Is this a 23.4 bug????

garo · 2003-06-13, 20:33

Why am I getting such terrible times with a P4? Nothing else is running on the machine - it's a linux machine running mprime. The CPU utilization by mprime is 99% and here is the benchmark result:

Intel(R) Pentium(R) 4 CPU 2.53GHz
CPU speed: 2518.42 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: 36.180 ms.
Best time for 448K FFT length: 82.717 ms.
Best time for 512K FFT length: 98.710 ms.
Best time for 640K FFT length: 125.363 ms.
Best time for 768K FFT length: 156.425 ms.
Best time for 896K FFT length: 188.587 ms.
Best time for 1024K FFT length: 215.378 ms.
Best time for 1280K FFT length: 295.141 ms.
Best time for 1536K FFT length: 362.408 ms.
Best time for 1792K FFT length: 452.708 ms.
Best time for 2048K FFT length: 518.228 ms.

This is just terrible and I don't know why!!

Ha! I just ran 22.12 and this is what I get:

Intel(R) Pentium(R) 4 CPU 2.53GHz
CPU speed: 2520.08 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: 64 bytes
TLBS: 64
Prime95 version 22.12, RdtscTiming=1
Best time for 256K FFT length: 9.605 ms.
Best time for 320K FFT length: 12.701 ms.
Best time for 384K FFT length: 15.452 ms.
Best time for 448K FFT length: 18.584 ms.
Best time for 512K FFT length: 20.755 ms.
Best time for 640K FFT length: 26.945 ms.
Best time for 768K FFT length: 33.202 ms.
Best time for 896K FFT length: 40.465 ms.
Best time for 1024K FFT length: 43.791 ms.
Best time for 1280K FFT length: 62.086 ms.
Best time for 1536K FFT length: 75.915 ms.
Best time for 1792K FFT length: 92.438 ms.

garo · 2003-06-13, 20:37

Ok!
Even 23.3 seems to be working fine.
Intel(R) Pentium(R) 4 CPU 2.53GHz
CPU speed: 2518.52 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 23.3, RdtscTiming=1
Best time for 384K FFT length: 14.862 ms.
Best time for 448K FFT length: 17.550 ms.
Best time for 512K FFT length: 19.975 ms.
Best time for 640K FFT length: 25.726 ms.
Best time for 768K FFT length: 30.548 ms.
Best time for 896K FFT length: 37.083 ms.
Best time for 1024K FFT length: 41.160 ms.
Best time for 1280K FFT length: 55.680 ms.
Best time for 1536K FFT length: 67.167 ms.
Best time for 1792K FFT length: 81.180 ms.
Best time for 2048K FFT length: 90.121 ms.

BTW, with 23.4 I checked the logfile as well and mprime is actually running slow. Not just the benchmark

garo · 2003-06-13, 21:58

Okay. This is getting curioser and curioser. I find that no matter what version I run, after about 10 minutes the throughput slows to about an eighth of what it should be. I redirect the mprime output to a log file and I found that the first 10000 iterations give the good per-iteration time and then it slows down dramatically. Instead of an expected time of 17ms I get anything between 100 to 130ms.

I am now running a SETI unit on the machine to see if the problem is the machine or the machine/Prime95 combo.

Someone, please help!!!

:( :?

ewmayer · 2003-06-13, 22:39

I noticed when I first switched to 23.4 on several p4s at work, right after restarting using the new version I got a dramatic improvement in the per-iteration time (typically 15-20%), but after a few minutes things slowed down somewhat, but still gave a 5-10% speedup. This suggests to me that perhaps 23.4 is stressing the CPU sufficiently that heat-induced throttle-down is negating some of the cache usage improvements. Nowhere near the kind of slowdown you're seeing though. Try getting some extra airflow around your CPU and see if that helps.

garo · 2003-06-13, 23:20

That is absolutely right!! i just checked and found that there is a fan failure because of which the CPU is throttling down a huge amount. Now it makes complete sense. Thanks a lot!!

