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

[mAJORD]

*bumpety*

So hows it coming along, Poke?

Looking fwd to the next version.

[poke349]

Thanks for the bump!

The program is looking good, but testing has been drastically slowed down because I'm RMA'ing one of the cpus in my "Nagisa" rig.

It started acting up on Friday and within a few hours, it was dead...
y-cruncher crashed in one of the tests. Thinking it was a bug, I tried to reproduce the crash... but instead, I got a BSOD... (so it sure as hell wasn't the program's fault)

After a few reboots and prime95 tests with different processor affinity settings, I finally confirmed that one of the processors was dying. Then within a few more reboots, it wouldn't post anymore. And it refused to post until I removed that dead processor - confirming it was dead.

Never overclocked... Never overheated... Almost always below 65C. The highest it ever got was 85C and only for a few seconds before I realized I didn't use enough thermal paste...
So I probably got a bad chip to begin with. It lasted only 9 months of near-contiguous 100% CPU computation.
I'm not exactly sure how Intel expects these to be used... But these ARE workstation chips, so they should be able to last years of torture @ stock.

So it's RMAing right now, and I'm running the rig with only one chip... (yeah... 4 cores with 64GB of ram... how's that for a spectacular ram/cores ratio?)

EDIT: It's not the CPU, it's the motherboard. Apparently the mobo is no longer able to supply sufficient power to that one socket. So it's motherboard RMA... In the meantime, I need to build a new computer to use... My laptop is waaaaaay toooooo crappy for code-testing. (it can't even play HD movies or Anime...)


Enough sad stuff... Version 0.4.1 should be out within another week or two...

As far as changes go:

• 2-8% faster for Pi, improvements for the other constants range from no improvement to 20%
• Added SuperPi benchmark sizes from 1M to 128G
• Current benchmark sizes extended to 100b
• Thread limit increased from 64 to 256 since Win7-based Windows Server now supports more than 64 cores.
• Re-added the digit comparision feature
• New Constant: generic square roots
• New Constant: Golden Ratio
• Fixed about a billion bugs in the x86 builds...
• Fixed a bunch of other rare bugs...

Nothing really spectacular though. I've stopped working on the program since mid-June and I'll be resuming in October (for a number of reasons). Currently the only changes I'm making are fixes. Most of the time since the previous release was spent rewriting a lot of code in preparation for a very large feature sometime in the future.

*As a side note: I haven't actually tested any of the benchmarks above 10b... I don't have that kind of ram.
But maybe somebody with a pretty looking rack-server will be able to.

Lastly (kinda technical though): There's a possibility that the program won't work properly above 41 billion digits - since that's the point where a number of variables will cross the 32-bit mark (and I'm not 100% sure I'm using 64-bit integers in all the places that I should be).

[Warboy]

Intel Core i7 920 @ 3980MHz (21 * 190)

[SquattingDog]

Hey guys, thought I would post up some results obtained from a couple of systems kicking around the place here currently. Got a couple more, but ran out of time as it's 2AM and I'm working tomorrow morning o.o

The 5200+ cannot complete validation - may not be stable, need to test and tune that a little more lol.

[SquattingDog]

And some more...

The last two are virtually identical machines - Gigabyte X48-DS5 Motherboards, Q6600's @ 3.2GHz, 4GB G.Skill PQ DDR2-1000 Memory @ DDR2-960...you will see the missus' machine beats mine just marginally *shakes fist*.

Thought it was worth having a look at other machines too, not just i7 and dual-socket beasts

Oh and that Atom in the previous post was running multithreaded mode, I will OC it past 2GHz later in the week, and post the results. Will also re-run the 25M place test on the P4 2.8Ghz...image got corrupted.

[poke349]

Thought it was worth having a look at other machines too, not just i7 and dual-socket beasts

I like that attitude. I haven't seen a P4 in a while.

The 5200+ cannot complete validation - may not be stable, need to test and tune that a little more lol.

It's probably a bug in the validation itself - probably being too aggressive with anti-cheat protection. But unless I can reproduce that on a machine that I have access to, I can't fix it.

Hardware errors don't give that message.

When a hardware error occurs, one of 4 things can happen:

1. Crash or BSOD - just to state the obvious...
2. y-cruncher catches the error, corrects it, and moves on. Validation will still fail. (see picture below)
3. y-cruncher catches the error, is unable to correct it, and prints an error message.
4. y-cruncher does not catch the error, and it tells you that the digits don't match at the end.

