Always copied, never equalled :)
:lsfight:
Printable View
2.0ghz 8 core sounds nice but I know it can't over clock :/
Ok guys, I've got this thing up and running on a test basis although my case isn't here yet. Currently running Windows Server 2008 R2 HPC Edition beta. I build these systems to do work and not really much for benchmarking, but if anyone has something they'd like to see run let me know.
there is only one benchamark i am interested in and thats wcg. Would you mind running it lets says for week just to see how it can do ?
Currently it looks like this:
4x Opteron 6174
16x Kingston KVR1333D3D4R9S DDR3 1333 CAS 9 ECC Ram
Supermicro H8QGi-F board
Using onboard G200 video for now.
Seagate 7200.12 1TB SATA drive
Cheapo DVD burner
I ran cinebench 11.5 last night but do not remember the score. I will run it again later today. I did find out though that it is limited to 32 cores. :mad:
nice man, how bout any benches that can use all them cores! wprime? this one http://www.xtremesystems.org/forums/...d.php?t=202139 ? :banana::banana::banana::banana: i have no idea what benches use that many, if it wasnt for you i probably wouldnt find out for like 10 more years :p:
A screenshot of the rig with open case is all I ask, beaker7 :)
Cinebench 11.5 hitting 32 core limit...:(
http://files.me.com/dougmeyer/k7ittu
Try wPrime 32m and 1024m..
Don't forget to go into advanced settings and set threads to 32 as it's 2 by default..
weird indeed... but nice work rig beaker
what kind of work your into???
1.55 is the "accepted" version to use and gives a time thats comparable to the others..
On his CB 11.5 score we can see the potential power here with that 19+ score.
Now think, thats just using 32 of his 48 cores, call it 70% for the sake of arguement.
My SR2 rig at 4350 got 21.55 points..
He's got a beast there..That's why I want to see wPrime..:D
With it set to 48 threads...
http://files.me.com/dougmeyer/0wskj9
With it set to 32 threads...
http://files.me.com/dougmeyer/zn2ask
With it set to 16 threads...
http://files.me.com/dougmeyer/puica1
what also interesting is that the 1024 times go the opposite direction.
it seems like for the 32m it cant "engage" all the cores fast enough
Interesting, I'd have thought better..
Your 1024m time isn't bad with 48 threads but the others are slower than I expected.
Honestly I expected to see you beat my 58.5s 1024m time.
could i use this for gaming.?
Interestingly WCG and its benchmark only seem to see 16 cores.
This is.....breathtaking. Hopefully you figure out wprime and put the smack down.
Best I can do right now is crappy phone pic.
System is not yet mounted in a case...
http://files.me.com/dougmeyer/9i3iho
beastly setup beaker, thanks for sharing. Can't wait to see more testing.
im too, looking forward benchamrks much :)
Tell you what would be interesting--7zip. Run the latest beta and test out LZMA2 compression. We'll need common settings and a common source file for the results to mean much, though.
Also interesting: Generic GPU Bench
A small update. I installed the Quadro card, and disabled the onboard G200 Matrox and got this:
http://files.me.com/dougmeyer/395t1q
On thing that seems strange to me is that CPUz lists the memory as Dual Channels instead of quad. I have 16 DIMMs installed. To test if this was affecting performance I took out 8 DIMMs and the benches were about the same.
http://files.me.com/dougmeyer/74iald
http://files.me.com/dougmeyer/ivg30o
http://files.me.com/dougmeyer/wbwgml
well your 1024m beats that 3.0ghz 24 threads one we have here :) so that's good.
I guess that Gt200 wasn't doing very much there it got off loaded to cpu :(
so it does look like 32m can't engage enough cores fast enough I guess clock speed will help that one the most in the end :/
Unfortunately WCG still only sees 32 cpus after updating the number in my device profile...
:shrug:
http://files.me.com/dougmeyer/6eq773
omg, this is awesome score
This might help:
Here is a link to the info:Quote:
NB: Reports on the forum disclosed that BOINC may not automatically run more than 32 threads simultaneous. This can be overcome by creating a cc_config.xml control file , or adding the line <ncpus>48</ncpus> to said files <options> section. See discussion: Maximum Number of CPU's that BOINC can see?
http://www.worldcommunitygrid.org/fo...ffset,0#254638
The last post has the info needed
A start..Got it running properly today.
