Cache, dual channel memory and Mprime performance

optim · 2004-06-23, 02:18

AMD Athlon64 in Socket754 has 1MB of L2 cache and single channel DDR memory. AMD Athlon64 in Socket939 has 512KB of L2 cache and dual channel DDR memory. Supposing the same core speed (2.2GHz in Athlon64 754-3400+/939-3500+), which one is better for Mprime?

How much the amount of cache and the presence of a dual channel memory affect the performance of Mprime?

If possible give me speeds for 1792K FFT benchmark, telling me what CPU you use along with its memory, frequency and L2 cache.

For example one of my boxen is an AthlonXP2600@2103MHz, 512K cache, 1GB DDR333@FSB169MHz which attains a 157ms time in the 1792 FFT benchmark with Mprime23.5 and Linux 2.6.5.

I want to upgrade to a faster processor. I suppose I will not see much difference with an Athlon3200@2200MHz and DDR400@FSB200MHz. So I am thinking about an Athlon 64bit...

Another question: How well Mprime23.5 runs in 64bit Linux?

error404 · 2004-06-23, 13:05

Optim

The following is the benchmark test for my AMD64-3400+ and my older
3000+. Note that the output for Windows is more verbose than the Linux
output.

mprime is a 32 bit binary and will only run under 32 bit Linux/UNIX, it will
will not run in 64 bit mode.

Tom

/******************************************************/

Slackware-9.1-CURRENT
Gigabyte GA-K8VT800M motherboard (AMD64-3400+)
K8-2200, 64K L1, 1M L2, FSB 800MHz, 1024M PC3200, Bus 200MHz

Your timings will be written to the results.txt file.
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.

Timing 73 iterations at 384K FFT length. Best time: 21.511 ms.
Timing 62 iterations at 448K FFT length. Best time: 25.634 ms.
Timing 54 iterations at 512K FFT length. Best time: 29.052 ms.
Timing 43 iterations at 640K FFT length. Best time: 36.449 ms.
Timing 36 iterations at 768K FFT length. Best time: 44.719 ms.
Timing 31 iterations at 896K FFT length. Best time: 53.794 ms.
Timing 27 iterations at 1024K FFT length. Best time: 60.608 ms.
Timing 21 iterations at 1280K FFT length. Best time: 80.654 ms.
Timing 18 iterations at 1536K FFT length. Best time: 98.725 ms.
Timing 15 iterations at 1792K FFT length. Best time: 118.297 ms.
Timing 13 iterations at 2048K FFT length. Best time: 133.803 ms.
Benchmark complete.

/******************************************************/

Windows-2000
Abit NF7-S motherboard (AthlonXP 3000+)
AMD-2167, 512K Cache, FSB 166 MHz, 768M PC2100, Bus 333 MHz

[Wed Jun 23 07:44:42 2004]
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.

AMD Athlon(tm) XP
CPU speed: 2171.67 MHz
CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE
L1 cache size: 64 KB
L2 cache size: 512 KB
L1 cache line size: 64 bytes
L2 cache line size: 64 bytes
L1 TLBS: 32
L2 TLBS: 256
Prime95 version 23.8, RdtscTiming=1
Best time for 384K FFT length: 33.347 ms.
Best time for 448K FFT length: 38.966 ms.
Best time for 512K FFT length: 41.337 ms.
Best time for 640K FFT length: 53.218 ms.
Best time for 768K FFT length: 62.968 ms.
Best time for 896K FFT length: 73.711 ms.
Best time for 1024K FFT length: 79.437 ms.
Best time for 1280K FFT length: 104.764 ms.
Best time for 1536K FFT length: 122.211 ms.
Best time for 1792K FFT length: 151.922 ms.
Best time for 2048K FFT length: 164.430 ms.

/******************************************************/
/******************************************************/
/******************************************************/

optim · 2004-06-23, 13:57

wow, thanks! :-)

delta_t · 2004-06-23, 15:22

This is somewhat on topic of your original question, but I found that dual channel memory (DDR333) in my dual processor Xenon box allows me to run two LL-instance of Prime95 at full speed. Others have reported slow downs with two LL-instances with dual Xenons but I believe they had single channel memory. Apparently the dual channel (in these cases at least) are better.

I'm not too sure of the Opterons though. They are better at memory intensive apps due to their integrated memory controller and better bus. Perhaps dual vs. single with mprime wouldn't make such a big difference?

optim · 2004-06-25, 03:20

what Xeon systemboard are you using?