When y-cruncher catches and corrects an error, it will look like this:

If you're curious, this screenshot was generated by intentionally introducing an error into the computation via the source code.
I don't have access to any unstable machines, and most hardware errors usually end with a BSOD.
So no, this feature hasn't actually been "truely" tested before.

The sanity check error that it gave you can (but not always) show up under the following circumstances:

1. Either the system clock or the BIOS clock has been tampered during the computation.
2. The program has detected an abnormal frequency* - possibly caused by time-slowing cheats.
3. The binary has been tampered with.
4. The base clock or the FSB has been tampered with.**

*Note that speed-step and any CPU throttling/power-saving feature does NOT trigger this. (I've made sure of that.)
**This is an unwanted side-effect of the anti-cheat protection. As a result, SetFSB and similar tools may not work as they may trigger a sanity check error.

It is VERY possible that there are other "legit" things that could set off a sanity check error.
I've fixed all the things I know of that I could reproduce on my machines. But I don't have a lot of machines to play with, so it's very likely that it's still buggy.

[Hoss331]

[SquattingDog]

It's probably a bug in the validation itself - probably being too aggressive with anti-cheat protection. But unless I can reproduce that on a machine that I have access to, I can't fix it.

Hardware errors don't give that message.

When a hardware error occurs, one of 4 things can happen:

1. Crash or BSOD - just to state the obvious...
2. y-cruncher catches the error, corrects it, and moves on. Validation will still fail. (see picture below)
3. y-cruncher catches the error, is unable to correct it, and prints an error message.
4. y-cruncher does not catch the error, and it tells you that the digits don't match at the end.

The sanity check error that it gave you can (but not always) show up under the following circumstances:

1. Either the system clock or the BIOS clock has been tampered during the computation.
2. The program has detected an abnormal frequency* - possibly caused by time-slowing cheats.
3. The binary has been tampered with.
4. The base clock or the FSB has been tampered with.**

*Note that speed-step and any CPU throttling/power-saving feature does NOT trigger this. (I've made sure of that.)
**This is an unwanted side-effect of the anti-cheat protection. As a result, SetFSB and similar tools may not work as they may trigger a sanity check error.

That 5200+ machine has been priming all night, and is still going with no issues. I had also passed LinPack (OCCT) and other stress-tests run on it, so it certainly appears to be stable.

I have just run this on my netbook again @ 2.08GHz, and it runs a LOT quicker, but I get the validation error as SetFSB is used to overclock. I will post the results later, at work on a dumb-terminal atm so cant.

Thanks for putting the time an effort into making a great program, I will continue to use this as my benchmark for speed of CPUs/OCs

[poke349]

Wow... 4.46GHz. On air? water?

EDIT: I just noticed that your old entry was clocked higher. What did you change?

Thanks for putting the time an effort into making a great program, I will continue to use this as my benchmark for speed of CPUs/OCs

Thanks for the complement.

The program was actually originally written for an entirely different purpose. After that was done, I realized that it could make a decent multi-threaded Pi benchmark. So I added the benchmarking features just for that purpose (though I'm incompetent in this area of programming so the benchmarking validation stuff is still very buggy).

[Hoss331]

Wow... 4.46GHz. On air? water?

EDIT: I just noticed that your old entry was clocked higher. What did you change?

The old run was done on air and slower ram. Im on water now with LV blades which are faster and also have better timings. This is my daily clock now mainly to keep the ram where I want it but it can go alot higher if needed, chip can do 5ghz.

50

[spdycpu]

Ditched the DFI X58 T3eH8 and got the EVGA Classified.

Here is 4.4GHz (shows 4.2 due to turbo), 3.6GHz Uncore and 1.65GHz memory on the x64/SSE3 binary:

Begin Computation:

Computing: Pi

Algorithm: Chudnovsky Formula

Summing Series:
Time:    6.339 seconds  ( 0.002 hours )
InvSqrt...
Time:    0.335 seconds  ( 0.000 hours )
Final Multiply...
Time:    0.158 seconds  ( 0.000 hours )

Compute Pi Time: 6.833 seconds  ( 0.002 hours )

Constructing Base Conversion Table:
Time:    0.475 seconds  ( 0.000 hours )
Base Converting (Primary Cutting Parameters):
Time:    1.575 seconds  ( 0.000 hours )

Writing Decimal Digits:   25,000,001  digits written


Total Computation Time:  8.886 seconds  ( 0.002 hours )



Benchmark Successful. The digits appear to be OK.

