On Monday 20 June 2005 18:49, George Woltman wrote: > At 02:02 AM 6/20/2005, you wrote: > >I have a question too. Regarding v24. I have a dual PIII system running LL > >tests on one CPU and factoring on the other. LL efficiency demands > > sticking to v23 but this could result in doing excess factoring according > > to the new cutoffs... advice please. > > First benchmark v24 and v23. The other improvements in v24 may have offset > the loss due to the new Athlon-friendly memory layout.
Thanks George, that hit the spot. FYI this is a dual PIII-1000 (Coppermine) system with a single memory controller, 512MB x PC133 SDRAM running at 100 MHz with timings 5/2/2, running linux 2.4.20 (a) v23.5: Intel(R) Pentium(R) III processor CPU speed: 1002.37 MHz CPU features: RDTSC, CMOV, PREFETCH, MMX, SSE L1 cache size: 16 KB L2 cache size: 256 KB L1 cache line size: 32 bytes L2 cache line size: 32 bytes TLBS: 64 Prime95 version 23.5, RdtscTiming=1 Best time for 384K FFT length: 80.433 ms. Best time for 448K FFT length: 96.319 ms. Best time for 512K FFT length: 107.134 ms. Best time for 640K FFT length: 140.853 ms. Best time for 768K FFT length: 167.865 ms. Best time for 896K FFT length: 199.521 ms. Best time for 1024K FFT length: 228.064 ms. Best time for 1280K FFT length: 304.675 ms. Best time for 1536K FFT length: 358.049 ms. Best time for 1792K FFT length: 449.207 ms. Best time for 2048K FFT length: 499.108 ms. (b) v24.12, affinity set to "any processor", other processor idle: Intel(R) Pentium(R) III processor CPU speed: 1002.46 MHz CPU features: RDTSC, CMOV, Prefetch, MMX, SSE L1 cache size: 16 KB L2 cache size: 256 KB L1 cache line size: 32 bytes L2 cache line size: 32 bytes TLBS: 64 Prime95 32-bit version 24.12, RdtscTiming=1 Best time for 512K FFT length: 92.389 ms. Best time for 640K FFT length: 124.077 ms. Best time for 768K FFT length: 150.051 ms. Best time for 896K FFT length: 184.520 ms. Best time for 1024K FFT length: 211.963 ms. Best time for 1280K FFT length: 288.633 ms. Best time for 1536K FFT length: 349.149 ms. Best time for 1792K FFT length: 417.301 ms. Best time for 2048K FFT length: 465.396 ms. Best time for 2560K FFT length: 610.460 ms. Best time for 3072K FFT length: 744.327 ms. Best time for 3584K FFT length: 911.517 ms. Best time for 4096K FFT length: 1031.383 ms. Best time for 58 bit trial factors: 14.151 ms. Best time for 59 bit trial factors: 14.399 ms. Best time for 60 bit trial factors: 14.072 ms. Best time for 61 bit trial factors: 14.530 ms. Best time for 62 bit trial factors: 25.929 ms. Best time for 63 bit trial factors: 26.373 ms. Best time for 64 bit trial factors: 59.393 ms. Best time for 65 bit trial factors: 60.170 ms. Best time for 66 bit trial factors: 61.033 ms. Best time for 67 bit trial factors: 61.375 ms. (c) v24.12, affinity set to CPU 0, other processor idle: Intel(R) Pentium(R) III processor CPU speed: 1002.19 MHz CPU features: RDTSC, CMOV, Prefetch, MMX, SSE L1 cache size: 16 KB L2 cache size: 256 KB L1 cache line size: 32 bytes L2 cache line size: 32 bytes TLBS: 64 Prime95 32-bit version 24.12, RdtscTiming=1 Best time for 512K FFT length: 92.252 ms. Best time for 640K FFT length: 122.676 ms. Best time for 768K FFT length: 150.749 ms. Best time for 896K FFT length: 181.436 ms. Best time for 1024K FFT length: 208.147 ms. Best time for 1280K FFT length: 286.569 ms. Best time for 1536K FFT length: 348.531 ms. Best time for 1792K FFT length: 423.336 ms. Best time for 2048K FFT length: 466.984 ms. Best time for 2560K FFT length: 612.132 ms. Best time for 3072K FFT length: 750.940 ms. Best time for 3584K FFT length: 912.906 ms. Best time for 4096K FFT length: 1014.040 ms. Best time for 58 bit trial factors: 14.162 ms. Best time for 59 bit trial factors: 14.090 ms. Best time for 60 bit trial factors: 14.112 ms. Best time for 61 bit trial factors: 14.060 ms. Best time for 62 bit trial factors: 25.953 ms. Best time for 63 bit trial factors: 25.959 ms. Best time for 64 bit trial factors: 59.690 ms. Best time for 65 bit trial factors: 60.151 ms. Best time for 66 bit trial factors: 61.100 ms. Best time for 67 bit trial factors: 61.406 ms. Note, affinity seems to make no difference, perhaps not surprising when running linux 2.4 family kernel. (d) v24.12, affinity set to CPU 1, other processor running LL test: Intel(R) Pentium(R) III processor CPU speed: 1002.12 MHz CPU features: RDTSC, CMOV, Prefetch, MMX, SSE L1 cache size: 16 KB L2 cache size: 256 KB L1 cache line size: 32 bytes L2 cache line size: 32 bytes TLBS: 64 Prime95 32-bit version 24.12, RdtscTiming=1 Best time for 512K FFT length: 101.971 ms. Best time for 640K FFT length: 131.915 ms. Best time for 768K FFT length: 160.634 ms. Best time for 896K FFT length: 194.316 ms. Best time for 1024K FFT length: 227.816 ms. Best time for 1280K FFT length: 331.071 ms. Best time for 1536K FFT length: 400.086 ms. Best time for 1792K FFT length: 470.086 ms. Best time for 2048K FFT length: 540.083 ms. Best time for 2560K FFT length: 691.039 ms. Best time for 3072K FFT length: 837.910 ms. Best time for 3584K FFT length: 1022.582 ms. Best time for 4096K FFT length: 1162.423 ms. Best time for 58 bit trial factors: 14.217 ms. Best time for 59 bit trial factors: 14.771 ms. Best time for 60 bit trial factors: 14.173 ms. Best time for 61 bit trial factors: 14.295 ms. Best time for 62 bit trial factors: 26.092 ms. Best time for 63 bit trial factors: 26.133 ms. Best time for 64 bit trial factors: 59.482 ms. Best time for 65 bit trial factors: 60.424 ms. Best time for 66 bit trial factors: 61.135 ms. Best time for 67 bit trial factors: 61.475 ms. Note LL test timings significantly worse due to memory bandwidth constraint, however trial factoring almost unaffected. I assume that the P-1 limits have been deepened to compensate for the reduced trial factoring depth (seems to be 2 bits less in the case of exponents around 29,000,000) Regards Brian Beesley _______________________________________________ Prime mailing list [email protected] http://hogranch.com/mailman/listinfo/prime