2004-06-23, 02:18	#1
optim Nov 2003 European Union 68₁₆ Posts	Cache, dual channel memory and Mprime performance AMD Athlon64 in Socket754 has 1MB of L2 cache and single channel DDR memory. AMD Athlon64 in Socket939 has 512KB of L2 cache and dual channel DDR memory. Supposing the same core speed (2.2GHz in Athlon64 754-3400+/939-3500+), which one is better for Mprime? How much the amount of cache and the presence of a dual channel memory affect the performance of Mprime? If possible give me speeds for 1792K FFT benchmark, telling me what CPU you use along with its memory, frequency and L2 cache. For example one of my boxen is an AthlonXP2600@2103MHz, 512K cache, 1GB DDR333@FSB169MHz which attains a 157ms time in the 1792 FFT benchmark with Mprime23.5 and Linux 2.6.5. I want to upgrade to a faster processor. I suppose I will not see much difference with an Athlon3200@2200MHz and DDR400@FSB200MHz. So I am thinking about an Athlon 64bit... Another question: How well Mprime23.5 runs in 64bit Linux?

2004-06-23, 13:05	#2
error404 Sep 2003 43 Posts	amd64-3400+ Optim The following is the benchmark test for my AMD64-3400+ and my older 3000+. Note that the output for Windows is more verbose than the Linux output. mprime is a 32 bit binary and will only run under 32 bit Linux/UNIX, it will will not run in 64 bit mode. Tom /****************************************************/ Slackware-9.1-CURRENT Gigabyte GA-K8VT800M motherboard (AMD64-3400+) K8-2200, 64K L1, 1M L2, FSB 800MHz, 1024M PC3200, Bus 200MHz Your timings will be written to the results.txt file. 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. Timing 73 iterations at 384K FFT length. Best time: 21.511 ms. Timing 62 iterations at 448K FFT length. Best time: 25.634 ms. Timing 54 iterations at 512K FFT length. Best time: 29.052 ms. Timing 43 iterations at 640K FFT length. Best time: 36.449 ms. Timing 36 iterations at 768K FFT length. Best time: 44.719 ms. Timing 31 iterations at 896K FFT length. Best time: 53.794 ms. Timing 27 iterations at 1024K FFT length. Best time: 60.608 ms. Timing 21 iterations at 1280K FFT length. Best time: 80.654 ms. Timing 18 iterations at 1536K FFT length. Best time: 98.725 ms. Timing 15 iterations at 1792K FFT length. Best time: 118.297 ms. Timing 13 iterations at 2048K FFT length. Best time: 133.803 ms. Benchmark complete. /**************************************************/ Windows-2000 Abit NF7-S motherboard (AthlonXP 3000+) AMD-2167, 512K Cache, FSB 166 MHz, 768M PC2100, Bus 333 MHz [Wed Jun 23 07:44:42 2004] 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. AMD Athlon(tm) XP CPU speed: 2171.67 MHz CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE L1 cache size: 64 KB L2 cache size: 512 KB L1 cache line size: 64 bytes L2 cache line size: 64 bytes L1 TLBS: 32 L2 TLBS: 256 Prime95 version 23.8, RdtscTiming=1 Best time for 384K FFT length: 33.347 ms. Best time for 448K FFT length: 38.966 ms. Best time for 512K FFT length: 41.337 ms. Best time for 640K FFT length: 53.218 ms. Best time for 768K FFT length: 62.968 ms. Best time for 896K FFT length: 73.711 ms. Best time for 1024K FFT length: 79.437 ms. Best time for 1280K FFT length: 104.764 ms. Best time for 1536K FFT length: 122.211 ms. Best time for 1792K FFT length: 151.922 ms. Best time for 2048K FFT length: 164.430 ms. /**************************************************/ /**************************************************/ /****************************************************/

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Quad Channel DDR4 vs Dual Channel DDR4	ATH	Hardware	17	2015-10-27 21:06
Single vs Dual channel memory	TObject	Hardware	5	2014-12-24 05:58
Importance of dual channel memory for dual core processors	patrik	Hardware	3	2007-01-07 09:26
Opteron 175, Asus A8V-Deluxe, OCZ dual channel pc4000	optyguy	Hardware	3	2006-01-21 08:06
MPrime: K7 has 64KB of L1 cache... but I thought it had 128K!	optim	Hardware	2	2004-07-10 19:59

2004-06-23, 13:57	#3
optim Nov 2003 European Union 2³×13 Posts	wow, thanks! :-)

2004-06-23, 15:22	#4
delta_t Nov 2002 Anchorage, AK 3×7×17 Posts	This is somewhat on topic of your original question, but I found that dual channel memory (DDR333) in my dual processor Xenon box allows me to run two LL-instance of Prime95 at full speed. Others have reported slow downs with two LL-instances with dual Xenons but I believe they had single channel memory. Apparently the dual channel (in these cases at least) are better. I'm not too sure of the Opterons though. They are better at memory intensive apps due to their integrated memory controller and better bus. Perhaps dual vs. single with mprime wouldn't make such a big difference?

2004-06-25, 03:20	#5
optim Nov 2003 European Union 2³·13 Posts	what Xeon systemboard are you using?