Program Version:    0.3.2 Alpha Build 6953 (fix 1) (x64 SSE3)
Processor(s):       Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz
CPU Frequency:      4,190,005,359 Hz  (frequency may be inaccurate)
Thread(s):          8
Digits:             25,000,000
Total Time:         8.8856 seconds
Checksum:           1c81bbd4c3983fa20937e66c692cdb33

Begin Computation:

Computing: Pi

Algorithm: Chudnovsky Formula

Summing Series:
Time:    13.989 seconds  ( 0.004 hours )
InvSqrt...
Time:    0.637 seconds  ( 0.000 hours )
Final Multiply...
Time:    0.335 seconds  ( 0.000 hours )

Compute Pi Time: 14.962 seconds  ( 0.004 hours )

Constructing Base Conversion Table:
Time:    0.923 seconds  ( 0.000 hours )
Base Converting (Primary Cutting Parameters):
Time:    3.368 seconds  ( 0.001 hours )

Writing Decimal Digits:   50,000,001  digits written


Total Computation Time:  19.257 seconds  ( 0.005 hours )



Benchmark Successful. The digits appear to be OK.

Program Version:    0.3.2 Alpha Build 6953 (fix 1) (x64 SSE3)
Processor(s):       Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz
CPU Frequency:      4,190,005,871 Hz  (frequency may be inaccurate)
Thread(s):          8
Digits:             50,000,000
Total Time:         19.2569 seconds
Checksum:           b5d98352d1a01c7b08e6c562c61b1681

Begin Computation:

Computing: Pi

Algorithm: Chudnovsky Formula

Summing Series:
Time:    30.866 seconds  ( 0.009 hours )
InvSqrt...
Time:    1.269 seconds  ( 0.000 hours )
Final Multiply...
Time:    0.677 seconds  ( 0.000 hours )

Compute Pi Time: 32.813 seconds  ( 0.009 hours )

Constructing Base Conversion Table:
Time:    1.801 seconds  ( 0.001 hours )
Base Converting (Primary Cutting Parameters):
Time:    7.341 seconds  ( 0.002 hours )

Writing Decimal Digits:   100,000,001  digits written


Total Computation Time:  41.962 seconds  ( 0.012 hours )



Benchmark Successful. The digits appear to be OK.

Program Version:    0.3.2 Alpha Build 6953 (fix 1) (x64 SSE3)
Processor(s):       Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz
CPU Frequency:      4,190,005,407 Hz  (frequency may be inaccurate)
Thread(s):          8
Digits:             100,000,000
Total Time:         41.9623 seconds
Checksum:           3ce11a173665c33b075c7f12f712e89c

Begin Computation:

Computing: Pi

Algorithm: Chudnovsky Formula

Summing Series:
Time:    87.054 seconds  ( 0.024 hours )
InvSqrt...
Time:    3.123 seconds  ( 0.001 hours )
Final Multiply...
Time:    1.724 seconds  ( 0.000 hours )

Compute Pi Time: 91.902 seconds  ( 0.026 hours )

Constructing Base Conversion Table:
Time:    4.640 seconds  ( 0.001 hours )
Base Converting (Primary Cutting Parameters):
Time:    20.603 seconds  ( 0.006 hours )

Writing Decimal Digits:   250,000,001  digits written


Total Computation Time:  117.163 seconds  ( 0.033 hours )



Benchmark Successful. The digits appear to be OK.

Program Version:    0.3.2 Alpha Build 6953 (fix 1) (x64 SSE3)
Processor(s):       Intel(R) Core(TM) i7 CPU 920 @ 2.67GHz
CPU Frequency:      4,190,005,727 Hz  (frequency may be inaccurate)
Thread(s):          8
Digits:             250,000,000
Total Time:         117.163 seconds
Checksum:           57109bb3167de5e60a525a0210a665ba

[poke349]

Ditched the DFI X58 T3eH8 and got the EVGA Classified.

Wow. New mobo? What did you do with the old one?

All your times beat my workstation. Congrats

[Newblar]

heres my newb system, on a gigabyte 790gx ds4h mainboard


i was wondering if this prog could be used as a stability test? it seems it would be very good that

[spdycpu]

Wow. New mobo? What did you do with the old one?

