had support for AVX for quite a while now, so I shouldn't have to worry about that.)

If the hardware doesn't support AVX in the first place, then it won't tell you anything about not being able to use AVX.

As for the compilers, GCC is fine. So the Linux version of the AVX binary will run on Bulldozer.
For Windows, I'll wait until closer to the release date of Bulldozer. There's no point in spending time on getting/learning/trying another compiler if the problem will go away by itself.
If neither the Intel Compiler nor Visual Studio has fixed their problem by Bulldozer's release, then I'll start looking for another compiler - MinGW is first on my list.

EDIT:
Here's what it looks like if you try running v0.5.5 on SB without OS support for AVX.

I actually have 3 OS's installed on this machine at the moment.
Windows 7: My default, has all my games and stuff installed.
Windows 7 SP1 Beta: Sole purpose -> AVX programming
Ubuntu 10.10: Linux programming.

The Windows 7 SP1 Beta install is just temporary. I put it on an extra laptop drive I had sitting around. I'm gonna get rid of it when MS officially releases SP1.

(click to enlarge)
http://www.numberworld.org/y-crunche..._AVX_small.png

In all my years, one of the things that I have learned is never assume people know stuff.

You'd be amazed at how stupid people can really be sometimes.

The last thing that you'd want to hear is somebody trying to run AVX without an AVX-supported OS or hardware. *rolls eyes*

I just suggested the other compilers that might be more "friendly" to you only because I had to look up what AVX is on wiki and it lists OS/compilers that support it.

Some Linux distros might still be using an older kernel (especially true when it comes to Enterprise linuxes), so, recompiling and updating the kernel may be required for AVX support.

I don't do anything that needs AVX, and as far as I know, none of the commerically available software that I use at work supports AVX, so I don't really have to worry about that for now.

If anything, I'd be moving over to GPU computing before I'd be moving to AVX computing.
02-07-2011, 01:10 PM
poke349

Quote:

Originally Posted by alpha754293

In all my years, one of the things that I have learned is never assume people know stuff.

You'd be amazed at how stupid people can really be sometimes.

The last thing that you'd want to hear is somebody trying to run AVX without an AVX-supported OS or hardware. *rolls eyes*

I just suggested the other compilers that might be more "friendly" to you only because I had to look up what AVX is on wiki and it lists OS/compilers that support it.

Some Linux distros might still be using an older kernel (especially true when it comes to Enterprise linuxes), so, recompiling and updating the kernel may be required for AVX support.

I don't do anything that needs AVX, and as far as I know, none of the commerically available software that I use at work supports AVX, so I don't really have to worry about that for now.

If anything, I'd be moving over to GPU computing before I'd be moving to AVX computing.

Yeah, I know what you mean. One the main reasons why I added the dispatcher (y-cruncher.exe) was because I was seeing waaaaay too many people run the 32-bit binaries when they could have run the 64-bit ones. (This was prior to v0.4.3..)

So the dispatcher got rid of that issue. But there isn't much I can do to stop someone from going into the "Binaries" folder and manually running something that's incompatible. I can add checks, but I don't feel like duplicating the CPU-detection code into the main program itself.

For the Linux versions, you are entirely correct. The problem is gonna remain until I make a dispatcher for the Linux release.

As for GPU computing... That's a different can of worms. I suppose anything that scales perfectly from SSE to AVX will probably run well on the Fermi GPUs. (provided it isn't bound by some other resource)
Later on, I'll probably get myself a Fermi card to experiment with. But I don't have my hopes up since the program is already at the point of diminishing return for vectorization with the current set of algorithms.

From what I've seen, anything that's bound by memory doesn't run that much better on GPUs (if at all). GPU <-> main memory is slower than CPU <-> main memory. I'd like to see a GPU with 64GB of memory... Then there's no need to go to main memory. :rofl:
02-07-2011, 06:00 PM
skycrane

hey poke, looks like your package is scheduladed to be there on wed. :) hope you enjoy it...

you will see when you get it that the power demands are a bit big. you need 5 power connectors for it to run. you need 1 24pin atx, 1 4pin molex, 2-8pin power, and 1-4pin power as well. the easiest way i found to hook it up was for one of the 8 pin, to use a 4 to 8 pin adapter, ive sent one with it. its been running at 100% load for the last 14 months like that with no problems :) now for the last 4 pin, use a pic-e and hook it up backwards. if you need a better explination, call me and ill talk you through it. but what you do is take the pci-e and go to hook it up normaly where the latch on it hooks on the board, flip it around so the latch is on the other side 180 degrees, and let the first pair of connectors hang over the side of the mobo, and give it a little pressure it will fit in nicely. to make sure its right, look at the 2 8pins and where the yellow pos12v is the pcie should have its 12pos on the same side .
02-07-2011, 07:22 PM
poke349
Quote:

Originally Posted by skycrane

hey poke, looks like your package is scheduladed to be there on wed. :) hope you enjoy it...

you will see when you get it that the power demands are a bit big. you need 5 power connectors for it to run. you need 1 24pin atx, 1 4pin molex, 2-8pin power, and 1-4pin power as well. the easiest way i found to hook it up was for one of the 8 pin, to use a 4 to 8 pin adapter, ive sent one with it. its been running at 100% load for the last 14 months like that with no problems :) now for the last 4 pin, use a pic-e and hook it up backwards. if you need a better explination, call me and ill talk you through it. but what you do is take the pci-e and go to hook it up normaly where the latch on it hooks on the board, flip it around so the latch is on the other side 180 degrees, and let the first pair of connectors hang over the side of the mobo, and give it a little pressure it will fit in nicely. to make sure its right, look at the 2 8pins and where the yellow pos12v is the pcie should have its 12pos on the same side .

Thanks! :D It's gonna be a very valuable development rig since it'll be the first NUMA machine I ever get. It'll definitely be put to good use. ;)

Woah... that's a big PSU requirement. :shocked:

I just took a closer look at my 1000W Corsair. It has:
- 1 x 8-pin connector - which can split into 2 x 4-pins
- 1 x modular 8-pin connector - which can also split into 2 x 4-pins
- 2 x 3/4-pin PCIe
- enough modular PCIes to run 4-way SLI... :rofl:
So I think it should work without using a PCIe. I can split one of the 8-pin connectors. Run one into the 4-pin on the board and hook up the 4 to 8-pin adapter to the other...

The PSU I have in my Harpertown rig has 2 x 8-pin + a 4-pin "ram power connector". (advertised as being able to handle fully-loaded quad-sockets with fully loaded ram configurations :rofl:)
So in the worst case, I can swap PSUs with that rig.

This 1000W Corsair was originally for my Harpertown rig - and it seems to run it with no issues.
But 16 high-capacity FB-DIMM ram modules pull 4.5A on the +5Vsb rail. That 1000W Corsair is only rated for 3.5A on that rail. So I ended up getting an even stronger PSU for that rig.
But I suppose they overbuild these PSUs anyway... :shrug: I'll see when it arrives. :D
02-07-2011, 09:07 PM
sy5tem2

hello!

here is my little result .

100M

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2011 Alexander J. Yee ( a-yee@northwestern.edu ) User: Carl Nepveu Processor(s): Intel(R) Core(TM)2 Quad CPU Q9400 @ 2.66GHz Logical Cores: 4 Physical Memory: 4,293,963,776 bytes ( 4.00 GB ) CPU Frequency: 3,600,057,472 Hz Program Version: 0.5.5 Build 9178 Alpha (x64 SSE4.1 - Windows ~ Nagisa) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 100,000,000 Hexadecimal Digits: Disabled Threading Mode: 4 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 536 MB Start Date: Mon Feb 07 23:34:57 2011 End Date: Mon Feb 07 23:35:48 2011 Computation Time: 48.579 seconds Total Time: 51.071 seconds CPU Utilization: 368.29 % Multi-core Efficiency: 92.07 % Last Digits: 9948682556 3967530560 3352869667 7734610718 4471868529 : 99,999,950 7572203175 2074898161 1683139375 1497058112 0187751592 : 100,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: a1b698af844046320e5be1d9efc2010f0e436a93392fbd9d18a1189011148184

500M

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2011 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): Intel(R) Core(TM)2 Quad CPU Q9400 @ 2.66GHz Logical Cores: 4 Physical Memory: 4,293,963,776 bytes ( 4.00 GB ) CPU Frequency: 3,600,051,840 Hz Program Version: 0.5.5 Build 9178 Alpha (x64 SSE4.1 - Windows ~ Nagisa) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 500,000,000 Hexadecimal Digits: Disabled Threading Mode: 4 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 2.42 GB Start Date: Mon Feb 07 23:11:53 2011 End Date: Mon Feb 07 23:17:18 2011 Computation Time: 312.729 seconds Total Time: 325.474 seconds CPU Utilization: 376.13 % Multi-core Efficiency: 94.03 % Last Digits: 3896531789 0364496761 5664275325 5483742003 7847987772 : 499,999,950 5002477883 0364214864 5906800532 7052368734 3293261427 : 500,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 8743fd538af6ad5e08eb772e4bb4cb4b39c8e2d6604b16560be1d569747931aa

ty

have a nice day
02-07-2011, 09:47 PM
skycrane
Quote:
Originally Posted by poke349

Thanks! :D It's gonna be a very valuable development rig since it'll be the first NUMA machine I ever get. It'll definitely be put to good use. ;)

Woah... that's a big PSU requirement. :shocked:

I just took a closer look at my 1000W Corsair. It has:

1 x 8-pin connector - which can split into 2 x 4-pins
1 x modular 8-pin connector - which can also split into 2 x 4-pins
2 x 3/4-pin PCIe
enough modular PCIes to run 4-way SLI... :rofl:

So I think it should work without using a PCIe. I can split one of the 8-pin connectors. Run one into the 4-pin on the board and hook up the 4 to 8-pin adapter to the other...

The PSU I have in my Harpertown rig has 2 x 8-pin + a 4-pin "ram power connector". (advertised as being able to handle fully-loaded quad-sockets with fully loaded ram configurations :rofl:)
So in the worst case, I can swap PSUs with that rig.

