Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Pentium 4 and hyperthreading (to Charles Worthington)

Author: Robert Hyatt

Date: 13:22:23 02/19/03

Go up one level in this thread


On February 19, 2003 at 14:57:18, Aaron Gordon wrote:

>Hyatt has two CPUs. Each CPU will provide about a 16% boost in crafty from
>Hyper-threading so 16% X 2 = 32%. Pretty much identical to his claim.
>
>Now, I'd prefer if when he said this he did it in a bit more detail instead of
>saying, "Oh, Hyperthreading gives you a 30-33% boost". IMO it should be worded,
>"With two Xeon chips hyperthreading produces around a 30% speedup in Crafty.
>This is around 15% per cpu".

Actually, for Crafty, it varies from 20% to 30% +per+ +cpu+.

IE if you run without SMT on one cpu, and then run with SMT on one
cpu, crafty will run 20-30% faster depending on the test set you choose.
Ditto for two physical processors.  IE if your single (noSMT) processor
does 1M, then the SMTon will do 1.2M to 1.3M.  If your dual (noSMT)
does 2M, then your dual with SMT on will do 2.4 to 2.6M.


>
>On February 19, 2003 at 14:17:55, Steffen Basting wrote:
>
>>Hi!
>>I've posted some benchmark results on my PIV 3.06 earlier and your 1.33 factor
>>in speed up seems to be a little too optimistic. Crafty benefits most from
>>hyperthreading (about 20%), the usual nodes per second increase (Deep Shredder 6
>>and Deep Fritz 7) is about 10% at maximum. Perhaps Xeons perform better, but I
>>don't think the hyperthreading performance differs too much.
>>
>>----- Crafty, 1 Thread -----
>>White(1): hash 96M
>>hash table memory = 96M bytes.
>>White(1): smpmt=0
>>parallel threads disabled.
>>White(1): bench
>>Running benchmark. . .
>>......
>>Total nodes: 48447803
>>Raw nodes per second: 1211195
>>Total elapsed time: 40
>>SMP time-to-ply measurement: 16.000000
>>
>>----- Crafty, 2 Threads -----
>>White(1): hash 96M
>>hash table memory = 96M bytes.
>>White(1): smpmt=2
>>max threads set to 2
>>White(1): bench
>>Running benchmark. . .
>>......
>>Total nodes: 48586305
>>Raw nodes per second: 1429008
>>Total elapsed time: 34
>>SMP time-to-ply measurement: 18.823529
>>
>>----- Shredder 6-Mark, 1 Thread -----
>>655+-44, 490kn/s
>>
>>----- Shredder 6-Mark, 2 Threads -----
>>696+-45, 502kn/s
>>
>>----- Deep Fritz 7-Mark, 1 Thread -----
>>1433+-0, 1193kn/s
>>
>>----- Deep Fritz 7-Mark, 2 Threads -----
>>1602+-0, 1352kn/s
>>
>>Regards, Steffen.
>>
>>PS: The Crafty version I used was 18.6-smp from Bob's site, perhaps newer
>>versions are optimized for hyperthreading?



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.