All your times beat my workstation. Congrats

Thanks Perhaps not at the longer runs, I'll have to test those later on.

It is sitting here in the original DFI box, I guess I'll sell it at some point. I really want to get some 4.5-4.6GHz runs but my PSU only has one 8-pin connector and the board has two. Obviously it'll function on one but I'm hoping the second should help a little. I'm about to run up to Fry's to snag one of those 4 pin molex to 8-pin connectors. Also 4.4 was w/o ice in the reservoir so I'll try that as well.

[poke349]

Thanks Perhaps not at the longer runs, I'll have to test those later on.

It is sitting here in the original DFI box, I guess I'll sell it at some point. I really want to get some 4.5-4.6GHz runs but my PSU only has one 8-pin connector and the board has two. Obviously it'll function on one but I'm hoping the second should help a little. I'm about to run up to Fry's to snag one of those 4 pin molex to 8-pin connectors. Also 4.4 was w/o ice in the reservoir so I'll try that as well.

What kind of temps are you hitting at those speeds (I know you're running water.)? 4.4GHz is already very impressive. And you sustained it long enough to do 250m.

My suitemate last year (Serotonnin) had trouble getting past 4.1GHz on air. The 500m run we did at 4.1GHz hit 92C on air with a 133cfm Ultra Kaze. He has a C0 and we needed 1.45volts to "barely" get stability @ 4.1.

4.2 boots into windows, didn't last more than 30 seconds of y-cruncher - and we didn't dare try prime95.
4.3 posts, no boot.
4.4 no post.

turbo disabled in all cases.


My new toy is a D0, and it's rock stable at 3.2 @ stock voltages. I haven't pushed it to see how far it'll go @ stock voltage. Right now, I'm running it at 3.34 GHz or (3.5 = 167 x 21 with turbo) with a slight vcore bump just to be safe. Under prime95 it peaks at 78C on the "stock fans" of a CM Hyper N520. I plan on getting some better fans - something I can actually hear...

My workstation went down... I'm guessing the MOSFET(s) for the top socket couldn't take anymore 24/7/52 150-watt power draw. I'll be RMAing it in a couple weeks, but in the meantime I needed a decent computer so I had to build this little comp. (by little, I mean physically small, not "crappy")
It's small enough to hand carry on a plane so I'll be taking it back to college.
It will also take over as my gaming, and programming rig since it games better and compiles faster. My workstation will stay at home (remote-controlled from school) and be dedicated to "extreme code-testing".

Benchmarks coming soon... Hopefully I can kill enough services to do a 2.5b run - which might be the first 2.5b done on a single-socket machine on record.

v0.4.1 should be out next week. Testing is just about done and now we're working on a new layout for the site. Once that's done too, I can release v0.4.1.

[poke349]

heres my newb system, on a gigabyte 790gx ds4h mainboard


i was wondering if this prog could be used as a stability test? it seems it would be very good that

No such thing as newb. Great to see some more older rigs.

It "can" be used as a stability test, but only if you do large runs or if you run multiple instances of the program. Even though it's multi-threaded, it isn't quite able to keep all cores busy at all times. Two instances is usually enough to keep all the cores busy. And under the right conditions, it will run as hot as prime95.

The biggest problem keeping it from being a good stability test is that it needs too much memory and it runs too quickly. So even if you're doing the largest computation that will fit into ram, it will be done in "a few minutes".

[StevensDE]

Hi ,

here a benchmark with a DELL Precision T7400 with 2x Intel Xeon E5420. (non oc, all default)













Why is my performance on a real 8 Core machine so bad in in comparison with a "little Core i7 920 250 Dollar CPU".. ? This is only one physical quadcore? And faster than a 3500 Dollar Dell Precision T7400 ???

[spdycpu]

Hi ,

here a benchmark with a DELL Precision T7400 with 2x Intel Xeon E5420. (non oc, all default)

Why is my performance on a real 8 Core machine so bad in in comparison with a "little Core i7 920 250 Dollar CPU".. ? This is only one physical quadcore? And faster than a 3500 Dollar Dell Precision T7400 ???