This 1000W Corsair was originally for my Harpertown rig - and it seems to run it with no issues.
But 16 high-capacity FB-DIMM ram modules pull 4.5A on the +5Vsb rail. That 1000W Corsair is only rated for 3.5A on that rail. So I ended up getting an even stronger PSU for that rig.
But I suppose they overbuild these PSUs anyway... :shrug: I'll see when it arrives. :D
yea that should work :) i had it running for 13 months with a old small PC&C 500w running full load for that long without a problem.... so i think 1kw will be ok...

when you get it, can you keep the white box, and the foam insert for when you send it back.

you should have the board, 4 optis, 8gb ram, hdd with server03 installed, booklet, cd-rom, and 4 hsf. the hsf might be a bit loud :D i cut the blue wire to have them run 100% to keep my temps at a nice 15c :)

the only thing i ask, is that when ur done testing for the day, just to let boinc run the eon project.

skycrane
02-07-2011, 11:02 PM
poke349

Quote:

Originally Posted by skycrane

yea that should work :) i had it running for 13 months with a old small PC&C 500w running full load for that long without a problem.... so i think 1kw will be ok...

when you get it, can you keep the white box, and the foam insert for when you send it back.

you should have the board, 4 optis, 8gb ram, hdd with server03 installed, booklet, cd-rom, and 4 hsf. the hsf might be a bit loud :D i cut the blue wire to have them run 100% to keep my temps at a nice 15c :)

the only thing i ask, is that when ur done testing for the day, just to let boinc run the eon project.

skycrane

Will definitely keep the packaging. (If it's absurdly big, then I might have flatten it to put behind my bed or something...)

I just created a WCG account (Mysticial) and put it on the XS team.
But I can't find the eon project on the lists, so I have it running something else right now. :shrug:

I've got one machine on it for now. (A Core i7 860 @ stock) I might put my i7 920 rig on it later, but I'll need to re-tune the OC since it does get a little hot under load.
Same applies to my SB rig - I have it at 1.4v for 24/7 - not sure if I'm willing to sustain it under load.

One thing is, I can't put my Harpertown rig on WCG. :( It's too loud to keep on 24/7 and probably isn't worth the power consumption with the overkill amount of ram it has.
Even though it's at stock, it has burned out 3 times in the past (even from idling... :shrug:) And everything is past warranty.
02-08-2011, 06:58 AM
skycrane

what burnt out on it? the board? or the ram?
wait a min, thats 8-8gb sticks of fbdimm.... holly crap :shocked: man i hope you got a really great deal on that :)
have you re-re-read the manual? all the servers i have are very picky on connection all the power connectors to the board. you might have missed a small fdd somewhere that gives extra 5v to the ram.
02-08-2011, 07:16 AM
poke349

Quote:

Originally Posted by skycrane

what burnt out on it? the board? or the ram?
wait a min, thats 8-8gb sticks of fbdimm.... holly crap :shocked: man i hope you got a really great deal on that :)
have you re-re-read the manual? all the servers i have are very picky on connection all the power connectors to the board. you might have missed a small fdd somewhere that gives extra 5v to the ram.

Tyan Tempest S5397... It was designed during the early Clovertown days and then recertified for the Harpertowns. Which is probably why.
The highest-end Clovertown at the time pulled 95W. I have a pair of 150W Harpertowns on it with another 200W of ram...

The first time it burned it, it was a mosfet on socket 0.
Second time it died it had something to do with the NB.
Third time it died, it had something to do with the memory controller...

Mobo every single time... And I'm sure nothing was overheating either.

Even with my ridiculous cooling setup, it only delayed those later two inevitable RMAs.
I've stopped putting the machine under 24/7 stress for about a year and a half now. I decided it wasn't worth it and moved all my low-memory computation work to my cheaper single-socket machines.
It's under a new mobo now - SuperMicro MBD-X7DWN+O. But its still gonna stay off most of the time. I only use it for things that will not run on my other machines - precisely anything that needs memory.

p.s. The rig has 16 x 4GB not 8 x 8GB. It had a very special purpose when I first built it a few years ago, but now it's just another one of my coding workstations...

EDIT:
I went through 3 PSUs before settling on one that I liked.
750W Silencer - No Post
1000W Corsair - Works, but +5Vsb rail wasn't strong enough.
1000W Enermax - Works, and is advertised for quad-SLI/quad-fire/quad-socket/32-DIMMs...

The 750W Silencer ended up going in my i7 920 rig. The 1000W Corsair became my open-air tester and will go in the rig you're sending me.
02-08-2011, 10:49 PM
alpha754293

Couple of things - a)

@poke
GPU might still be helpful if it's computationally bound. While yes, if you're memory bound, you probably won't be able to load the entire problem into the GPU RAM, and I would think that swapping to the CPU RAM MAY be possible, but costly. But it would probably still likely be faster than trying to run the whole thing on CPU. A speedup is a speedup is a speedup. I wouldn't know anything about it because I've been pulled off the analysis side of things at work. (Boo hoo... :o() Otherwise, the next thing that I would have been asking for is a Fermi card for my system and take 4 hour FEA and run it in like 15 minutes on the card. But...that doesn't matter. And I still can't justify spending $4k on a card myself for "home" use. (If I could run my CFD on it, that'd be a different story).

b) @skycrane/poke
That's quite the power demands. What kind of board it is? Actually, I've heard about quite a number of the Tyan Intel boards failing at work too. I don't know what happens to them. My Opteron Tyan boards have all been very stable/solid.

Even with our latest quad Socket G34 board, it only takes I think 1x 24pin, and 3x 8pin for the power. And after a little snafu with improperly load balancing the PSU, causing it to trip the overcurrent protection -- it's fine now that it's rebalanced, and running on a Corsair 1kW PSU. So I can't imagine why that system would have that crazy power supply requirements.

The only system that I've got now that's too loud to run 24/7 is my Sun Fire X4200 (dual Opteron 2.8 GHz, single core, 8 GB of RAM, 4x 73 GB 10krpm 2.5" Fujitsu SAS drives, quad GbE + one 10/100 dedicated BMC/SNMP/IPMI/GRASP management NIC). That thing gives me a headache. Everything else, I can pretty much 24/7.
02-09-2011, 10:35 AM
skycrane

hmmmm :D i see something just got delivered.......
02-09-2011, 10:39 AM
poke349

Quote:

Originally Posted by skycrane

hmmmm :D i see something just got delivered.......

Yep :D It's in my room sitting on a static mat right now. Then I realized I had class. :rofl::rofl::rofl:

It looks preeeettttty slick! Then I realized it's not gonna fit in my CM HAF 932. I'm not sure if it will fit even if I cut out the upper PSU mounts that are blocking it. :ROTF:
02-09-2011, 01:20 PM
poke349

Okay... I see what you mean by loud. ;) :D

That's louder than my Harpertown rig (and high pitch). And even louder than my i7 920 rig when I turn the Deltas up to max for suicide OCs. :rofl:

So I'm not gonna be able to sleep with it on.
I also don't think I can put it in my office either, everyone will disembowel me :ROTF:.

To run it 24/7, I'm gonna need to find a way to speed control them.
I don't see anything in the BIOS that doesn't rely on the blue-line, so I'll probably grab one of these off the egg:
http://www.newegg.com/Product/Produc...-065-_-Product

I'll probably grab another 4 x 2GB set of ram to put everything in dual-channel.

(click to enlarge)
http://www.numberworld.org/screenie_...2011_small.jpg

Anyways, I'm gonna try to mount it in my CM HAF 932. I've bent the PSU mounts out of the way and pulled the top fan out. So I think it should work.
02-09-2011, 03:53 PM
skycrane

poke, it looks great, hope your able to make it fit in the case.

you can hook the blue wires back up, just strip the ends off and a little bit of black tape and twist them together should work, without you having to get the speed controller

im sorry about that, i should have fixed them before i sent them out.
02-09-2011, 04:09 PM
poke349

Quote:

Originally Posted by skycrane

poke, it looks great, hope your able to make it fit in the case.

you can hook the blue wires back up, just strip the ends off and a little bit of black tape and twist them together should work, without you having to get the speed controller

im sorry about that, i should have fixed them before i sent them out.

I was thinking along the lines of bringing them into a soldering room to do that, but it might be overkill :rofl:.

The temps barely hit 35 - 39C under prime with 74F ambient. (3 of them are at 32 - 35, one is at 39 so I'll probably try reseating that last one.)
Either way, I'm thinking that's pretty far below the safe 24/7 limit.

It fits! Barely... 3mm taller and the case fans at the top will hit the ram modules. Half an inch taller and it hits the very top of the case.
Since there are no mounting holes up there, I had to stack standoffs to keep the mobo above the tray. So it isn't screwed down up there, but at least it won't touch the mobo tray.
(I did the same with the new EEATX mobo in my Harpertown rig in a CM Cosmos case.)

(click to enlarge)
http://www.numberworld.org/screenie_...se_0_small.jpg

http://www.numberworld.org/screenie_...se_1_small.jpg

Just wondering, were you running these boards open-air? I see no scratches on any of the mounting holes.

Oh and the I/O panel just happens to be compatible with that of my old Tyan S5397... lolz :D
02-09-2011, 04:43 PM
skycrane

yea they were all run in open air, all sitting on a piece of plywood, with small standoffs screwed into the bottom of the cpu mounting plates :) thats amazing that the I/O panel fits.

that is one tight fit in there. how is the HDD i sent working, no problems getting server03 to work?
02-09-2011, 05:00 PM
poke349

Quote:

Originally Posted by skycrane

yea they were all run in open air, all sitting on a piece of plywood, with small standoffs screwed into the bottom of the cpu mounting plates :) thats amazing that the I/O panel fits.

that is one tight fit in there. how is the HDD i sent working, no problems getting server03 to work?

Ah... see, I'm in a dorm. So I don't have that much surface area. :rofl:
Prior to the one you just sent, I had 5 machines in my tiny dorm - 4 of which stay on 24/7. And I'm already taking advantage of that 3rd dimension - up :ROTF:.

I have my i7 920 on WCG. My SB rig will follow once I finish a few things.
I'm gonna try to setup the quad you sent me on Linux WCG since I need it as an SSH server with my group-mates for a NUMA project we're doing.
(I'm gonna put another drive in there for Linux.)

As for the ones I'll keep off WCG:
My laptop idles at 60C. My Harpertown rig is too fragile and too loud to stay on 24/7, and my file server is a Prescott...

Yep, Server 2003 works fine. :D I think there's gonna be a few drivers I need update. I'll see.
02-10-2011, 07:43 AM
Alpha

wow.... Looks nice poke and skycrane! wanna give it a run on multi-prime? (link in sig).

also, that could be a mean gaming rig with that 9800gtx, lol
02-10-2011, 09:53 AM
poke349

Quote:

Originally Posted by Alpha

wow.... Looks nice poke and skycrane! wanna give it a run on multi-prime? (link in sig).

also, that could be a mean gaming rig with that 9800gtx, lol

Will do. But next week though. I can't use the rig until I get fan control and mosfet cooling setup.