Tried XP Pro 64 yesterday and itshowed 2cores.. Yes..TWO..:rofl:
Win 7 Ultimate 64 fixed the issue.
Now how to get it "up to speed"
Tried AMD Overdrive and the only thing that wasn't grayed out was the HTT option..Moved that one notch and BANG..reboot..
All thoughts ideas and suggestions are welcome.:D
http://img268.imageshack.us/img268/9682/magnycours1.jpg
SetFSB?
You'd need to know the clock controllers though.
Well...that sucks.
It was a far off guess...not like I actually thought it would work myself.
You could ask ket to mod the BIOS...if its possible. Of course it would take time though.
That isn't going to help much. Opterons are all locked and will simply ignore any writes to the P-tables that specify a FID or VID out of the range the CPU is programmed to accept. The value is overwritten instantly. Since you can't change HT Ref from there either, there is no facility for overclocking an Opteron using K10Stat.
You'll have to find a register that changes your HT ref generator's output, if it's even on an editable bus.
nice guys, now we can try Cinebench R10 and R11 ? 10pts?:)
Finally up and running :up:
I hope you will crack HTT clocking problem Movieman!
Cinebench running 48 cores...
http://files.me.com/dougmeyer/by92oi
lol, this is crazy, 26pts :), n1 !
That is an amazing score. If we could only overclock those Magny Cours to 3+Ghz now.
hehe 4 of these 5+ ghz :eek:
On LHe!
AMD should make a statement with their processing power...they should design a custom 4P or 8P board with 6+2 phase per CPU and huge overclocking capabilities.
Or release and Extreme server CPU with a really high TDP/clock :ROTF:
Sadly, when you are a sensible person everything I just said is crazy talk. :shrug:
damn 26 points in Cinebench R11.5....nice beaker
And I was thinking my 7.5+ points is fast :rofl:
Massive performance out of stock setup :clap:
I think it is time to solder some new quartz to that board :p:
I've been following this thread for two weeks, but only got post privileges today...
SuperMicro H8QG6-F motherboard, integrated video, bios version 1?, bios date April 2010, supermicro sc484 chassis, 4x opteron 6172 (2.1ghz), 64gb ecc ram, 2k8 server R2 trial (64-bit). $9 cpu heatsinks (startech fank8am2) from microcenter, modified to fit width.
wprime 2.03, 1024m test with 48 threads:
Best: 52.353 seconds.
Usual: 53 - 54 seconds.
wprime 1.55, 1024m test with 48 threads:
Best: 52.103 seconds
Usual: 52.5 - 54 seconds
cinebench 11.5 set to 48 threads:
Best cpu test: 24.99
Usual: 24.52 - 24.94
I'm not sure how many threads cinebench is running. Cinebench insists that there are 32 cores in the system. I manually set the preferences to 48 threads, but if I check the preferences a second later, it's back at 32 threads. So i'm not sure if the cinebench results are based on 32 or 48 threads.
Based on another thread, I tried the k10stat program. No go. I eventually figured out that you can only oc the engineering samples.
Hey thanks simplesam, post up some screenshots :)
4x opteron 6172
wprime 32m at various thread counts.
Threads Time
12t 6.851
16t 5.473
20t 4.868
24t 4.511
28t 4.292
32t 4.134
36t 4.165
40t 4.365
44t 4.289
48t 4.37
Both 32m and 1024m at 48 thtreads, current run. 2010-06-15, 1pm
http://i879.photobucket.com/albums/a...rime48th-1.jpg
I wonder how efficient wprime is creating and deleting the threads. In the 48t 32m test, the first thread finished at 1.638.
Power management seems to clock these cpus down very quickly. The 1024m scores were much higher and had a huge swing from one run to another, until I turned off power management in windows.
4x opteron 6172
http://i879.photobucket.com/albums/a...ebench48th.jpg
I do have a kill-a-watt, but I still can't test it. The chassis has triple 900w power supplies. It is supposed to draw from two of them, and use the third as a hot spare.