I believe this is due to you having the older quad core. It does not have an integrated memory controller like the i7, yours is on the chipset rather than the processor. Also the i7 is triple channel DDR3, I've seen memory reads in excess of 29gb/s. I think the main problem with the Core 2 Quad and the Xeons based off the Core 2 Quad is the older bus. Even a single Core 2 Quad has two dual cores under the heat spreader. For example, Core 0 and 1 is a die, Core 2 and 3 is another. If core 0 and 1 do the work it is quick, if core 0 and 3 do the work the speed is hindered quite a bit due to the traffic still needing to go to the bus. Also since you have two quad cores the bus gets saturated with that kind of traffic, it isn't quite as efficient as the AMD and i7 bus.

Also what helps is the "little" i7 was overclocked from 2.66GHz to 4.4GHz and has the benefit of Hyperthreading. The L3 cache was also overclocked from 2.4GHz to 3.6GHz. My i7 was $199, not too bad of a price/performance ratio I think.

[spdycpu]

What kind of temps are you hitting at those speeds (I know you're running water.)? 4.4GHz is already very impressive. And you sustained it long enough to do 250m.

At 4.4GHz, 1.38v it was around 75-80c load. This is on an Apogee waterblock, custom hold down plate and a 370 GPH pump with a 30 pass radiator (copper tubing, aluminium fins). I really need a new waterblock, this old Apogee just doesn't cut it. Water temps are usually pretty good though, that radiator is a heater core out of a car.

My suitemate last year (Serotonnin) had trouble getting past 4.1GHz on air. The 500m run we did at 4.1GHz hit 92C on air with a 133cfm Ultra Kaze. He has a C0 and we needed 1.45volts to "barely" get stability @ 4.1.

4.2 boots into windows, didn't last more than 30 seconds of y-cruncher - and we didn't dare try prime95.
4.3 posts, no boot.
4.4 no post.

turbo disabled in all cases.


My new toy is a D0, and it's rock stable at 3.2 @ stock voltages. I haven't pushed it to see how far it'll go @ stock voltage. Right now, I'm running it at 3.34 GHz or (3.5 = 167 x 21 with turbo) with a slight vcore bump just to be safe. Under prime95 it peaks at 78C on the "stock fans" of a CM Hyper N520. I plan on getting some better fans - something I can actually hear...

My workstation went down... I'm guessing the MOSFET(s) for the top socket couldn't take anymore 24/7/52 150-watt power draw. I'll be RMAing it in a couple weeks, but in the meantime I needed a decent computer so I had to build this little comp. (by little, I mean physically small, not "crappy")
It's small enough to hand carry on a plane so I'll be taking it back to college.
It will also take over as my gaming, and programming rig since it games better and compiles faster. My workstation will stay at home (remote-controlled from school) and be dedicated to "extreme code-testing".

Benchmarks coming soon... Hopefully I can kill enough services to do a 2.5b run - which might be the first 2.5b done on a single-socket machine on record.

v0.4.1 should be out next week. Testing is just about done and now we're working on a new layout for the site. Once that's done too, I can release v0.4.1.

Sorry to hear about your board, that is unfortunate and does happen occasionally. I've seen a friends server go up in smoke because of the mosfets, one of them basically turned to ash. I'm very interested to try the latest version, speed increases are always welcomed.

Have fun with the D0, some of them are pretty good. I'm disappointed I can't get 4.4GHz stable, that chip is a picky one though. It only likes 1.21 or 1.34v VTT, anything else and it'll eventually blue screen. Also it is picky with the Vcore. The other D0 and my old C0 weren't picky, just throw volts at them and up they go. I'll see about getting another D0, hopefully a 3845B027.

[poke349]

I believe this is due to you having the older quad core. It does not have an integrated memory controller like the i7, yours is on the chipset rather than the processor. Also the i7 is triple channel DDR3, I've seen memory reads in excess of 29gb/s. I think the main problem with the Core 2 Quad and the Xeons based off the Core 2 Quad is the older bus. Even a single Core 2 Quad has two dual cores under the heat spreader. For example, Core 0 and 1 is a die, Core 2 and 3 is another. If core 0 and 1 do the work it is quick, if core 0 and 3 do the work the speed is hindered quite a bit due to the traffic still needing to go to the bus. Also since you have two quad cores the bus gets saturated with that kind of traffic, it isn't quite as efficient as the AMD and i7 bus.

Also what helps is the "little" i7 was overclocked from 2.66GHz to 4.4GHz and has the benefit of Hyperthreading. The L3 cache was also overclocked from 2.4GHz to 3.6GHz. My i7 was $199, not too bad of a price/performance ratio I think.