Meh... That's just a leftovered and unused card I had sitting around. It sucks for games since it's so old, so all it's good for now is to get full resolution. (Which is definitely helpful for coding... :yepp:)

Charts Updated! :D
02-12-2011, 04:56 AM
skycrane

alpha herre are my times
16 - 2.032
32 - 5.625
64 -15.5
128-42.797
256-118.375
512-327.532
1b - 876.156

you dont have a win 64bit?

these are all on my quad socket quad core opti 8356@ 2.32ghz 16 true cores :)

hey poke, its been 2 days and no update !!!!!! :(
02-12-2011, 08:42 AM
poke349

Quote:

Originally Posted by skycrane

alpha herre are my times
16 - 2.032
32 - 5.625
64 -15.5
128-42.797
256-118.375
512-327.532
1b - 876.156

you dont have a win 64bit?

these are all on my quad socket quad core opti 8356@ 2.32ghz 16 true cores :)

hey poke, its been 2 days and no update !!!!!! :(

I haven't gotten the fans fixed yet. Don't have the tools to do it here. (or the time to do it)
I also don't have enough spare small fans to put on the mosfets, nor do I have a spare optical drive.
My newegg order is coming on Tuesday so I'll have everything I need by then.

So I've been tuning my SB rig for WCG in the meantime. I just put it on yesterday, so I won't know how much it generates until the point system stabilizes in a couple days.

Also, when I shrank the partition on the drive you sent me to make room for Linux, it somehow broke the activation on the machine. :shrug:
I did make a ghost copy of the drive before I did that. But now I'm debating whether to reload the ghost copy, or just install server 2008.

Any idea how many WCG points I can expect from this thing? (can't wait to get it up and running)
02-13-2011, 07:40 PM
alpha754293

@poke/skycrane

What are the specs of the system?
02-13-2011, 08:02 PM
poke349

Quote:

Originally Posted by alpha754293

@poke/skycrane

What are the specs of the system?

4 x Opteron 8356 @ 2.3 GHz
Not sure how much ram he's running.

The one skycrane sent me is the same, but with 8GB.
I should have it up and running 24/7 on Tuesday when my newegg order of fans and fan controllers come in.
02-15-2011, 04:05 PM
poke349

Alright, the thing is good to go for 24/7. It's quieter than the AC and the load temps are all good. The mosfet sinks are cold to touch under load - same with the south bridges...

(click to enlarge)
http://www.numberworld.org/screenie_...2011_small.jpg

Gonna put it on WCG for a few days to see what it gets before I switch it over to Linux for some real NUMA programming! ;)
02-15-2011, 06:11 PM
skycrane

alpha, i had 3 systems all the same, tyan s4985 mobo, 4 8356 optis@2.31ghz, 2 with 8gb of ddr2-4300ecc, and one with 16gb.

i gave one to poke, so he can get the NUMA working for his pi program :)

hey poke, it looks good, glad to see its up and running. do you think there are enough fan connnectors on the mobo :) i think there are 10 of them hehe
02-16-2011, 08:13 AM
poke349

Quote:

Originally Posted by skycrane

alpha, i had 3 systems all the same, tyan s4985 mobo, 4 8356 optis@2.31ghz, 2 with 8gb of ddr2-4300ecc, and one with 16gb.

i gave one to poke, so he can get the NUMA working for his pi program :)

hey poke, it looks good, glad to see its up and running. do you think there are enough fan connnectors on the mobo :) i think there are 10 of them hehe

The pi program is a small thing. It's more about learning how to program for NUMA. :D That's more important.

There's definitely enough fan connectors on the mobo... I haven't used all of them yet. :ROTF:

Just wondering... what's the ram rated for in terms of heat? There's no temperature sensor on them but my heat gun is showing 52 - 55C under load. I'm not sure if that's too hot (and those numbers are just the top of the ICs, not sure about inside...) The FB-DIMMs in my Harpertown rig will hit 70 - 75C, but they are rated for 105C so I'm not concerned about them. But we're talking normal ECC ram here...

EDIT:
@ Alpha
You still want me to run the benchmark? Skycrane already ran it and like he said, the machine is identical.
02-16-2011, 09:19 PM
poke349

Here, I ran it on my other two machines:

I didn't feel like rebooting my Sandy Bridge rig to clock it up. So it's only at 4.4 GHz.

Is there a reason why this thing runs so well on my Harpertown rig? It absolutely destroyed my SB rig.
Even if you cut in a factor of two for having two sockets, my Harpertown rig still beats the SB rig in clock-for-clock efficiency?

(click to enlarge)
http://www.numberworld.org/screenie_...2011_small.png

(click to enlarge)
http://www.numberworld.org/screenie_...2011_small.png
02-17-2011, 05:55 AM
Alpha

it doesnt like HT very much at all for some reason. Now that i have a toy to play with, i can see if i can work something out - lost all my original code though a few years back...

my 9775's destroy this benchmark as well compared to my i7. not sure why...

Here's to learning! :toast:
02-17-2011, 11:45 AM
alpha754293

HTT is an interesting oddity. Sometimes it works really well. Otherwise, not so much. And there doesn't really seem to be any general trend one way or another.

Nice system. I'm still hoping to go up to 32 cores and 128 GB of RAM by June, but I don't know about that as I'm also planning on finish off my master's. (I don't really NEED 32-cores right now, but it would be a fun (and expensive) toy to play with.)
02-17-2011, 11:45 AM
alpha754293

HTT is an interesting oddity. Sometimes it works really well. Otherwise, not so much. And there doesn't really seem to be any general trend one way or another.

Nice system. I'm still hoping to go up to 32 cores and 128 GB of RAM by June, but I don't know about that as I'm also planning on finish off my master's. (I don't really NEED 32-cores right now, but it would be a fun (and expensive) toy to play with.)
02-18-2011, 12:10 PM
poke349

OMG... I just found a HUGE bug in the program... :shocked::eek: I feel like Intel now. lol
It affects Log(2), Log(10), and the Euler-Mascheroni Constant for very large computations...

It's probably been there since v0.4.1... One of the optimizations I did botched the Arc Hyperbolic Cotangent code and affects everything that depends on it. :(:shakes:

Hardly anyone (including me) runs these 3 constants, hence why the bug has been allowed to linger for sooooooo long.

Damn... I'm gonna need to take my SB rig off WCG for a few days to fix it... :(

EDIT:
My Harpertown rig should finish verifying the fix in a couple of days... I'll probably release the patched version if it passes.
The more comprehensive tests will be done simultaneously on my Harpertown rig and my SB rig. But since my SB rig doesn't have as much ram and disk bandwidth, I'm starting it now. My Harpertown rig will probably catch up in no time - hopefully they'll finish at the same time 4 - 5 days from now.
02-18-2011, 02:15 PM
st0ned

Kicked a i980x a ss :D

St0ned - i2600k @ 5.1 8GB ram 2133Mhz 10-12-10-28 1T

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2011 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz Logical Cores: 8 Physical Memory: 8,569,065,472 bytes ( 8.00 GB ) CPU Frequency: 3,400,077,024 Hz Program Version: 0.5.5 Build 9178 Alpha (x64 AVX - Windows ~ Hina) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 1,000,000,000 Hexadecimal Digits: Disabled Threading Mode: 8 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 4.75 GB Start Date: Fri Feb 18 22:06:04 2011 End Date: Fri Feb 18 22:11:04 2011 Computation Time: 284.704 seconds Total Time: 300.343 seconds CPU Utilization: 742.08 % Multi-core Efficiency: 92.76 % Last Digits: 6434543524 2766553567 4357021939 6394581990 5483278746 : 999,999,950 7139868209 3196353628 2046127557 1517139511 5275045519 : 1,000,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 890fdff76350813617e46f49a997a7ee37d4c700fd4c643b19a26237424fa2e7

Pic:

http://img52.imageshack.us/img52/2065/cruncher.jpg

Ok, after the 1st nice result, I went for the killing, yes I was a bit greedy but hey! can you blame me ? Such a nice piece of silicon, I'm in love, volts for both runs were the same.

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2011 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): Intel(R) Core(TM) i7-2600K CPU @ 3.40GHz Logical Cores: 8 Physical Memory: 8,569,065,472 bytes ( 8.00 GB ) CPU Frequency: 3,400,035,360 Hz Program Version: 0.5.5 Build 9178 Alpha (x64 AVX - Windows ~ Hina) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 1,000,000,000 Hexadecimal Digits: Disabled Threading Mode: 8 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 4.75 GB Start Date: Fri Feb 18 22:20:59 2011 End Date: Fri Feb 18 22:25:53 2011 Computation Time: 279.048 seconds Total Time: 294.340 seconds CPU Utilization: 740.41 % Multi-core Efficiency: 92.55 % Last Digits: 6434543524 2766553567 4357021939 6394581990 5483278746 : 999,999,950 7139868209 3196353628 2046127557 1517139511 5275045519 : 1,000,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: c496b0ffc62ffcdf077409de852166e56c13134b07ed842710e229eb0c670dfc

Pics!!:
http://img263.imageshack.us/img263/4240/cruncher52.jpg

2% improve oh well could have been worst :P
02-18-2011, 09:40 PM
poke349

Quote:

Originally Posted by st0ned

Kicked a i980x a ss :D

St0ned - i2600k @ 5.1 8GB ram 2133Mhz 10-12-10-28 1T

Ok, after the 1st nice result, I went for the killing, yes I was a bit greedy but hey! can you blame me ? Such a nice piece of silicon, I'm in love, volts for both runs were the same.

2% improve oh well could have been worst :P

Nice!!! 5+ GHz :D:D:D

Well... what did you expect? 5.1 GHz -> 5.2 GHz is less than 2% :p:

I'll update the charts later.
03-05-2011, 03:28 PM
skycrane

hey poke, its been a few days, any updates ??????
03-05-2011, 04:28 PM
poke349

Quote:

Originally Posted by skycrane

hey poke, its been a few days, any updates ??????

Nothing big so far. Been using it to run little code snippets to see why certain things run so well (or poorly) on AMDs.

Currently, there's one critical subroutine in y-cruncher, (and two more in development) that runs horrifically slow clock-for-clock on K10 compared to Core 2/ix. (this is all in cache so NUMA isn't a factor)
Been trying to figure out the cause of this and improve it if possible.

For example, I'm finding out that it's nearly impossible to sustain the theoretical 1 SSE add + 1 SSE multiply per cycle even when I write meaningless code designed specifically to sustain that throughput.
This is something I can easily achieve in all my post-Core 2 machines but I'm having a hard time even reaching 70% of the theoretical throughput - perhaps I misread the AMD K10 docs or something.
(And I'm getting consistent results across 3 different compilers - so it's not ICC being "unfair". lol)

In the meantime, I've been doing some NUMA ray-tracing on it with 2 other people. So it's my "learning process" for NUMA.
Ray-tracing is obviously easier to parallelize and less memory intensive than something like computing Pi...