2003-06-13, 20:33	#1
garo Aug 2002 Termonfeckin, IE 2²·691 Posts	Is this a 23.4 bug???? Why am I getting such terrible times with a P4? Nothing else is running on the machine - it's a linux machine running mprime. The CPU utilization by mprime is 99% and here is the benchmark result: Intel(R) Pentium(R) 4 CPU 2.53GHz CPU speed: 2518.42 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: 36.180 ms. Best time for 448K FFT length: 82.717 ms. Best time for 512K FFT length: 98.710 ms. Best time for 640K FFT length: 125.363 ms. Best time for 768K FFT length: 156.425 ms. Best time for 896K FFT length: 188.587 ms. Best time for 1024K FFT length: 215.378 ms. Best time for 1280K FFT length: 295.141 ms. Best time for 1536K FFT length: 362.408 ms. Best time for 1792K FFT length: 452.708 ms. Best time for 2048K FFT length: 518.228 ms. This is just terrible and I don't know why!! Ha! I just ran 22.12 and this is what I get: Intel(R) Pentium(R) 4 CPU 2.53GHz CPU speed: 2520.08 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: 64 bytes TLBS: 64 Prime95 version 22.12, RdtscTiming=1 Best time for 256K FFT length: 9.605 ms. Best time for 320K FFT length: 12.701 ms. Best time for 384K FFT length: 15.452 ms. Best time for 448K FFT length: 18.584 ms. Best time for 512K FFT length: 20.755 ms. Best time for 640K FFT length: 26.945 ms. Best time for 768K FFT length: 33.202 ms. Best time for 896K FFT length: 40.465 ms. Best time for 1024K FFT length: 43.791 ms. Best time for 1280K FFT length: 62.086 ms. Best time for 1536K FFT length: 75.915 ms. Best time for 1792K FFT length: 92.438 ms.

2003-06-13, 20:37	#2
garo Aug 2002 Termonfeckin, IE 101011001100₂ Posts	Ok! Even 23.3 seems to be working fine. Intel(R) Pentium(R) 4 CPU 2.53GHz CPU speed: 2518.52 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 23.3, RdtscTiming=1 Best time for 384K FFT length: 14.862 ms. Best time for 448K FFT length: 17.550 ms. Best time for 512K FFT length: 19.975 ms. Best time for 640K FFT length: 25.726 ms. Best time for 768K FFT length: 30.548 ms. Best time for 896K FFT length: 37.083 ms. Best time for 1024K FFT length: 41.160 ms. Best time for 1280K FFT length: 55.680 ms. Best time for 1536K FFT length: 67.167 ms. Best time for 1792K FFT length: 81.180 ms. Best time for 2048K FFT length: 90.121 ms. BTW, with 23.4 I checked the logfile as well and mprime is actually running slow. Not just the benchmark

2003-06-13, 21:58	#3
garo Aug 2002 Termonfeckin, IE 2²·691 Posts	Okay. This is getting curioser and curioser. I find that no matter what version I run, after about 10 minutes the throughput slows to about an eighth of what it should be. I redirect the mprime output to a log file and I found that the first 10000 iterations give the good per-iteration time and then it slows down dramatically. Instead of an expected time of 17ms I get anything between 100 to 130ms. I am now running a SETI unit on the machine to see if the problem is the machine or the machine/Prime95 combo. Someone, please help!!! :( :?

2003-06-13, 22:39	#4
ewmayer ∂²ω=0 Sep 2002 República de California 2D7E₁₆ Posts	I noticed when I first switched to 23.4 on several p4s at work, right after restarting using the new version I got a dramatic improvement in the per-iteration time (typically 15-20%), but after a few minutes things slowed down somewhat, but still gave a 5-10% speedup. This suggests to me that perhaps 23.4 is stressing the CPU sufficiently that heat-induced throttle-down is negating some of the cache usage improvements. Nowhere near the kind of slowdown you're seeing though. Try getting some extra airflow around your CPU and see if that helps.

2003-06-13, 23:20	#5
garo Aug 2002 Termonfeckin, IE 2²×691 Posts	That is absolutely right!! i just checked and found that there is a fan failure because of which the CPU is throttling down a huge amount. Now it makes complete sense. Thanks a lot!!