Author: Robert Hyatt
Date: 17:11:57 12/31/02
Go up one level in this thread
On December 31, 2002 at 14:54:49, Christopher A. Morgan wrote: > >Bob, > >Thanks for the clarification. > >In case (2) does the second program actually have to be executing code, that is >running, or just open, for the chess program that is running, to experience the >speed up? Seems to me if two different programs are running they both slow >down, although I understand under HT with two programs running they would both >run faster than on a chip without HT. Hyperthreading depends on two instruction streams being executed in parallel. If you have a chess engine that runs two threads, it will run faster with SMT on. If you run two programs in parallel, they will run faster overall than they will with a single cpu enabled... A single thread will _not_ run faster on a SMT-enabled machine, whether there is another thread to execute or not. > >You also say, “For a non- threaded application, HT will not buy much...” Does >this mean even with two non-threaded programs running on HT chip will not speed >things up that much? By that I mean either program’s speed on the HT chip >compared to its speed running alone on a comparable chip but without HT will be >~ the same. That is correct. however, both _together_ will run a bit faster. For example, program A takes 300 cpu seconds to run, program B takes 300 cpu seconds to run. If you run them on a single-cpu, no hyper-threading, they will take 600+ seconds total. The + comes from O/S overhead and cache trashing as context switches are done. If you run both at the same time on a SMT processor, they might both run in something like 500-520 seconds total. Note that _neither_ runs in less than 300 seconds, but together they run in less than 600, which is the point of SMT. > >Thanks. > >Chris > > >On December 31, 2002 at 10:54:54, Robert Hyatt wrote: > >>On December 31, 2002 at 03:08:54, Christopher A. Morgan wrote: >> >>> >>>Bob, >>> >>>Are you saying a P4 chip with HT enabled should run 20-30% faster than a P4 chip >>>with the same GHz that does not have HT, or anyway to use HT, at least in >>>Crafty, and presumably in other chess engines as well? Thanks. >>> >> >>Yes, assuming one of two things: >> >>(1) the program you run is threaded. IE the "deep" programs, or a program >>like Crafty that understands parallel search; >> >>(2) you run two programs at the same time, so that there are two threads of >>executable code ready to run at any instant... >> >>For either of the above, HT will speed things up significantly. For a non- >>threaded application, HT will not buy much... >> >>> >>> >>>On December 30, 2002 at 22:13:43, Robert Hyatt wrote: >>> >>>>On December 30, 2002 at 19:39:23, Frank Koenig wrote: >>>> >>>>>Two questions. >>>>> >>>>>One) Will Intel's HT technology be able to help chess programs above and beyond >>>>>just allowing one CPU to appear as two? >>>>> >>>>>Second) If you are running XP, will HT require XP Pro instead of XP Home to take >>>>>advantage of it? >>>>> >>>>>Thanks, >>>>> >>>>>Frank >>>> >>>> >>>>I can't answer the last question.... As to the first, you can expect a parallel >>>>searcher to run 20-30% faster using two "virtual cpus" than using one real cpu. >>>> >>>>YMMV depending on the program however, as my numbers are numbers produced only >>>>by Crafty on a 2.8ghz xeon...
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.