One thing is: NUMA programming is HARD - a lot harder and a lot trickier than shared-memory or MPI... :rofl:
At first I thought it was as simple as duplicating resources into all the nodes, but that doesn't really work when it won't fit on a node. :ROTF::shakes:
03-16-2011, 10:07 AM
poke349

Looks like I finally broke the SATA on my SB rig... :rofl::ROTF::shakes:
That was fast - 2 months only.

One of the 4 HDs dropped out on me. Then it refuses to get past the splash screen on boot.
None of the HDs show up in the BIOS. Once I unplug all of them, everything is fine...

Still waiting for that newegg RMA email...
03-23-2011, 04:00 PM
alpha754293

Pity that the i2600K is only 4 cores and that most of the motherboards doesn't support (I think) > 32 GB.
03-24-2011, 08:28 PM
poke349

Quote:

Originally Posted by alpha754293

Pity that the i2600K is only 4 cores and that most of the motherboards doesn't support (I think) > 32 GB.

I'd be all over a Sandy Bridge machine with more than 32GB of ram... :yepp:

if it was affordable that is... :rofl:

The dual-socket Sandies don't come out till later this year - but I don't plan on building another rig for a long time.

A 2 x 8-core Sandy with 128GB or 256GB of ram would be a pretty amazing code-testing machine... :yepp::yepp::yepp: (more so than my Harpertown rig back in 2009)
But that would probably cost more than my car... :rofl:
03-28-2011, 12:28 AM
poke349

Been playing with Linux and NUMA for some time now on the rig Skycrane sent me.

If anyone (with a NUMA machine) is interested, you might be able to get a speedup if you install the NUMA control package (numactl) and play with the settings.

Try running y-cruncher like this:

numactl --interleave=all "./x64 SSE3.out"

or whatever version is compatible.
I haven't checked if there is a Windows equivalent.

It's doesn't "make" the program NUMA aware, but it does seem to distribute the memory traffic more evenly across the cores.
y-cruncher has a problem of having a huge amount of memory/interconnect traffic concentrated on the "primary" NUMA node. Interleaving the memory seems to spread it out a bit.
03-28-2011, 01:02 AM
Nikolasz

Thats my upload, Validation - Pi - 500,000,000
main computer not oced :up:
03-28-2011, 05:12 AM
alpha754293

Quote:

Originally Posted by poke349

Been playing with Linux and NUMA for some time now on the rig Skycrane sent me.

If anyone (with a NUMA machine) is interested, you might be able to get a speedup if you install the NUMA control package (numactl) and play with the settings.

Try running y-cruncher like this:

numactl --interleave=all "./x64 SSE3.out"

or whatever version is compatible.
I haven't checked if there is a Windows equivalent.

It's doesn't "make" the program NUMA aware, but it does seem to distribute the memory traffic more evenly across the cores.
y-cruncher has a problem of having a huge amount of memory/interconnect traffic concentrated on the "primary" NUMA node. Interleaving the memory seems to spread it out a bit.

I don't think that there's numactl for cygwin. I tried looking for it and also trying to see if I compile it from source and get the following error:

Code:

$ make Makefile:179: .depend: No such file or directory cc -MM -DDEPS_RUN -I. bitops.c libnuma.c distance.c memhog.c numactl.c numademo. c numamon.c shm.c stream_lib.c stream_main.c syscall.c util.c mt.c clearcache.c test/*.c > .depend.X && mv .depend.X .depend cc -g -Wall -O2 -I. -c -o numactl.o numactl.c cc -g -Wall -O2 -I. -c -o util.o util.c util.c: In function `memsize': util.c:80: warning: array subscript has type `char' cc -g -Wall -O2 -I. -c -o shm.o shm.c shm.c: In function `attach_shared': shm.c:129: error: storage size of `st' isn't known shm.c:140: warning: implicit declaration of function `fstat64' shm.c:143: warning: implicit declaration of function `ftruncate64' shm.c:157: warning: implicit declaration of function `mmap64' shm.c:157: warning: assignment makes pointer from integer without a cast shm.c:129: warning: unused variable `st' make: *** [shm.o] Error 1

Source from http://oss.sgi.com/projects/libnuma/
03-29-2011, 02:43 AM
poke349

Quote:

Originally Posted by alpha754293

I don't think that there's numactl for cygwin. I tried looking for it and also trying to see if I compile it from source and get the following error:

Code:

$ make Makefile:179: .depend: No such file or directory cc -MM -DDEPS_RUN -I. bitops.c libnuma.c distance.c memhog.c numactl.c numademo. c numamon.c shm.c stream_lib.c stream_main.c syscall.c util.c mt.c clearcache.c test/*.c > .depend.X && mv .depend.X .depend cc -g -Wall -O2 -I. -c -o numactl.o numactl.c cc -g -Wall -O2 -I. -c -o util.o util.c util.c: In function `memsize': util.c:80: warning: array subscript has type `char' cc -g -Wall -O2 -I. -c -o shm.o shm.c shm.c: In function `attach_shared': shm.c:129: error: storage size of `st' isn't known shm.c:140: warning: implicit declaration of function `fstat64' shm.c:143: warning: implicit declaration of function `ftruncate64' shm.c:157: warning: implicit declaration of function `mmap64' shm.c:157: warning: assignment makes pointer from integer without a cast shm.c:129: warning: unused variable `st' make: *** [shm.o] Error 1

Source from http://oss.sgi.com/projects/libnuma/

I'm getting speedups of up to 30% using numactl interleave on Linux. :eek: The speedup depends on the size - for smaller sizes, it sometimes backfires.
I'll accept any settings as valid benchmarks. Tuning the OS like this is part of the game, lol.

As far as I can tell, only Windows Server has any sort of NUMA-awareness support. But you need Win Server anyway to get more than 2 sockets in the first place.
There might be a setting somewhere in the OS that can be set to force interleaved memory allocation. If there isn't, maybe there's a special WinAPI malloc() function that does interleaved allocation. If I find it, I'll try it out.

But in any case, interleaved memory allocation isn't a "solution" to NUMA.
It lets the program get the full memory bandwidth, but it doesn't get rid of the interconnect contention and latency.

The program will still need to be redesigned to run well on NUMA.

EDIT:
Got my B3 stepping today. I finally get my SATA ports back. :D (I still don't get how I managed to kill them in only 2 months...)
Gonna have to re-test my OC tomorrow after the TIM sets. That thermal pad on the H50 was great, but it's only good for the first use, afterwards it's too uneven so I had to scrape it all off and use some of my leftovered Arctic Silver.
03-29-2011, 10:39 AM
alpha754293

Quote:

Originally Posted by poke349

I'm getting speedups of up to 30% using numactl interleave on Linux. :eek: The speedup depends on the size - for smaller sizes, it sometimes backfires.
I'll accept any settings as valid benchmarks. Tuning the OS like this is part of the game, lol.

As far as I can tell, only Windows Server has any sort of NUMA-awareness support. But you need Win Server anyway to get more than 2 sockets in the first place.
There might be a setting somewhere in the OS that can be set to force interleaved memory allocation. If there isn't, maybe there's a special WinAPI malloc() function that does interleaved allocation. If I find it, I'll try it out.

But in any case, interleaved memory allocation isn't a "solution" to NUMA.
It lets the program get the full memory bandwidth, but it doesn't get rid of the interconnect contention and latency.

The program will still need to be redesigned to run well on NUMA.

EDIT:
Got my B3 stepping today. I finally get my SATA ports back. :D (I still don't get how I managed to kill them in only 2 months...)
Gonna have to re-test my OC tomorrow after the TIM sets. That thermal pad on the H50 was great, but it's only good for the first use, afterwards it's too uneven so I had to scrape it all off and use some of my leftovered Arctic Silver.

Okay - correct me if I'm wrong, but I thought that NUMA meant that CPU fastest when it works on data that's closest to the CPU?

I know that I had to play around with NUMA settings for Folding@Home (at the behest of the folks at the forum over there), so that when I ran "coreinfo", it showed that each socket (on a quad-socket system) was on it's own NUMA node.

And I suppose you're right too - < quad-sockets - probably doesn't matter as much (unless you're going in-between nodes).

Hmmm....*confused* :confused:
03-29-2011, 12:37 PM
poke349
Quote:

Originally Posted by alpha754293

Okay - correct me if I'm wrong, but I thought that NUMA meant that CPU fastest when it works on data that's closest to the CPU?

I know that I had to play around with NUMA settings for Folding@Home (at the behest of the folks at the forum over there), so that when I ran "coreinfo", it showed that each socket (on a quad-socket system) was on it's own NUMA node.

And I suppose you're right too - < quad-sockets - probably doesn't matter as much (unless you're going in-between nodes).

Hmmm....*confused* :confused:

You are correct, but there's an even bigger problem with the program.
y-cruncher allocates all the memory it needs in a single gigantic malloc(). The OS will put that entire memory block on the node that called the malloc(). So the result is that most (if not all) of the memory will be concentrated in a single node.

The 4 nodes will all try to access that same memory - all of which is in one node. So not only do you have contention on the interconnect, but the aggregate memory bandwidth is only equal to that of one socket. (since all the memory is concentrated there)
If the memory is interleaved, you still don't get rid of the interconnect traffic, but at least now you have 4 sockets of memory bandwidth.

This problem does not occur in "normal" applications that do lots of small allocations. For these normal apps, the OS will bias all the allocations on the nodes that called them - so memory accesses are mostly local.
But since y-cruncher does one massive malloc() at the beginning, the OS is not able to do this.

There were several major reasons why I made the program do a single malloc() and it's own memory management as opposed to letting the OS handle it.
1. Memory allocation is expensive - (hundreds/thousands of cycles?)
2. malloc() is not thread-safe. Placing locks around it is not scalable. Even if it is thread-safe (with the right compiler options), it's still implemented with locks.
3. Memory allocation is not deterministic. The address that malloc() returns differs between program instances even with the exact same settings.
  The memory manager I put into the program is deterministic - even with multi-threading.
  It's a lot easier to hold together 100,000 lines of multi-threaded code if its deterministic. :yepp:
I made this decision at around October 2008 or so (quite a while ago) - but still before the first version of y-cruncher. It's worked fine since then.

For NUMA, this approach obviously backfires like crazy. It's still possible to make it work without giving up determinism or scalability - but it's messy. It's something I'm probably gonna put off for a few major releases while I redesign/rewrite the program.
There's about 80,000 lines of (nearly) completed code that haven't been enabled yet because the new internal interface is incompatible with what y-cruncher currently uses.
It's gonna take a while to migrate the program to that new interface.
The NUMA layer will eventually be built on top of all the existing "normal" threading code - though I have yet to work out the details.
03-30-2011, 06:18 AM
alpha754293
Quote:
Originally Posted by poke349

You are correct, but there's an even bigger problem with the program.
y-cruncher allocates all the memory it needs in a single gigantic malloc(). The OS will put that entire memory block on the node that called the malloc(). So the result is that most (if not all) of the memory will be concentrated in a single node.

