Author: Robert Hyatt
Date: 21:18:31 12/14/03
Go up one level in this thread
On December 14, 2003 at 23:47:22, Ricardo Gibert wrote: >On December 14, 2003 at 23:05:08, Robert Hyatt wrote: > >>On December 14, 2003 at 08:25:31, Ricardo Gibert wrote: >> >>>On December 14, 2003 at 07:15:08, stuart taylor wrote: >>> >>>>On December 14, 2003 at 03:50:21, Jeremiah Penery wrote: >>>> >>>>>On December 14, 2003 at 03:31:47, stuart taylor wrote: >>>>> >>>>>>On December 13, 2003 at 23:56:06, enrico carrisco wrote: >>>>>> >>>>>>>On December 13, 2003 at 22:09:22, stuart taylor wrote: >>>>>>> >>>>>>>>Including speed, for all applications? Or for all chess programs? >>>>>>>>If not, why did they now make 2.2 (or 2.0?). >>>>>>>>S.Taylor >>>>>>> >>>>>>>What "3.2GHz" cpu do you speak of? A P4? If so, yes, you would see a nice >>>>>>>speed up even on 32-bit apps, chess or non-chess. >>>>>> >>>>>>I'm speaking of the same company, AMD. >>>>>>If the difference varies from program to program, which Ghz. of 64 bit-AMD would >>>>>>you say, is safely a speed-up (over amd/32/3.2Ghz) on ANY application (32 or >>>>>>64)? >>>>>>S.Taylor >>>>> >>>>>There is no 3.2GHz AMD processor. >>>> >>>>Oh! >>>> >>>>So whatever it was. 3 Ghz? 2.8 Ghz? >>>> >>>>Funny, I thought the 3.2 was Amd. >>>> >>>>S.Taylor >>> >>>You're thinking of the intel PIV. >>> >>>The fastest amd cpu is 2.2 Ghz. To "help" people make performance comparisons, >>>amd has instituted "performamce ratings." For instance, the 2.2 Ghz athlon xp >>>has a 3200+ PR. >> >> >>One thing I can say beyond a doubt. I have been running on this quad 2.0ghz >>opteron, and I have now seen 10M NPS in several positions. I have _never_ >>seen any single CPU machine that will produce 2.5M nps. For example, my 2.8 >>will do about 1/2 of that using one cpu, while the opteron is producing 2.5M >>on a single processor, and acutally a bit more as the NPS is not scaling >>perfectly. >> >>It is an impressive procssor, at least for what I am doing. > >Before you disdained Amd in favor of Intel. Now you are sounding very much like >a fan. More or less correct. I witnessed some K5 problems that caused a _lot_ of debugging that was unnecessary. however, I continue to look at everything and run on whatever I can, which is really almost everything being made. And the opteron has certainly impressed me, probably more than any single 64 bit chip I have tried. The new Itaniums are not bad, but they don't seem to have any significant advantage over the Opteron, and I happen to like the assembly language the opteron uses. :) > >Amd has consistently made greater strides with its series of processors than >Intel. It started with the mediocre K5. It was only a little inferior with the >K6-2. It more or less equalized with the Athlon. Now it has pulled solidly ahead >with the Hammer. A remarkable story considering it was up against a giant with >vastly greater resources. > >The Itanium is the worst processor ever produced when you consider how much >money Intel threw at it and how much they got in return. Most companies would >have been put out of business by the chip many times over. The Itanium 2 >improved, but compared to opteron, it is way too expensive. A dual opteron >system is cheaper than a single Itanium 2 and performance-wise can run circles >around it to boot. > >The Itanium 2 only gets interesting in an 8-way configuration, but the market is >too tiny for it to recoup the incredible amount if money Intel has spent. Intel >seems to be running a charity as far as I can see. > >When Intel announced it had no plans to follow suit with its own version of the >hammer, I wondered how it could be that wasn't a major blunder. What did Intel >have up its sleave? When the Hammer release slipped by about year, I started to >think maybe Intel knew Amd was going to fall on its ass. Now that opteron is >here, the only one who looks like they might fall on their ass is Intel. > >However, with Intel's resources, it can afford a lot of expensive misadventures >and still remain the industry leader, but you have to wonder how long this can >remain true. Amd continues to execute with amazing consistency. Very impressive >for such a relatively tiny company. I agree. Intel isn't stupid. But ignoring the IA-64 stuff AMD has done is a significant error, IMHO. Nothing like real X86 compatibility, with the added boon of more registers and 64 bits when you want to use 'em. There are a few compatibility issues, dealing with pointers. But in general, the thing _really_ looks good. I had no idea that I might see 10M nps on a quad-anything in 2003. The last time I saw near 10M was on the Cray 32-cpu T90, and there CB was bumping along at about 7M. Crafty runs the benchmark at 7M+ using gcc. Eugene's compiler is significantly faster. So consider me an AMD believer when it comes to Opteron. The NUMA stuff is problematic, but definitely solvable. And for duals/quads, it really is not that noticable if you are careful with shared stuff.
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.