New 800 MHz FSB - Page 2

guido72 · 2003-05-29, 13:47

Quote:

Originally Posted by MAtthes

Hey Gudio,

thanx for the warm welcome.

I'm using an SLK900U from Thermalright (one monster of a copper heatsink) with a Pabst 2000RPM 92mm fan. It is kinda running on the hotside with 61C. I am planning on upgrading the fan and getting a better thermal grease so I can keep it at 55C I hope. Voltage is 1.55 for Vcore (although my powersupply is kinda crappy, so it is actually somewhere around 1.52). Vdimm is 2.8 with PC3500 Geil DDR @ 2-3-3-6 running at DDR416.

EDIT: Oh yeah that is 512MB of RAM (2 Sticks of 256MB so the dualchannel stuff works)

I might give your HT tests a try in a while, I'll keep you posted.

Regards and my congrats to Italy for winning the Champions League (although that was clear before they played :-))

Matthes

Hi once again!
I know your heatsink: really a monster!
Well, waiting for your benchies and Thank you for your congratulations for our Soccer Clubs. Never been in Dortmund, but some time ago, I saw F.C. Bayern Vs. Borussia at Olimpia Stadion, in Munich (I think Bayern won 3-1, two goals by Olaf Thon, but I'm not sure). Maybe it was the Stadium (wonderful!) but I liked that match as much as Juventus-Milan...
Regards

Matthes · 2003-05-29, 16:56

Hi again,

this post might get a little long so bear with me ... ;)

First of all I have rebenched everything with as little stuff running in the background as possible (sometimes it is hard to determine what task may survive and which won't). I have also run the test (except factoring) multible times so any quirks might be evened out. I chose to post the one which I thought was quickest (all in all there was very little variance in the timings).

First benchies are with HT on and just one LL running:

Intel(R) Pentium(R) 4 CPU 2.40GHz
CPU speed: 3121.34 MHz
CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2
L1 cache size: 8 KB
L2 cache size: 512 KB
L1 cache line size: 64 bytes
L2 cache line size: 128 bytes
TLBS: 64
Prime95 version 23.4, RdtscTiming=1
Best time for 384K FFT length: 11.418 ms.
Best time for 448K FFT length: 13.557 ms.
Best time for 512K FFT length: 15.431 ms.
Best time for 640K FFT length: 18.521 ms.
Best time for 768K FFT length: 22.490 ms.
Best time for 896K FFT length: 26.752 ms.
Best time for 1024K FFT length: 29.992 ms.
Best time for 1280K FFT length: 39.231 ms.
Best time for 1536K FFT length: 48.456 ms.
Best time for 1792K FFT length: 57.495 ms.
Best time for 2048K FFT length: 64.930 ms.
Compare your results to other computers at http://www.mersenne.org/bench.htm
That web page also contains instructions on how your results can be included.

Next is HT off and one LL:

Intel(R) Pentium(R) 4 CPU 2.40GHz
CPU speed: 3121.35 MHz
CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2
L1 cache size: 8 KB
L2 cache size: 512 KB
L1 cache line size: 64 bytes
L2 cache line size: 128 bytes
TLBS: 64
Prime95 version 23.4, RdtscTiming=1
Best time for 384K FFT length: 11.474 ms.
Best time for 448K FFT length: 13.624 ms.
Best time for 512K FFT length: 15.491 ms.
Best time for 640K FFT length: 18.633 ms.
Best time for 768K FFT length: 22.598 ms.
Best time for 896K FFT length: 26.837 ms.
Best time for 1024K FFT length: 30.208 ms.
Best time for 1280K FFT length: 39.447 ms.
Best time for 1536K FFT length: 48.734 ms.
Best time for 1792K FFT length: 57.830 ms.
Best time for 2048K FFT length: 65.231 ms.
Compare your results to other computers at http://www.mersenne.org/bench.htm
That web page also contains instructions on how your results can be included.

Some conclusions on my part (or call them wild guesses). HT does very little if you decide to run just one LL, if you are picky you might even say it helps a little, this could be due to some task stealing some cycles from Prime95, which with HT on Prime95 would get some part of. But to be quite honest, I would have to say that it lies in the margin of error.

Now HT on and two LLs running.

The benchmarking as someone has pointed out in another thread are not accurate, probably because there is a pause between the tests for different FFTs.

Anyhow running two LLs in normal mode i get timings of around .119 secs on both of the tests. Running one instance gives timings of about .058 secs. This seems to indicate, that either HT has some overhead or both LLs are fighting over memorybandwith or a little of both. I'm still impressed with it, and would have suspected worse timings with two LLs running.

On to HT on and one LL + one factoring.