But I had 25 drives in it last night, + a monitor, + a few other things on the same circuit, and I didn't blow the 15 amp fuse. I have blown that circuit with 2 poweredge 2950s + the other things on circuit.
ufff, next nice scores (we are loosers wuth Thubans now :-D )
But I don't know which two are powering the system... Here's what I did. Wall, kill-a-watt, 1-to-3 extension cord, server.
No pci cards installed, one drive, ipmi chip off, sas2008 chip on, sasx36 expander installed and on, 4x cpu fans, 2x integrated chipset fans, 7x "make you deaf" chassis fans, 16 dimms, windows power profile: performance.
The power supplies have fans that run, even when the system is off.
The chassis fans are so loud, I couldn't tell you if the power supply fans ramp up when the server starts.
Three power supplies only, system off: 72w
Booting peak: 540w
Idle, logged on via rdp: 370w
wprime 1024m running: 714w
Movieman, what does your 2-way system pull?
simplesam can you also try the same test with balanced power plan?
Thanks!
BTW monstrous setup! What will be it target workload when you finish playing with it?
Any chance you guys can watch the task manager while making a run on Cinebench and see if it's actually using all 48 cores?
It looks like it is... All 48 cores hit 100% when set to 48 threads. But all 48 cores hit 100% when it's at the default 32 threads too.
Balanced idle: 340 watts.
Because of all the fans and such, this probably isn't the best system to test for low power.
I planned on using for server consolidation, but I may need to sell it due to a sudden change in finances.
I had an OS crash early in my initial setup, and the previous results are from that config.
A fresh install returns consistent wprime 32m results around 3.85, and 1024m around 51.4
But now I've found that I have a bad stick of ram.. Waiting on a replacement.
Hey guys, its finally time to get this thing off the bench and into a case as my custom MMODS case finally came in for it.
To that end, I'm looking to replace the ludicrously loud fans that come with the Dynatron A6. Does anyone have experience using a fan adapter to upgrade them to something larger?
Any recommended fans that will perform similarly but be much quieter?
http://www.dynatron-corp.com/en/prod...v=&id=214&in=0
Overall Dimension 114.0 x 75.5 x 66.7 mm
Fan Dimension 77 x 77 x 20 mm
Speed At Duty Cycle 0~20%: 1000 RPM
At Duty Cycle 50%: 2600 RPM
At Duty Cycle 100%: 5500 RPM
Bearing 2 Ball Bearing
Rated Voltage 12 V
Power At Duty Cycle 0~20%: 0.54 W
At Duty Cycle 50%: 1.68 W
At Duty Cycle 100%: 9.0 W
Air Flow At Duty Cycle 0~20%: 8.36 CFM
At Duty Cycle 50%: 25.5 CFM
At Duty Cycle 100%: 53.9 CFM
Noise Level At Duty Cycle 0~20%: 20.6 dBA
At Duty Cycle 50%: 35.1 dBA
At Duty Cycle 100%: 51.4 dBA
Air Pressure At Duty Cycle 0~20%: 0.24 mm-H2O
At Duty Cycle 50%: 1.25 mm-H2O
At Duty Cycle 100%: 5.59 mm-H2O
I used Noctua NH-U12DO A3 CPU Cooler (Socket G34/C32) for my Supermicro H8QG6-F (Socket G34)
Cause I want to have a silent system. The other CPU Cooler you can get are build for 1HE/2HE server or something else.
This kind of cpu-coolers are to noisy in stead of 4300 RPM up to 6500 RPM.
The noctua coolers are not cheap but itīs very good quality.
Regards
O-Brian
2 year old thread is 2 year old.
Sorry for the necromancing ...
This is a bump of an old bump...
I started laughing when I read that... I have had my 4p G34 ES rig at 3ghz since July of last year. I was running a 2p before that at 3.1ghz.
Now after a system rebuild to a heftier SM board the GL I am currently at 3.48ghz... and yes its ballz.
I am just finishing its overclock... I will try for more this weekend... Sig is original build.
If only 3ghz mc came sooner.
Hello I use NOCTUA CPU-Cooler NH-U12DO A3 for Cooling of my 4 Opteron 6173
The Noctua Cooler are very silent and are working for 4 Years now. (24/7)
You can use the Noctua NH-U9DO (90mm x 90mm) too (in my second Server)
The Dynatron Coolers are too noisy with a speed of 5300 RPM
Regards O-Brian
Ooof, Movieman . . .