The 4 nodes will all try to access that same memory - all of which is in one node. So not only do you have contention on the interconnect, but the aggregate memory bandwidth is only equal to that of one socket. (since all the memory is concentrated there)
If the memory is interleaved, you still don't get rid of the interconnect traffic, but at least now you have 4 sockets of memory bandwidth.

This problem does not occur in "normal" applications that do lots of small allocations. For these normal apps, the OS will bias all the allocations on the nodes that called them - so memory accesses are mostly local.
But since y-cruncher does one massive malloc() at the beginning, the OS is not able to do this.

There were several major reasons why I made the program do a single malloc() and it's own memory management as opposed to letting the OS handle it.

Memory allocation is expensive - (hundreds/thousands of cycles?)
malloc() is not thread-safe. Placing locks around it is not scalable. Even if it is thread-safe (with the right compiler options), it's still implemented with locks.
Memory allocation is not deterministic. The address that malloc() returns differs between program instances even with the exact same settings.
The memory manager I put into the program is deterministic - even with multi-threading.
It's a lot easier to hold together 100,000 lines of multi-threaded code if its deterministic. :yepp:

I made this decision at around October 2008 or so (quite a while ago) - but still before the first version of y-cruncher. It's worked fine since then.

For NUMA, this approach obviously backfires like crazy. It's still possible to make it work without giving up determinism or scalability - but it's messy. It's something I'm probably gonna put off for a few major releases while I redesign/rewrite the program.
There's about 80,000 lines of (nearly) completed code that haven't been enabled yet because the new internal interface is incompatible with what y-cruncher currently uses.
It's gonna take a while to migrate the program to that new interface.
The NUMA layer will eventually be built on top of all the existing "normal" threading code - though I have yet to work out the details.
More stupid questions:

will MPI or OMP help?

How is it doing the multi-threading/multi-processing now anyways?

Can threads do malloc()? (You might have actually said it already, but I'm a dummy non-programmer trying to wrap my head around programming concepts.)

I've always been under the impression/belief that multi-processing can happen ONLY if it spawns slave processes. So the whole idea of multi-threading != multi-processing in my mind (or at least it doesn't seem to be as efficient or doesn't use the full resources available to it).
03-30-2011, 07:28 AM
repman

Had some spare time to run this on one of my older rigs, just for fun :rofl:

It's a HP ML350G4 as shown in my sig.

http://img153.imageshack.us/img153/1571/ml350g4.th.jpg

25M

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2011 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): Intel(R) Xeon(TM) CPU 3.00GHz Logical Cores: 4 Physical Memory: 4,227,399,680 bytes ( 3.93 GB ) CPU Frequency: 3,000,194,895 Hz Program Version: 0.5.5 Build 9179 (fix 1) (x64 SSE3 - Windows) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 25,000,000 Hexadecimal Digits: Disabled Threading Mode: 4 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 207 MB Start Date: Wed Mar 30 14:03:49 2011 End Date: Wed Mar 30 14:04:32 2011 Computation Time: 40.462 seconds Total Time: 43.022 seconds CPU Utilization: 373.11 % Multi-core Efficiency: 93.27 % Last Digits: 3803750790 9491563108 2381689226 7224175329 0045253446 : 24,999,950 0786411592 4597806944 2455112852 2554677483 6191884322 : 25,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 67dec58e652f6b1d0fea1feafad483d2799e90c7c5db8b200c858b7f86d6f727

50M

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2011 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): Intel(R) Xeon(TM) CPU 3.00GHz Logical Cores: 4 Physical Memory: 4,227,399,680 bytes ( 3.93 GB ) CPU Frequency: 3,000,202,463 Hz Program Version: 0.5.5 Build 9179 (fix 1) (x64 SSE3 - Windows) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 50,000,000 Hexadecimal Digits: Disabled Threading Mode: 4 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 316 MB Start Date: Wed Mar 30 14:06:08 2011 End Date: Wed Mar 30 14:07:43 2011 Computation Time: 89.224 seconds Total Time: 94.464 seconds CPU Utilization: 380.51 % Multi-core Efficiency: 95.12 % Last Digits: 4127897300 0153683630 8346732220 0943329365 1632962502 : 49,999,950 5130045796 0464561703 2424263071 4554183801 7945652654 : 50,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 1db010eae8d71aad16d348835969f377f22f7419e93594ee04695171dcf2eb5e

100M

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2011 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): Intel(R) Xeon(TM) CPU 3.00GHz Logical Cores: 4 Physical Memory: 4,227,399,680 bytes ( 3.93 GB ) CPU Frequency: 3,000,202,223 Hz Program Version: 0.5.5 Build 9179 (fix 1) (x64 SSE3 - Windows) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 100,000,000 Hexadecimal Digits: Disabled Threading Mode: 4 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 536 MB Start Date: Wed Mar 30 14:08:34 2011 End Date: Wed Mar 30 14:12:04 2011 Computation Time: 200.057 seconds Total Time: 210.020 seconds CPU Utilization: 386.74 % Multi-core Efficiency: 96.68 % Last Digits: 9948682556 3967530560 3352869667 7734610718 4471868529 : 99,999,950 7572203175 2074898161 1683139375 1497058112 0187751592 : 100,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 8f72a3e29cc2c459101624ad135c899f8894f4d30000bad9123c253fa1ba71ee

250M

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2011 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): Intel(R) Xeon(TM) CPU 3.00GHz Logical Cores: 4 Physical Memory: 4,227,399,680 bytes ( 3.93 GB ) CPU Frequency: 3,000,188,655 Hz Program Version: 0.5.5 Build 9179 (fix 1) (x64 SSE3 - Windows) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 250,000,000 Hexadecimal Digits: Disabled Threading Mode: 4 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 1.25 GB Start Date: Wed Mar 30 14:13:07 2011 End Date: Wed Mar 30 14:23:09 2011 Computation Time: 576.015 seconds Total Time: 602.238 seconds CPU Utilization: 390.76 % Multi-core Efficiency: 97.69 % Last Digits: 3673748634 2742427296 0219667627 3141599893 4569474921 : 249,999,950 9958866734 1705167068 8515785208 0067520395 3452027780 : 250,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 2d5e582a64c81add039e51df177d4a69a1f84aa395364b6915776ddfb6e1841f

500M

Code:

Validation Version: 1.1 Program: y-cruncher - Gamma to the eXtReMe!!! ( www.numberworld.org ) Copyright 2008-2011 Alexander J. Yee ( a-yee@northwestern.edu ) User: None Specified - You can edit this in "Username.txt". Processor(s): Intel(R) Xeon(TM) CPU 3.00GHz Logical Cores: 4 Physical Memory: 4,227,399,680 bytes ( 3.93 GB ) CPU Frequency: 3,000,192,863 Hz Program Version: 0.5.5 Build 9179 (fix 1) (x64 SSE3 - Windows) Constant: Pi Algorithm: Chudnovsky Formula Decimal Digits: 500,000,000 Hexadecimal Digits: Disabled Threading Mode: 4 threads Computation Mode: Ram Only Swap Disks: 0 Working Memory: 2.42 GB Start Date: Wed Mar 30 14:25:08 2011 End Date: Wed Mar 30 14:47:20 2011 Computation Time: 1,275.159 seconds Total Time: 1,331.560 seconds CPU Utilization: 392.47 % Multi-core Efficiency: 98.11 % Last Digits: 3896531789 0364496761 5664275325 5483742003 7847987772 : 499,999,950 5002477883 0364214864 5906800532 7052368734 3293261427 : 500,000,000 Timer Sanity Check: Passed Frequency Sanity Check: Passed ECC Recovered Errors: 0 Checkpoint From: None ---- Checksum: 10ef6c30486f7b58013d61d24782c8cda6cc98b3c21e5dbc19074b403e5e0b98

I also ran a few on my DL380 but it's powered down ATM so I can't get the results I think I did up to 1B there with 3 disks as swap.
I'll take some time once and do some more on the ML350 using swap :)
For the DL320 I need to take the RAM out of the 380 since it only has 1GB and 2 SATA 80GB disks which are in RAID 1 so those won't help...much:ROTF:
03-30-2011, 07:50 AM
soya_crack

I saw that nobody had yet uploaded a result of a AMD X6 so I searched my old screens and found one, maybe it helps you.

http://www.abload.de/img/ycrunchercuti7d1.png

Soya
03-30-2011, 08:30 AM
rcofell
Quote:

Originally Posted by poke349

I'm getting speedups of up to 30% using numactl interleave on Linux. :eek: The speedup depends on the size - for smaller sizes, it sometimes backfires.
I'll accept any settings as valid benchmarks. Tuning the OS like this is part of the game, lol.

As far as I can tell, only Windows Server has any sort of NUMA-awareness support. But you need Win Server anyway to get more than 2 sockets in the first place.
There might be a setting somewhere in the OS that can be set to force interleaved memory allocation. If there isn't, maybe there's a special WinAPI malloc() function that does interleaved allocation. If I find it, I'll try it out.

But in any case, interleaved memory allocation isn't a "solution" to NUMA.
It lets the program get the full memory bandwidth, but it doesn't get rid of the interconnect contention and latency.

The program will still need to be redesigned to run well on NUMA.

I've been meaning to post in here for awhile, as I've got access to an 8-socket Barcelona machine (w/ ~80GB ram) here in our research group at school that I've tried running things on. The NUMA nature of the machine definitely causes sub-optimal scaling for the smaller workloads (those below 1B).

I tried numactl for 1B and performance goes from 648sec (without) to 466sec (interleave all), so it does make a noticeable difference. Almost a 1.4X speedup in this case.

For comparison (without numactl), 10B takes 5,379sec and 100M takes 110sec.
Quote:
Originally Posted by poke349

You are correct, but there's an even bigger problem with the program.
y-cruncher allocates all the memory it needs in a single gigantic malloc(). The OS will put that entire memory block on the node that called the malloc(). So the result is that most (if not all) of the memory will be concentrated in a single node.

The 4 nodes will all try to access that same memory - all of which is in one node. So not only do you have contention on the interconnect, but the aggregate memory bandwidth is only equal to that of one socket. (since all the memory is concentrated there)
If the memory is interleaved, you still don't get rid of the interconnect traffic, but at least now you have 4 sockets of memory bandwidth.