Yes, 4.4 GHz is almost double the frequency of those 2.5 GHz Harpertowns.
The main thing bottlenecking the Harpertowns (as well as all the Core 2-based quad-cores) is the memory bandwidth.

Core i7s actually have more bandwidth than the program even needs... Dual-channel vs. Tri-channel only made a 3% difference on a 100m test.

Sorry to hear about your board, that is unfortunate and does happen occasionally. I've seen a friends server go up in smoke because of the mosfets, one of them basically turned to ash. I'm very interested to try the latest version, speed increases are always welcomed.

Have fun with the D0, some of them are pretty good. I'm disappointed I can't get 4.4GHz stable, that chip is a picky one though. It only likes 1.21 or 1.34v VTT, anything else and it'll eventually blue screen. Also it is picky with the Vcore. The other D0 and my old C0 weren't picky, just throw volts at them and up they go. I'll see about getting another D0, hopefully a 3845B027.

It was never overclocked and it's definitely been less than it's 2 year warranty, so it's still covered. But I have to wait until my parents come visit in about 2 weeks to bring the box for the mobo. And then, there's another 2 week turn-around time for the RMA...

The mobo still works if I only fill one socket (the working one) - which basically makes it a non-OCable QX9770 system with a stupid amount of ram...
But I'm not gonna keep on mounting and remounting everything just to use it for 2 weeks... So it's offline for now.
I'm not in dire need of 64GB of ram at the moment and all the HDs in it can be wired directly into my mini-i7 rig if I keep the two cases close enough together.


As for v0.4.1, there are two tests that I've been running at home on my mom's computer for about 10 days now, and it doesn't look like they'll be done for another week.
I'm 95% sure they'll finish without any problems and they aren't that important. There also hasn't been any major fixes since June, so I'm pretty sure that v0.4.1 is ready.
The only thing holding us back right now is the website. My roommate from last year and I have been working on it for about a week now. The only thing that isn't done yet is the homepage - which he says he can knock up in 20 minutes tonight (and he's two timezones ahead of me too). So I'll take that as a yes that it'll be by tonight.


Hopefully, my D0 will be a decent one. But I don't plan on pushing it past 4 GHz anytime soon. Just in case I fry it, I don't have another computer to use...


Just as a teaser, here's some screenies: (click on them to enlarge)

32M pi with tons of background programs... (same as SuperPi 32M)



One of the final tests for e...

Even though I'm not a serious overclocker, all those tweaking tools are perfect for letting me manually throttle the speed without needing a reboot.
When I'm testing code that will sustain 100% cpu for more than a few hours, it'll get (and stay) above 75C and that's a little too high for my confort... I typically set it down to 3GHz or even back down to stock depending on what I'm testing and how intensive it is.

[Serotoninn]

Version 0.4.1.awesomeness.win is out!!!

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

goodies include OP getting a sub 10 minute 1billion run on his new little mini-itx core i7 rig at 4ghz
also he ran the first 2.5b single socket benchmarks which he'll probably post soon!

[StevensDE]

Version 0.4.1.awesomeness.win is out!!!

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

Thanks for this update.

Here the same benchmark as yesterday with Version 0.4.1














And even better: it´s still a little bit faster than 0.3.2 :-)

@poke349

a small question: Did you compile it "only" with Visual Studio? If you just use the intel compiler and build an intel optimized version it could be still faster on an intel platform ?

[poke349]

Yes, v0.4.1 is out!

Here's a couple of large benchmarks on my LanBox.
And yes:
Sub-10 minutes for 1b on a single-socket.
A 2.5b benchmark on a single-socket.

Both at a mere modest overclock...

@poke349

a small question: Did you compile it "only" with Visual Studio? If you just use the intel compiler and build an intel optimized version it could be still faster on an intel platform ?

Yes, it was complied only with Visual Studio. I haven't tried the Intel compiler yet for a couple of reasons...

• It has a tendency to use instructions that are "too new". Which defeats the whole purpose of the x86 binary.
• It would skew benchmarks in favor of Intel processors.
• I have no freaking clue where to get the intel compiler without having to pay with my entire internship money...

And I sense some massive ownage coming... *hint**hint*

[Movieman]

a few to add.. Some slightly faster cpu's..
1m;

32m;

25m;

50m;

100m;

250m;

[poke349]

Yes, the sanity check errors for the really small SuperPi sized benchmarks is a known problem.
I'm gonna try to fix it sometime this week.

This has something to do with the timers being too imprecise for such fast timings...