Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Anyone ever considered using SSE or 3DNow! to get more NPS?

Author: Vincent Diepeveen

Date: 14:39:19 09/04/00

Go up one level in this thread


On September 03, 2000 at 20:36:42, Aaron Gordon wrote:

Yes i investigated the instructions in MMX, SIMD and 3dnow.

Can't use it in my chessprogram. the basic problem is the parallel
problem. how do you WAIT for certain things to be done?

So basically preventing the parallel problem is using ONLY MMX for
some parts. Now this is impossible as it doesn't have enough instructions
to do things. Also it's dead slow if you use arrays a lot.

>Hello.. I've asked a couple of programmers about this but I didn't think they
>knew enough to give me a real good answer about it so maybe some of you guys
>would know. I think SIMD (Intel's SSE & AMD's 3DNow!) would help a ton if
>properly optimized. Here's a couple of benchmarks I did with my Celeron 566 @
>1057mhz. Here's the regular test comparing to a Celeron 366..

MMX is if you have gigs of data like graphics which are processed by
thousands at a time preferably. It's not meant for chess at all.
you have a parallel problem then.

>My cpu's MIPS = 2868
>My cpu's MFLOPS = 1419
>Celeron 366's MIPS = 987 MIPS
>Celeron 366's MFLOPS = 490 MFLOPS
>
>MIP wise this is only a 2.905 fold increase & MFLOP wise only 2.895
>Now.. take the SSE benchmarks into account...
>
>My cpu's integer MMX = 3335 it/s
>My cpu's floating-point SSE = 4439 it/s
>
>Celeron 366 integer MMX = 1130 it/s
>Celeron 366 floating-point FPU = 526 it/s
>
>Again integer speed increase is 2.951x and..
>FPU speedup now is 8.439x faster than a Celeron 366.. which means
>instruction-wise it would be like a 3094MHz non-SSE Celeron which I figure it's
>insanely fast. =) I'm not sure which instructions need to be used or could be
>used/converted but hopefully if at all possible SSE/3DNow! could be somewhat
>used. I think coding it for SSE would be the best bet for now since people who
>do have Athlon's could still use it and see an increase of performance to some
>extent (Athlon's, Durons & Thunderbird's have 3DNow!1/2 and partial SSE
>support). Celeron-2's, Coppermine P3's and regular P3's have regular SSE
>support.
>
>Anyway if this is at all possible I'd like to know.. I think it would be very
>interesting to tinker with to say the least. :)



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.