DFI 875P-T @ 6Ghz

[IntelInside]

The "patch" is a "optimaization" much like what we have in 3dmark, alas many consider it a cheat. It's for prescott only AFAIK and a very bad thing for the benching community.

patch is not "optimization" like in 3DMark because it does not affect the computation result. PI is still calculated with the same precision. You are welcome to compare the results of unpatched and patched version. You will see that every digit is the same. Hopefully you will do binary compare of two output files instead of checking by hand.

I have made that patch and posted it first on www.intelfanboy.com using the same nick as here -- you can look it up there if you do not believe.

there are two versions of the patch. One that uses SSE3 FISTTP instruction and another one that is using SSE2 code so it can also be run on Northwood and Athlon 64. It can give up to 12% speedup.

I am soo sick of so called benching community "ethics". testing the latest CPUs with dated, unoptimized FPU code fair, when in reality no one will use such a crappy software. The software that needs optimization the most is already optimized so why not use it? Why SuperPI? Why not something that uses the full potential of a CPU and not just one legacy unit? If my CPU has a feature called SSE2 or SSE3 I want to be able to use it because I paid for it.

[computerpro3]

alright, lets compare a linux run to your patched one. The linux will have you beat by 10 seconds......

fair?

[IntelInside]

Cos it makes it impossible to compare results w/ the original SuperPi?

Who gives a ... (insert your favorite curse word here)?!?

Cos the patched version only works on Prescott?

There are two versions as I said above. Look them up.

Cos it makes Prescott look better than it actually is?

No, it does not. It just enables me to use what I paid for. For me that result is legit as much as any other. Why? Because:

- method of calculation hasn't changed -- algorithm is the same
- accuracy is the same
- PI result is exactly the same

You should have clearly renamed the software to Prescott-SuperPi (or SuperPi 1.1 or sumthing). That way it would have been easier to see which version of SuperPi was used and it would be easier to compare results.

I am sorry but I didn't have the source code and honestly I was lazy to patch the GUI so it says "SSE2" or "SSE3" in window title. Disassembling and patching the right instruction(s) is hard enough when you are doing it on someone else's code let alone the code is more than 10 years old and poorly optimized (if at all). Also, I have released the patches against the original SuperPI, not the patched SuperPI executable itself.

Let me remind you that I believe that everyone should use maximum power they can squeeze from their CPU. If you have SSE2 and/or SSE3 than that is the result you should publish. If you don't -- well, you are out of luck because you are going to be slower than those who do.

Tell me honestly, if you have Intel CPU do you disable SSE2 when you use VirtualDub, encode divx or mp3, work in Photoshop, etc? No? Do you disable Hyper-Threading? No? Well, why should you disable any of it in SuperPI or any other benchmark for chrissakes? Just a rhetorical question, don't have to answer it.

[IntelInside]

alright, lets compare a linux run to your patched one. The linux will have you beat by 10 seconds......

fair?

Fair enough for me. That doesn't prove that Linux is better OS, just that the Linux version of SuperPI (if that is what you were talking about) is optimized better. All that matters to me is that the program used is still SuperPI and the calculation result is exactly the same.

[STEvil]

Who gives a ... (insert your favorite curse word here)?!?

Those who use it, apparently.

[macci]

There are two versions as I said above. Look them up.

Link to the SSE2 version? Thx

[IntelInside]

Link to the SSE2 version? Thx

Here you go:

Prescott PI
Northwood PI
Technical details of the patch

Everyone, sorry I got a bit pissed off. Fugger, great job done with O/C. Keep it up.

[DudeMiester]

So what about a version that uses 3DNow and AMD64?

[IntelInside]

So what about a version that uses 3DNow and AMD64?

AMD64 can use SSE2 code and sorry but I don't have the SuperPI source code to recompile for AMD64 in 64-bit mode and I don't have AMD64 CPU to develop and test it on. Note that recompiling would also bring advantage to Intel CPUs because the whole application would get optimized. 3DNow would hardly give any advantage. You are welcome to try and make 3DNow and AMD64 patch though. I did all I could (without source code) to utilize full power of my Pentium 4 520.

[ablaze]

FUGGER, insane overclocks you got going!

got a question. I've got an 875P-T, but I've got throttling issues at higher cpu clocks. There's no option to disable thermal throttling in the bios. Did you encounter throttling issues or found a way round them?

[Webzta]

FUGGER, insane overclocks you got going!

got a question. I've got an 875P-T, but I've got throttling issues at higher cpu clocks. There's no option to disable thermal throttling in the bios. Did you encounter throttling issues or found a way round them?

he's running at -119c I'm guessing He hasn't had any thermal issues mate
as throttling wouldn't come into play till you were nearing the instability temps.. 60c +

[ablaze]

I think you're right about that. haha
I was initially under the impression that throttling came into play at the edges of system stability cos though load temps shot up to 66°C for me, even when running benches at idle temps of 40+ the cpu still throttled.
but now, I just re-tweaked my setup and load temps are at 55, and throttling's gone!

anyway, just for confirmation, FUGGER, did you encounter any throttling on your setup?