This problem does not occur in "normal" applications that do lots of small allocations. For these normal apps, the OS will bias all the allocations on the nodes that called them - so memory accesses are mostly local.
But since y-cruncher does one massive malloc() at the beginning, the OS is not able to do this.

There were several major reasons why I made the program do a single malloc() and it's own memory management as opposed to letting the OS handle it.

Memory allocation is expensive - (hundreds/thousands of cycles?)
malloc() is not thread-safe. Placing locks around it is not scalable. Even if it is thread-safe (with the right compiler options), it's still implemented with locks.
Memory allocation is not deterministic. The address that malloc() returns differs between program instances even with the exact same settings.
The memory manager I put into the program is deterministic - even with multi-threading.
It's a lot easier to hold together 100,000 lines of multi-threaded code if its deterministic. :yepp:
Yeah, tying all of the memory allocation to one thread/node doesn't do so well with NUMA I imagine. Beyond just the interconnect bandwidth impact, coherency traffic must become a significant overhead as well.

Determinism definitely is desirable with parallel programs, though in what capacity are you using the deterministic malloc()? Is it mainly for debug, or structured memory/pointer accesses? As for thread-safe, are you saying just using barriers to properly synchronize isn't good enough?

At the expense of memory capacity trade-off, could you have each thread malloc() a (relatively) small buffer to work out of? In the interest of creating some locality and doing burst transfers.

Quote:

Originally Posted by alpha754293

More stupid questions:

will MPI or OMP help?

How is it doing the multi-threading/multi-processing now anyways?

Can threads do malloc()? (You might have actually said it already, but I'm a dummy non-programmer trying to wrap my head around programming concepts.)

I've always been under the impression/belief that multi-processing can happen ONLY if it spawns slave processes. So the whole idea of multi-threading != multi-processing in my mind (or at least it doesn't seem to be as efficient or doesn't use the full resources available to it).

I'm curious of the current implementation as well :)

Is it all p-thread based? MPI would definitely be the way to go for ensuring proper locality on a NUMA system, at the expense of all the icky manual data sharing. I've only done some basic use of OpenMP, so I don't know too much of the advanced details like optimizations for NUMA systems, but is it really applicable to your current algorithm?

I remember reading sometime ago about your implementation spawning a large amount of threads, do you limit the number of work threads to the available hardware threads, or go beyond? Having excessive threads would likely thrash around too much with larger NUMA systems and be counter productive.

How independent are the threads? Is there much synchronization/sharing of data between them, or does each basically work on its own independent chunk of the series and combine in a tree-reduce fashion?
03-30-2011, 10:32 AM
poke349

Quote:

Originally Posted by alpha754293

More stupid questions:

will MPI or OMP help?

How is it doing the multi-threading/multi-processing now anyways?

Can threads do malloc()? (You might have actually said it already, but I'm a dummy non-programmer trying to wrap my head around programming concepts.)

I've always been under the impression/belief that multi-processing can happen ONLY if it spawns slave processes. So the whole idea of multi-threading != multi-processing in my mind (or at least it doesn't seem to be as efficient or doesn't use the full resources available to it).

Yes, threads can malloc(), but malloc() isn't thread safe. Some compilers can make it thread-safe with the right options, but that's done using locks - which limits scalability. I'll get to that later.

Quote:

Originally Posted by rcofell

I've been meaning to post in here for awhile, as I've got access to an 8-socket Barcelona machine (w/ ~80GB ram) here in our research group at school that I've tried running things on. The NUMA nature of the machine definitely causes sub-optimal scaling for the smaller workloads (those below 1B).

I tried numactl for 1B and performance goes from 648sec (without) to 466sec (interleave all), so it does make a noticeable difference. Almost a 1.4X speedup in this case.

For comparison (without numactl), 10B takes 5,379sec and 100M takes 110sec.

Yeah, tying all of the memory allocation to one thread/node doesn't do so well with NUMA I imagine. Beyond just the interconnect bandwidth impact, coherency traffic must become a significant overhead as well.

Determinism definitely is desirable with parallel programs, though in what capacity are you using the deterministic malloc()? Is it mainly for debug, or structured memory/pointer accesses? As for thread-safe, are you saying just using barriers to properly synchronize isn't good enough?

At the expense of memory capacity trade-off, could you have each thread malloc() a (relatively) small buffer to work out of? In the interest of creating some locality and doing burst transfers.

I'm curious of the current implementation as well :)

Is it all p-thread based? MPI would definitely be the way to go for ensuring proper locality on a NUMA system, at the expense of all the icky manual data sharing. I've only done some basic use of OpenMP, so I don't know too much of the advanced details like optimizations for NUMA systems, but is it really applicable to your current algorithm?

I remember reading sometime ago about your implementation spawning a large amount of threads, do you limit the number of work threads to the available hardware threads, or go beyond? Having excessive threads would likely thrash around too much with larger NUMA systems and be counter productive.

How independent are the threads? Is there much synchronization/sharing of data between them, or does each basically work on its own independent chunk of the series and combine in a tree-reduce fashion?

For computation threads, threading is done using recursive thread-forking.
For I/O threads, they are done using iterative thread-spawning.

I'll only focus on the computation threads.

Basically, there's a ton of binary divide-and-conquer crap in the program. Since each of these divide-and-conquer algorithms typically involve some sort of data merging/splitting at each stage, you can't easily split them into N threads at the start. So instead, I integrate the recursive thread-splitting into these binary divide-and-conquer algorithms.
At each stage, I split the current thread into two. In Windows, this is done by spawning another thread. In Linux, it's done using recursive OpenMP.
There's no pthread version yet, but it would basically be a copy/paste of the Windows code with some renaming. So far, OpenMP works just fine, so there's no need for pthreads yet.
By adjusting the depth of the thread-splitting, I can change the # of threads. This is the primary reason why y-cruncher is locked to power-of-two threads.

Some of the new code that I'm working on right now is able to do non-power-of-two threads using the same recursive splitting method. So in the future, I might be able to lift the power-of-two restriction on the program.

If the routine needs memory, then the memory heap is split in half (or some pre-determined ratio). This is also done recursively along with the thread-splitting.
All memory allocations that a thread does are on it's own part of the heap. They never touch another thread's memory - hence why its deterministic. No locks, no synchronization, no overhead.
The only non-deterministic bugs are the ones where one thread accidentally touches another thread's memory. But these are rare unless an out-of-bound array access happens to cross a thread boundary in the heap.
(In fact, I have yet to encounter a bug like this since I started this project)

There's a lot of fancy math needed to prove that a routine will never use more than XX memory - especially when you factor in memory fragmentation.
Another fancy protocol is needed to join threads when a large value needs to be returned via the heap.

So the memory allocation is deterministic, and completely free of locks/synchronization. A memory allocation is typically 20-100 cycles depending on how much stuff is on the heap.

Locking/synchronization isn't scalable, so there's none of it in the program. Thread-joining is the only form of "synchronization" that's used.

OS-managed memory allocation isn't scalable for this reason. All that page committing stuff needs to access global data - using locks...
To top it off the OS needs to zero newly allocated memory before the program can use it. (This is for security reasons: so you can't like steal a password by logging into to something, closing the program and allocating that physical memory into another program.)

As for pre-allocating a buffer for each thread to get better locality in NUMA - that's essentially the same as MPI as you have to explicitly move data in and out of that buffer.

At any given point in the program, all the threads are entirely independent.
When all threads complete their given tasks, they are assigned new tasks on different parts of memory. So no single part of memory is accessed by only one thread.

Implementing this in MPI will require inter-node matrix transposes - which is a huge research topic in my field.
In shared memory, these transposes can be eliminated by directly accessing the memory that's needed - this is what y-cruncher does right now.

In any case, an MPI implementation will pretty much require a rewrite of all the high-level code. I don't have immediate plans for this. And when I do, it'll be built on top of the NUMA/shared-memory layer.

EDIT:
Yes, it's a tree-reduction. :)

EDIT 2:
The program does in fact use more threads than hardware threads. This is to reduce the impact of load imbalance.
For systems with a non-power-of-two hardware threads, it needs to be rounded up to the next power-of-two anyways.
04-01-2011, 09:29 AM
alpha754293

Very interesting. I gotta say, I'm learning a LOT from this thread, although I also honestly, still don't understand alot of it as well only because I'm not a programmer. BUT...I THINK that I sorta get the jist of it though (and pity that you're so far otherwise I think that I can really learn programming from you), since you're one of the few people that I've EVER met who can explain it and explain it well!

Here's something interesting that I've noticed -- on the 48-core system that I've got at work - it's four sockets. I THINK that I have it set up so that there are four NUMA nodes.

And when I run y-cruncher on it, it doesn't really seem to quite take full advantage of the hardware and that's evident in the CPU utilization reported by Windows Task Manager and also by the program itself.

Is that (^ the above aforementioned ^) why it's like that?

I'm only used to commercially available software that's MPI and/or OpenMP and maybe it's cuz it's commercial, so it has very high CPU utilization. ?
04-01-2011, 09:51 AM
poke349

Quote:

Originally Posted by alpha754293

Very interesting. I gotta say, I'm learning a LOT from this thread, although I also honestly, still don't understand alot of it as well only because I'm not a programmer. BUT...I THINK that I sorta get the jist of it though (and pity that you're so far otherwise I think that I can really learn programming from you), since you're one of the few people that I've EVER met who can explain it and explain it well!

Here's something interesting that I've noticed -- on the 48-core system that I've got at work - it's four sockets. I THINK that I have it set up so that there are four NUMA nodes.

And when I run y-cruncher on it, it doesn't really seem to quite take full advantage of the hardware and that's evident in the CPU utilization reported by Windows Task Manager and also by the program itself.

Is that (^ the above aforementioned ^) why it's like that?

I'm only used to commercially available software that's MPI and/or OpenMP and maybe it's cuz it's commercial, so it has very high CPU utilization. ?

lol... I thought I was always bad at explaining stuff. ;)
Also, I don't always know what I'm talking about. So take my posts with a grain of salt. :D:rofl:

What you're seeing is load imbalance due to NUMA. What's happening is that some nodes run faster than others depending on where the data is. So you'll find one or two nodes lagging far behind the others - and when the threads on the faster nodes finish first, the slower threads will hang around a lot longer afterwards. So you see low CPU usage.

I was confused as well when I first noticed this on the 4 x 4 that Skycrane sent.
From the beginning, I had already suspected this was the cause. But I couldn't confirm it until I wrote some mini-benchmarks to specifically test for this.

There may be more reasons to it, but so far that's the only explanation I have.
04-03-2011, 07:35 PM
alpha754293

Quote:

Originally Posted by poke349

lol... I thought I was always bad at explaining stuff. ;)
Also, I don't always know what I'm talking about. So take my posts with a grain of salt. :D:rofl:

What you're seeing is load imbalance due to NUMA. What's happening is that some nodes run faster than others depending on where the data is. So you'll find one or two nodes lagging far behind the others - and when the threads on the faster nodes finish first, the slower threads will hang around a lot longer afterwards. So you see low CPU usage.

