Author: Robert Hyatt
Date: 06:46:08 10/06/03
Go up one level in this thread
On October 05, 2003 at 14:56:38, Vincent Diepeveen wrote: >On October 05, 2003 at 14:45:00, Robert Hyatt wrote: > >>On October 05, 2003 at 13:44:15, Vincent Diepeveen wrote: >> >>>On October 04, 2003 at 23:44:03, Robert Hyatt wrote: >>> >>>>On October 04, 2003 at 21:09:23, Vincent Diepeveen wrote: >>>> >>>>>On October 04, 2003 at 21:00:34, Tom Kerrigan wrote: >>>>> >>>>>>I had the chance to run my program on a dual P4 Xeon (with hyperthreading). >>>>> >>>>>which OS and what version number of the os and what release number? >>>>> >>>>>pretty crucial. >>>>> >>>>>>First off, there have been some involved arguments about the design and >>>>>>performance of hyperthreading on this board in the past. I'd like to settle one >>>>>>argument, namely that single threaded programs do not slow down when >>>>>>hyperthreading is on. Actually, my program did slow down by 1.3% but I think >>>>>>this is marginal and easily attributed to the scheduler, not hyperthreading. >>>>>> >>>>>>The odd part is that hyperthreading DOES slow down my program when running 2 >>>>>>threads. With HT off, my program searches 90% more NPS with a 2nd thread. With >>>>> >>>>>>HT on, it only searches 53% more NPS. The idle time reported by each thread is >>>>>>low and the nodes are split evenly, so it seems both processors are slowed down >>>>>>equally. What must be happening is that HT is activated some (or all?) of the >>>>>>time while searching but I have no idea what might be activating it. >>>>>> >>>>>>Also odd is that HT seems to be decreasing the efficiency of the search. With HT >>>>>>off, my program's time-to-ply is 64% faster with 2 threads but with HT on, it's >>>>>>only 21% faster. The time-to-ply:NPS ratios are 0.86 and 0.79 respectively. >>>>>> >>>>>>Running 4 threads with HT on results in a 15% NPS/6% time-to-ply speedup over 2 >>>>>>threads. >>>>>> >>>>>>In other words, there's no contest between running 2 threads (HT off) vs. >>>>>>running 4 threads (HT on). The former wins hands down for my program. >>>>>> >>>>>>-Tom >>>>> >>>>>Your thing is searching parallel nowadays and we do talk about a chessprogram >>>>>here? >>>>> >>>>>Doesn't take away that it is not easy to profit from HT. >>>>> >>>>>Basically HT only works well at intel test machines it seems. >>>>> >>>>>those do HT a lot better than non-test machines. >>>>> >>>>>it is confirmed again in www.aceshardware.com >>>>> >>>>>25% speedup (in nodes a second) for diep is just too much (single P4 EE 3.4Ghz) >>>>>i bet production machines that we can buy in the shops soon won't show at single >>>>>cpu P4 EE 3.4Ghz a speedup of 25% like aceshardware.com has tested. Anyway i >>>>>kept the executable to proof my guess there in the future when the p4 ee is >>>>>released or when i can run at a P4 3.2Ghz C (also showed 25% speedup in nps >>>>>thanks to HT for current diep version). >>>>> >>>>>best regards, >>>>>vincent >>>> >>>> >>>>Several have run this test with Crafty. SMT on is 20-30% faster in NPS for >>>>my program, on my dual 2.8, which is not a "test machine". Eugene posted >>>>similar numbers for a dual he has. Others have also reproduced this with >>>>no problems. >>> >>>Not really, all reports i saw here from non-Hyatt and non-Nalimov machines >>>report for the same versions 10-15% for crafty. >> >>And 10%-15% is _drastically_ different than 20%, right? >> >>learn some math. >> >>this varies significantly, on the same machine... > >You tested just 6 postions, so that renders your results pretty useless. No... I just posted the results with 6 positions. I have tested with hundreds of positions. WAC. LCT. Nolot. A set of 24 positions I use frequently that includes 8 tactical positions, 8 middlegame (non-tactical) positions and 8 endgame positions. Etc. My results have been pretty consistent in the 20-30% range... > >The others had tested around 30 positions. > >So even if we still take the average it's closer to 10% than it is to 30%. However, I quoted 20%-30% and it is closer to 20% than 10%. > >Nalimov just said 30% without much of a proof. Other than running the test and posting the output. > >Best regards, >Vincent
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.