[chilly1]

Cos it makes it impossible to compare results w/ the original SuperPi? Cos the patched version only works on Prescott? Cos it makes Prescott look better than it actually is?
You should have clearly renamed the software to Prescott-SuperPi (or SuperPi 1.1 or sumthing). That way it would have been easier to see which version of SuperPi was used and it would be easier to compare results.
Just personal thoughts nothing more.
Nice to see what the prescott can do once the software is optimized for it thou.

What happens when the next generation of AMD processors have a new set of instructions that make it work better should we perclude their use? And in teh original superPI is tehre any other code that is processor unique?

[OPPAINTER]

What happens when the next generation of AMD processors have a new set of instructions that make it work better should we perclude their use? And in teh original superPI is tehre any other code that is processor unique?

AMD rocks :banana4:

OPP

[Kane]

IntelInside has a point. If we can't use his SuperPI, because it uses SSE3, why should people be allowed to use HT in benchmarks? What's legal, and what's not? Personally I don't se it as a cheat. Don't know about you guys, but would be fun to hear your opinions. Oh, by the way, I own a AMD, so I'm not a FanBoy

[Charles Wirth]

Ablaze, no throttling. You are thinking of Nvidia based video cards.

Pi is a touchy issue for most when it comes to version and optimizations. we had a version we trusted the result but had very poor ability show exact results. meaning that it did not break down to tenths or hundredth of a second.

We dont really know how it was coded other than it was coded poorly in the first place. We see other operating systems scoring much better at the exact same calculation.

My G5's can pull 11 second pi and I am sure the new 2.5Ghz duals can pull under 10 seconds.

Now with all the fuss over verison and the inability to distinguish verions I may hold off on Pi untill we can to a happy medium where results are accepted without question on the Pentium platform w/Prescott.

[Xassius]

IntelInside has a point. If we can't use his SuperPI, because it uses SSE3, why should people be allowed to use HT in benchmarks? What's legal, and what's not? Personally I don't se it as a cheat. Don't know about you guys, but would be fun to hear your opinions. Oh, by the way, I own a AMD, so I'm not a FanBoy

as i have mentioned before...do we need DOOM 3 Prescott Version and DOOM3 Winchester Version?

[STEvil]

I have no problem with people running it, just as long as it can be distinguished from "normal" super-pi.

that said, is 40 seconds @ 2211mhz on A64 2800+ clawhammer with 512k L2 good?

EDIt

maybe its time for a new version of super-pi? one that you can choose between raw calculation ("divide this into this until you get to this many decimal places, period"), SSE2, SSE3, and maybe someone with enough motivation could throw in 3DNow for fun.

Myself I want to see a video card calculate pi to 1m places.. im sure someone from www.gpgpu.org might think it an interesting thing to try?

[OPPAINTER]

Super PI sucks anyway

OPP

[STEvil]

lol

[OPPAINTER]

lol

I suck at that benchmark, 24s

OPP

[CodeRed]

Here you go:

Prescott PI
Northwood PI
Technical details of the patch

Everyone, sorry I got a bit pissed off. Fugger, great job done with O/C. Keep it up.

hmm, this patch feels unnatural

I struggled to hit 25 sec on win2k@3232MHz (FX53). The NW (SSE2) patch took off 0.72 sec, so 24 sec now looks a possibility with win2003 + patch.

This is going to make it very hard if not impossible to compare superpi scores, thus invalidating it as a benchmark.


benchmark(ing):
n., A standard by which something can be measured or judged
tr.v., To measure (a rival's product) according to specified standards in order to compare it with and improve one's own product.

[Xassius]

Excuse me, but why are you calling my SuperPI patch crap?
Just curious...

excuse me...but why have you offically wrecked a benchmark many of us use?

by no means am I saying that enabling optimizations in a program is bad...it's just that enabling them in a Benchmark for COMPARISON purposes and NOT BEING ABLE TO DIFFER BETWEEN PATCHED AND UNPATCHED is really wrecking and destroying the purpose of superpi...

[blinky]

we had a version we trusted the result but had very poor ability show exact results. meaning that it did not break down to tenths or hundredth of a second.

just use codered's processtimer, it tells u ur exact time to the hundredths

[cpulloverclock]

just use codered's processtimer, it tells u ur exact time to the hundredths

Where I can dl it?

hum...

u can see your time on superpi down to the hundreth using codeRed's processtimer.exe PM him for info where to get it