I was confused as well when I first noticed this on the 4 x 4 that Skycrane sent.
From the beginning, I had already suspected this was the cause. But I couldn't confirm it until I wrote some mini-benchmarks to specifically test for this.

There may be more reasons to it, but so far that's the only explanation I have.

Well, if you need another quad-socket system to test stuff with, my quad dual-core (8-cores total) at home that's available. It's not the fastest system by any stretch of the imagination now, and I'll still have access to my 48-core system at work for about another week (I'm switching jobs).

And you're actually pretty good at explaining stuff. Better than most profs and even some programmers that I've met before. I'm pretty sure that you have a pretty good idea what you're talking about because otherwise, you probably wouldn't be writing the programs and doing what it is that you're doing. (Last I recall, you're doing your grad school...so I'd presume that you gotta know SOMETHING.)
04-04-2011, 09:39 PM
poke349

Quote:

Originally Posted by alpha754293

Well, if you need another quad-socket system to test stuff with, my quad dual-core (8-cores total) at home that's available. It's not the fastest system by any stretch of the imagination now, and I'll still have access to my 48-core system at work for about another week (I'm switching jobs).

And you're actually pretty good at explaining stuff. Better than most profs and even some programmers that I've met before. I'm pretty sure that you have a pretty good idea what you're talking about because otherwise, you probably wouldn't be writing the programs and doing what it is that you're doing. (Last I recall, you're doing your grad school...so I'd presume that you gotta know SOMETHING.)

Well, I may be in grad school, but I'm still just a first year... There's still a lot I don't know - especially supercomputer related things.
The only difference between me and other grad students is that I have a bunch of machines to toy with. :D Everyone else needs to share the school's resources.
And there's a lot you can't do when you don't have root or physical access to a machine you're trying to study.

Stuff like playing with the CPU clock, messing with memory timings, enabling/disabling cores... are some of the unconventional things I do to see what's going on in a program.

If you try that on a school machine, you'll get expelled. :rofl:

And I have some good news for the AMD fans...

Visual Studio 2010 SP1 fixes the bug that was miscompiling AVX.

It looks like I have a Windows binary that will run on Bulldozer. :yepp:
Interestingly, Visual Studio 2010 SP1 compiles a faster AVX binary than the Intel Compiler (about 1% faster)... :shrug:

Also, VS 2010 SP1 has support for FMA4 and XOP instructions. But since I don't plan on building another rig for a while, I won't be able to add support for FMA4 and XOP anytime soon.
04-05-2011, 04:55 AM
skycrane

so when are you going to update the program to take advantage of the 1.4x speed up :) :party:

good to hear you can fix the bug in it... now maybe the amd will give the intel some competiton....
04-05-2011, 06:49 AM
poke349

Quote:

Originally Posted by skycrane

so when are you going to update the program to take advantage of the 1.4x speed up :) :party:

good to hear you can fix the bug in it... now maybe the amd will give the intel some competiton....

I've been searching through the Windows docs and I can't find any option that'll do interleaved memory allocation. Some of the NUMA control functions "might" let me do an interleaved malloc() - I'm not sure yet.

So for now, Linux is the only way to get that 1.4x speed up. :( At least until I find a way to do interleaving in Windows...

All the NUMA control features in Windows seem to require Vista or later. So when I use them, I'm probably gonna drop support for x86 (without SSE3).
Any processor old enough to not have SSE3 is not gonna be able to run Vista. :rofl: So that's one less code-path I need to maintain... lol

Anyways, I'm gonna release a new patch for the program soon. It'll have the Bulldozer-compatible AVX binary, and my updated email address.
(My school finally terminated my northwestern.edu email since I graduated last year and I'm not a student there anymore.)
04-05-2011, 07:23 AM
alpha754293

Quote:

Originally Posted by poke349

Well, I may be in grad school, but I'm still just a first year... There's still a lot I don't know - especially supercomputer related things.
The only difference between me and other grad students is that I have a bunch of machines to toy with. :D Everyone else needs to share the school's resources.
And there's a lot you can't do when you don't have root or physical access to a machine you're trying to study.

Stuff like playing with the CPU clock, messing with memory timings, enabling/disabling cores... are some of the unconventional things I do to see what's going on in a program.

If you try that on a school machine, you'll get expelled. :rofl:

And I have some good news for the AMD fans...

Visual Studio 2010 SP1 fixes the bug that was miscompiling AVX.

It looks like I have a Windows binary that will run on Bulldozer. :yepp:
Interestingly, Visual Studio 2010 SP1 compiles a faster AVX binary than the Intel Compiler (about 1% faster)... :shrug:

Also, VS 2010 SP1 has support for FMA4 and XOP instructions. But since I don't plan on building another rig for a while, I won't be able to add support for FMA4 and XOP anytime soon.

Quote:

Originally Posted by skycrane

....1.4x speed up...

Really? AVX makes that much of a difference?

Hmm....I wonder if programs like LINPACK would be able to pick up/use that difference?

Way I look at it - there's always going to be someone who knows more than I do, and there's A LOT of people who know less. (It isn't an ego thing, it's a reality thing, or so I've been told. I'm usually too busy doing my own thing to notice anyways...or care.)

I still wonder why don't you go with something like PGI or Pathscale or some of the more "generic" (less stupid/specific) compilers?

Haha...I got kicked off the school systems for a while because I was running the distributed.net client on their Sun SPARC servers. Then I was like "screw this" and ran it on my home systems instead cuz it was faster. When I went to re-activate my account, they're like "it looks like you're blocked. Hmmm...I dunno why." and I'm like "*shrug* I unno either..." LOL

Stupid class that actually needed access to the labs.
04-05-2011, 07:44 AM
poke349

Quote:

Originally Posted by alpha754293

Really? AVX makes that much of a difference?

Hmm....I wonder if programs like LINPACK would be able to pick up/use that difference?

Way I look at it - there's always going to be someone who knows more than I do, and there's A LOT of people who know less. (It isn't an ego thing, it's a reality thing, or so I've been told. I'm usually too busy doing my own thing to notice anyways...or care.)

I still wonder why don't you go with something like PGI or Pathscale or some of the more "generic" (less stupid/specific) compilers?

Haha...I got kicked off the school systems for a while because I was running the distributed.net client on their Sun SPARC servers. Then I was like "screw this" and ran it on my home systems instead cuz it was faster. When I went to re-activate my account, they're like "it looks like you're blocked. Hmmm...I dunno why." and I'm like "*shrug* I unno either..." LOL

Stupid class that actually needed access to the labs.

The 1.4x is from the NUMA interleaved memory, not AVX.
Right now, AVX is only giving about 10% since the program is mostly integer and memory bound. I hope to change this situation in v0.7.x, but that's still a long way away.

I started with Visual Studio because it's easy to use and is user-friendly (integrated with GUI). Then I switched to the Intel Compiler since everyone says it compiles the fastest code and stays up to date with instruction sets.
I still use Visual Studio for most of the dev work. Only the release builds are compiled using ICC.

Switching compilers is actually a pretty big hassle. I have to rewrite all my batch scripts and re-tune a lot of the code if I want to get maximum performance.

EDIT:

Oh cool. I can compile FMA4 and XOP with VS 2010...
Too bad there's no way for me to run it. :rofl:

Code:

vmulpd ymm0, ymm0, ymm15 vfmaddpd ymm1, ymm1, ymm12, ymm0 vroundpd ymm5, ymm3, 0 vroundpd ymm14, ymm7, 0 vmulpd ymm0, ymm5, ymm4 vsubpd ymm3, ymm3, ymm5 vroundpd ymm2, ymm8, 0 vroundpd ymm11, ymm13, 0 vsubpd ymm8, ymm8, ymm2 vfrczpd ymm2, ymm1 vfmaddpd ymm1, ymm10, ymm3, ymm0 vmovapd ymm10, YMMWORD PTR [r15+32] vfmaddpd ymm15, ymm15, ymm12, ymm2 vfrczpd ymm2, ymm1 vsubpd ymm13, ymm13, ymm11 vfmaddpd ymm0, ymm4, ymm3, ymm2 vsubpd ymm7, ymm7, ymm14 vroundpd ymm6, ymm9, 0 vmovapd YMMWORD PTR _a2$2$[rbp], ymm0 vmulpd ymm0, ymm14, ymm8 vsubpd ymm9, ymm9, ymm6 vroundpd ymm1, ymm10, 0 vfmaddpd ymm1, ymm1, ymm7, ymm0 vmulpd ymm0, ymm6, ymm13 vfrczpd ymm2, ymm1
04-06-2011, 06:28 AM
alpha754293

I thougth that VS was just the programming interface and that you still have to have the compilers separate from VS (or that it has to be like integrated/installed afterwards or something)?

I don't know. My prof used to use VS just to hit 'F5' to compile (or something like that) but otherwise, he used to use Compaq Visual Fortran, which became Intel Visual Fortran.

The very very very miniscule amount of programming that I've done, I've used gcc/g77 and did it using in cygwin.
04-06-2011, 07:02 PM
poke349

Quote:

Originally Posted by alpha754293

I thougth that VS was just the programming interface and that you still have to have the compilers separate from VS (or that it has to be like integrated/installed afterwards or something)?

I don't know. My prof used to use VS just to hit 'F5' to compile (or something like that) but otherwise, he used to use Compaq Visual Fortran, which became Intel Visual Fortran.

The very very very miniscule amount of programming that I've done, I've used gcc/g77 and did it using in cygwin.

Visual Studio has it's own compiler. It's actually pretty decent. VS 2010 is actually competitive with the Intel Compiler for optimizing code.

Also, I just released v0.5.5.9180 (fix 2).
The AVX binary should work on Bulldozer. But of course I won't know for sure until someone tries it.
04-07-2011, 12:02 PM
Boogerlad

is the vs2010 sp1 compiler faster in sse4.2 as well compared to the intel compiler? It would simplify my build script for open source emulators a lot.
04-07-2011, 12:31 PM
poke349

Quote:

Originally Posted by Boogerlad

is the vs2010 sp1 compiler faster in sse4.2 as well compared to the intel compiler? It would simplify my build script for open source emulators a lot.

I haven't used SSE4.2. It's mostly string-processing stuff - which isn't that useful for computation-heavy apps.

But for SSE4.1 and down, it is very very close. The Intel compiler is less than 1% faster than VS 2010 SP1 for y-cruncher.
At this point, it'll depend heavily on the type of code you're doing.

y-cruncher has a lot of very large blocks of branchless straight-line code. (often more than > 1000 instructions)
This is pretty typical of HPC apps.