Unfortunately I could not copy and paste the timings for the factoring out of the Prime95 window, so I took screenshots (about 150K each) which can be found here for factoring + LL http://fb.design.fh-dortmund.de/matt...ctoringwll.jpg and here for just factoring http://fb.design.fh-dortmund.de/matt...toringwoll.jpg . It might show different exponents, this is because I forgot to do just the factoring the first time around. Timings for the LL instance were around .099 secs. Notice how this is faster then doing two LLs at the same time. I dont have any thoughts on wether it makes much sense to do both at the same time, but since LL does make more sense being able to find positive answers and not just negatives as factoring, I will personally stick to just doing one LL at a time. I did unregister the exponents that were send to me for factoring.

Well if anyone is still reading this, I'm done now :D

If there is anything else, someone wants me to try just drop a post in this thread, I'll be watching.

Regards, Matthes

2003-05-29, 16:56	#13
Matthes May 2003 2³·3 Posts	Hi again, this post might get a little long so bear with me ... ;) First of all I have rebenched everything with as little stuff running in the background as possible (sometimes it is hard to determine what task may survive and which won't). I have also run the test (except factoring) multible times so any quirks might be evened out. I chose to post the one which I thought was quickest (all in all there was very little variance in the timings). First benchies are with HT on and just one LL running: Intel(R) Pentium(R) 4 CPU 2.40GHz CPU speed: 3121.34 MHz CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2 L1 cache size: 8 KB L2 cache size: 512 KB L1 cache line size: 64 bytes L2 cache line size: 128 bytes TLBS: 64 Prime95 version 23.4, RdtscTiming=1 Best time for 384K FFT length: 11.418 ms. Best time for 448K FFT length: 13.557 ms. Best time for 512K FFT length: 15.431 ms. Best time for 640K FFT length: 18.521 ms. Best time for 768K FFT length: 22.490 ms. Best time for 896K FFT length: 26.752 ms. Best time for 1024K FFT length: 29.992 ms. Best time for 1280K FFT length: 39.231 ms. Best time for 1536K FFT length: 48.456 ms. Best time for 1792K FFT length: 57.495 ms. Best time for 2048K FFT length: 64.930 ms. Compare your results to other computers at http://www.mersenne.org/bench.htm That web page also contains instructions on how your results can be included. Next is HT off and one LL: Intel(R) Pentium(R) 4 CPU 2.40GHz CPU speed: 3121.35 MHz CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE, SSE2 L1 cache size: 8 KB L2 cache size: 512 KB L1 cache line size: 64 bytes L2 cache line size: 128 bytes TLBS: 64 Prime95 version 23.4, RdtscTiming=1 Best time for 384K FFT length: 11.474 ms. Best time for 448K FFT length: 13.624 ms. Best time for 512K FFT length: 15.491 ms. Best time for 640K FFT length: 18.633 ms. Best time for 768K FFT length: 22.598 ms. Best time for 896K FFT length: 26.837 ms. Best time for 1024K FFT length: 30.208 ms. Best time for 1280K FFT length: 39.447 ms. Best time for 1536K FFT length: 48.734 ms. Best time for 1792K FFT length: 57.830 ms. Best time for 2048K FFT length: 65.231 ms. Compare your results to other computers at http://www.mersenne.org/bench.htm That web page also contains instructions on how your results can be included. Some conclusions on my part (or call them wild guesses). HT does very little if you decide to run just one LL, if you are picky you might even say it helps a little, this could be due to some task stealing some cycles from Prime95, which with HT on Prime95 would get some part of. But to be quite honest, I would have to say that it lies in the margin of error. Now HT on and two LLs running. The benchmarking as someone has pointed out in another thread are not accurate, probably because there is a pause between the tests for different FFTs. Anyhow running two LLs in normal mode i get timings of around .119 secs on both of the tests. Running one instance gives timings of about .058 secs. This seems to indicate, that either HT has some overhead or both LLs are fighting over memorybandwith or a little of both. I'm still impressed with it, and would have suspected worse timings with two LLs running. On to HT on and one LL + one factoring. Unfortunately I could not copy and paste the timings for the factoring out of the Prime95 window, so I took screenshots (about 150K each) which can be found here for factoring + LL http://fb.design.fh-dortmund.de/matt...ctoringwll.jpg and here for just factoring http://fb.design.fh-dortmund.de/matt...toringwoll.jpg . It might show different exponents, this is because I forgot to do just the factoring the first time around. Timings for the LL instance were around .099 secs. Notice how this is faster then doing two LLs at the same time. I dont have any thoughts on wether it makes much sense to do both at the same time, but since LL does make more sense being able to find positive answers and not just negatives as factoring, I will personally stick to just doing one LL at a time. I did unregister the exponents that were send to me for factoring. Well if anyone is still reading this, I'm done now :D If there is anything else, someone wants me to try just drop a post in this thread, I'll be watching. Regards, Matthes