Author: Hristo
Date: 03:04:05 02/09/02
Go up one level in this thread
On February 09, 2002 at 05:34:09, Gian-Carlo Pascutto wrote: >On February 09, 2002 at 04:50:22, Hristo wrote: > >>Here is a report from the field. >>Exactly one week ago I bought a Mac. My very first Mac. >>A dual 1GHz G4 + 1GRam and all the other normal stuff. >>I?m absolutely impressed!!! After the initial shock of everything being new and >>different I must admit that this is the easiest ?thing? to use, administer, >>configure and it is by far the prettiest interface and hardware out there. I >>still have the AMD and Intel boxez (indeed) but they have not seen much use >>lately. The only drawback, which to me is significant, is that linux+XWindows >>doesn?t run on this Mac (yet). The video card ... again, other than that >>Yellowdog linux boots just fine. >> >>The interesting thing is that the G4-1GHz outperforms the PIII-1GHz, P4-1.8GHz >>and Xeon 1.7GHz. My test was an FFT routine (class) I use at work. The code is >>not optimized in any shape or form for the G4. The worst case is when doing >>64K-samples FFT (using floats). >> >>G4 - 18 fft/sec >>Xeon 1.7 - 6.04 fft/sec >>P4 1.8 - 5.93 fft/sec >>pIII 1 - 5 fft/sec >> >>Yes ... the cache got the best out of the Intel Chipsen. ;-) >>Until this point the G4 was still better than the rest of them. All of the CPUs >>scaled ?n log(n)? in the range of 1K-...-32K samples. >> >>If interested I can send you the source code to test yourself. >> >>It, almost, seems that one can write a chess program using ?float? instead of >>?int? and still get the same performance. >> >>BTW >>Does anyone have test(s) you would like to try on a G4?! If reasonable I might >>do it. > >Keep in mind the FPU of the P4 is horrible. Much slower than the Athlon. Yes ... I remember testing the Athlons some time ago and they also beat the Intel, however, the difference was not so huge. >However, it's SSE2 module is very fast _and_ optimized for just the kind >of thing you're testing. So, if possible, I'd be interesting to optimize >your code for Pentium 4 (if only by recompiling with Intel >C+autovectorization) and see what comes out. I forgot to mention. The code is entirely in C++ and templates. On the Mac I use OSX an the native compiler (modified version of gcc 2.95.xxx) with -02. On the Intel I use MSVC with /G6 /Ox and other flags for max performance. I do not have Intel compiler. Do you want the code?! Is the address in your profile correct? I’m curious to see what SSE2 does, as well as, the Altivec thing. I might spend some time doing the Altivec, cause it is so simple. ;-) Cheers. hristo > >-- >GCP
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.