Visual Studio has always sucked for this type of code.
The Intel Compiler has always been pretty good - if it's done right.
GCC seems to be competitive with the Intel Compiler - but it's hard to compare since it's Windows vs. Linux.

But now I'm tempted to think that VS2010 SP1 is much better at optimizing this type of code. Though I'm gonna need to run some benchmarks to see for sure.
04-08-2011, 05:53 AM
repman

poke, I have a question regarding instructions that y-cruncher uses. I noticed that on Intel CPU's it uses SSE4.1, on AMD it uses SSE3, but as I see it in CPU-Z AMD has SSE4A. I don't know much about this stuff, but is there any use of the SSE4A?
04-08-2011, 06:40 AM
poke349

Quote:

Originally Posted by repman

poke, I have a question regarding instructions that y-cruncher uses. I noticed that on Intel CPU's it uses SSE4.1, on AMD it uses SSE3, but as I see it in CPU-Z AMD has SSE4A. I don't know much about this stuff, but is there any use of the SSE4A?

SSE4a doesn't have anything that's useful for this application. Same applies to SSE4.2.
Also, I didn't have reliable access to an SSE4a machine until February this year.
04-08-2011, 11:42 PM
alpha754293

Quote:

Originally Posted by poke349

I haven't used SSE4.2. It's mostly string-processing stuff - which isn't that useful for computation-heavy apps.

But for SSE4.1 and down, it is very very close. The Intel compiler is less than 1% faster than VS 2010 SP1 for y-cruncher.
At this point, it'll depend heavily on the type of code you're doing.

y-cruncher has a lot of very large blocks of branchless straight-line code. (often more than > 1000 instructions)
This is pretty typical of HPC apps.

Visual Studio has always sucked for this type of code.
The Intel Compiler has always been pretty good - if it's done right.
GCC seems to be competitive with the Intel Compiler - but it's hard to compare since it's Windows vs. Linux.

But now I'm tempted to think that VS2010 SP1 is much better at optimizing this type of code. Though I'm gonna need to run some benchmarks to see for sure.

You do realize that you can kinda get gcc for Windows (via Cygwin).

I'm not sure how that all works out (whether it's emulated or whatnot--again not a programmer), but suffice it to say that for someone such as yourself, you might be able to make heads or tails of it and be able to at least run tests/benchmarks with it to be able to tell.

And for like some, they'd like recompile the kernel, the compiler itself, etc...so...it's a whole new giant can of worms.

But it IS available.

And I HAVE used it (a very insy tinsy tiny wee bit) with g77 (GNU Fortran 77 compiler), nothing like what you're doing. (Not even remotely CLOSE to it).
04-09-2011, 12:23 AM
poke349

Quote:

Originally Posted by alpha754293

You do realize that you can kinda get gcc for Windows (via Cygwin).

I'm not sure how that all works out (whether it's emulated or whatnot--again not a programmer), but suffice it to say that for someone such as yourself, you might be able to make heads or tails of it and be able to at least run tests/benchmarks with it to be able to tell.

And for like some, they'd like recompile the kernel, the compiler itself, etc...so...it's a whole new giant can of worms.

But it IS available.

And I HAVE used it (a very insy tinsy tiny wee bit) with g77 (GNU Fortran 77 compiler), nothing like what you're doing. (Not even remotely CLOSE to it).

I used cygwin GCC and Dev-C++ way back in my late high school and early undergraduate days. Then I "discovered" Visual Studio and found it a lot easier to use. :ROTF:

I've been too lazy to benchmark cygwin GCC now. I remember setting up the environment was a pain in the @$$. And then having to hook it up to Dev-C++ to override its archaic version of GCC.
Visual Studio was just double-click setup.exe and it's all ready - GUI and everything. Beats the hell outta all the Linux solutions... :rofl:

I think my productivity went up like 3-fold when I switched from cygwin GCC to Visual Studio 2005.
The biggest part was the integrated debugger that would jump directly to the line in the source code that crashed - even with optimizations enabled.
It's a huge time-saver since it eliminates a lot of the guess work. There's a way to setup something like this in Linux, but I hear it's nightmare to setup unless you're experienced.
05-10-2011, 10:50 PM
poke349

Thought I'd give this thread a little bump - since it fell off the first page.

I've been busy as hell this school year, so I haven't been able to make too many improvements.

I have a slightly faster base conversion code. But it's ram only for now, I haven't extended it to swap yet.
Current Conversion Code (left): 1.837 + 65.386 seconds
New Conversion Code (right): 44.487 seconds
http://www.numberworld.org/y-crunche....6_preview.png

This new base conversion code won't be enabled until v0.6.x.

I've been working a completely new bignum interface. It's better, faster, more flexible, and more future-proof than the one y-cruncher currently uses. :D But it's also largely incompatible with the existing one. :(

This new base conversion is written using the new interface. For that screenshot above, I had to compile both the old and new interfaces together and then "hack" the new base conversion into the old code. :rofl:
It works, but it's ugly and I don't think it's "stable" enough for production. It's gonna take a while to migrate y-cruncher to the new interface.

Sometime over the next year, I'll probably be releasing a Dynamic Link Library the exposes some of the low-level functions use by y-cruncher.
It'll be for Windows first, but I'll support Linux later.

Here's a couple of tiny sample programs. Both examples are under 200 lines and use OpenMP and "ymp.dll". YMP is the name I've given to the library.
Sometime before I release the library itself, I'll be releasing the interface/documentation as well a large number of sample programs (including these).
http://www.numberworld.org/y-crunche..._5_11_2011.png

If anyone's looking for a factorial benchmark... :wasntme: (20,000,000! is 137 million digits long!!!)
The other example computes e to a billion digits. It's not optimized at all - yet is almost as fast as y-cruncher.

Once I release the library, I'll be curious to see how people (if any) will use it.
Unless someone beats me to it, it'll probably be the first public bignum library to have multi-threading. :)
05-12-2011, 01:48 AM
Demo

I tried this today and it seems to run 2c cooler than Prime95 26.6 but I do see that my VID under RealTemp is going from 1.3811 to 1.4111v basically its asking more juice!

For overclock stability test how many runs under stress test would you consider as 24/7 stable for gaming/movies/internet ?
05-12-2011, 12:24 PM
Meizuman

A test with an old unpopular CPU :up:

http://i270.photobucket.com/albums/j...uncher250K.jpg
05-13-2011, 10:31 AM
poke349

Quote:

Originally Posted by Demo

I tried this today and it seems to run 2c cooler than Prime95 26.6 but I do see that my VID under RealTemp is going from 1.3811 to 1.4111v basically its asking more juice!

For overclock stability test how many runs under stress test would you consider as 24/7 stable for gaming/movies/internet ?

It would depend on how stable you want. I typically run a combination of y-cruncher/prime95/Linpack for several hours, and then pull the frequency down by about 5 - 10%.
But that's just me. I do a lot of compiling work, so I can't afford any instability at the wrong times.

Quote:

Originally Posted by Meizuman

A test with an old unpopular CPU :up:

http://i270.photobucket.com/albums/j...uncher250K.jpg

Neat. :) I'll update these over the weekend.
05-13-2011, 12:56 PM
theELVISCERATOR

1 Attachment(s)

neat program

This any good? Ran at 4500mhz...

Attachment 114333
05-17-2011, 10:49 PM
poke349

XS went down on Sunday or so... then I got tied down with other things.
I'll try to update the lists tomorrow.

btw, for any software devs interested in very large multi-precision arithmetic...
I've launched the website for the YMP library. Though it's still gonna be a while before it's ready for release.

http://www.numberworld.org/ympfl/

And here's another screenshot of the library.
This is computing Pi to 100 million digits using the same (slow) algorithm that SuperPi uses.
The source code for this is on that webpage (along with other examples), but you won't be able to compile it without the YMP library itself.
http://www.numberworld.org/ympfl/samples/pi_agm.png
05-23-2011, 09:40 AM
Alpha

i love big numbers! i'll be shooting you an email later in the hopes i get to play with the library! I can see multi-prime getting a bump....
05-23-2011, 11:22 PM
poke349

Quote:

Originally Posted by Alpha

i love big numbers! i'll be shooting you an email later in the hopes i get to play with the library! I can see multi-prime getting a bump....

Got your email. I'll respond properly tomorrow.
I was at ACEN (Anime Central) the whole weekend... :D then I crashed for all of today. :rofl:

So my sleep schedule is completely #$%^ed up right now... :ROTF:

Lesson: Do not drive 85mph on Interstate 57 with a 30mph crosswind when you're not fully awake... :rofl::rofl::rofl:
(I AM alive btw...)
05-24-2011, 05:16 AM
Alpha

i was reading through some of the documentation you have for you library and am i correct when i see that you have not implemented the modulo function?
05-24-2011, 05:52 AM
poke349
Quote:

Originally Posted by Alpha

i was reading through some of the documentation you have for you library and am i correct when i see that you have not implemented the modulo function?

Integer Modulus has been implemented, but it's not very well optimized.

There's two versions of it:

Modulus by a small integer (< 2^32):
This one is very slow (several times slower than GMP). That's because y-cruncher doesn't need this function at so there's no efficient implementation of it.
But for completeness, I had to implement it even if it's slow.

Modulus by a large integer (no size restriction):
This one is fast for large operands. (both operands > 1000 digits)
But there are some cases where it's horrifically slow compared to other libraries:
- When the operands are very small. (< 100 digits)
- When the dividend it much longer than divider.
y-cruncher doesn't use integer division (yet), so this function hasn't been properly optimized.

When I say "fast" or "slow", I mean fast/slow compared to other libraries. Larger operands will almost always be slower (in run-time) than small operands.
06-12-2011, 11:53 AM
poke349

I've opened the documentation (for the YMP library) to the public.

It's here:
http://www.numberworld.org/ympfl/doc...n/general.html

I've also released the source code for the rest of the sample programs:

http://www.numberworld.org/ympfl/#Samples

You still won't be able to compile them since I haven't released the library yet. But it's something to look at for the curious.
The "pi_chudnovsky.c" example beats gmp-chudnovsky 4-to-1 on a Core i7. It's almost as fast as y-cruncher and it isn't even optimized. :D
06-12-2011, 02:02 PM
Morphling

Here is my own rig:
http://img4.uploadhouse.com/fileuplo...bfc43e073c.jpg
Not so much, but it was just to participate ;)
06-13-2011, 09:49 AM
poke349

Cool, I'll update this in a bit.

Show 100 post(s) from this thread on one page

All times are GMT -8. The time now is 08:10 PM.

XtremeSystems