Author: Eelco de Groot
Date: 19:49:39 10/25/99
Go up one level in this thread
On October 25, 1999 at 17:04:13, James T. Walker wrote: >On October 25, 1999 at 16:09:17, Michel Langeveld wrote: > >>On October 25, 1999 at 15:47:23, Doody Ungson wrote: >> >>>Does anyone out there know the reasons ahy the AMD K6 is the favored CPU as >>>compared to the Intel processors for the same megahertz speed? If so has anyone >>>compared the K6-2 vs. K6-3 chess computing benchmarks for the same megahertz >>>speed? Any reference to a URL or a separate article about this subject matter >>>will help. >> >>http://www.rebel.nl/bench.htm > >The Rebel site may give misleading results. The Rebel 10 program benefits >greatly from the K6-3 CPU. Not all programs do this. For example Fritz seems >about the same on the K6-2 and K6-3 for the same Mhz. For comparison to the PII >vs K6-3 using Rebel I found that Rebel on the K6-3-450 runs 2X faster than my >PII-333 in the middle game and 1.72X faster than the PII-333 in the endgame. >This from only a few positions I tested. That makes the K6-3-450 more like a >PII-600 for Rebel. Fritz/Junior/Nimzo99 do not gain much more than the expected >difference in Mhz. Hiarcs 7.32 also gains slightly more than the Mhz would >indicate. >Jim Walker As stated before in some very long messages, size of the second level cache compared to size of the evaluation function seems to be an important factor as well as the fact whether the L2 cache resides "on die" as in K6 III, the newer Celerons and in the very new 'Coppermine' from Intel. (There are probly more, I suppose Alpha also falls in this category?) Some further news about Coppermine, the "on chip" second level cache will be 256 kB by the way, same as K6-III where Celeron has 128 kB, so not 512 kB as I wrongly assumed, to be read at: http://news.cnet.com/news/0-1003-200-851003.html I now think a 733MHz Coppermine might just beat the Kryotech AMD K6-III 600MHz in the Rebel Benchmarks! Still, Celerons don't do that badly either for many chessprograms. Eelco (Quotes from some earlier posts:) ------------------------------------------------------------------------------------------------------------------------------ Posted by David Blackman on September 23, 1999 at 04:01:38: In Reply to: Re: Benchmarks, question about Coppermine posted by Eelco de Groot on September 23, 1999 at 00:42:28: On September 23, 1999 at 00:42:28, Eelco de Groot wrote: >For years second level cache runs at 66 Mhz and recently this was >increased to 100 Mhz. > >The AMD-K6-III is a breakthrough because the second level cache runs >at full processor speed. If you have (say) a 450 Mhz K6-III the >second level cache will run at 450 Mhz too instead of the old 100 Mhz. > >Ed Schroder This second level cache at full processor speed is made up by the marketing departments. The L2 on AMD K6-III and Intel Celeron are actually a bit quicker than most of the off-chip caches, but nowhere near true CPU clock speed. Celeron is usually about 11 clock cycles to L2 cache. K6-III is unpredictable. About 11 clock cycles is the fastest i saw in testing, but sometimes it was much slower for no obvious reason. The 100MHz or 66MHz off chip caches are also a bit misleading. You might think this means an access time of 10ns or 15ns, but most chip sets will actually be 50ns to 100ns. ---------------------------------------------------------------------------------------------------------------------------- David, isn't it strange that there is such variability in results for the K6-III? I mean if the benchmark would be a simple assembly language program, what could then cause the differences that you mention? Thanks for your comments! I found one article in Dutch, by the way, coming back to my original question, that said that Coppermine will have L2 cache "at full speed". So I think that for chessprograms with large evaluation functions we can expect an increase in speed compared with the Celeron. Rebel's Kryotech benchmark for the K6-III 600MHz will be hard to beat though I think, because of its extra fast memory, L3 cache I suppose, at 400 MHz. ----------------------------------------------------------------------------------------------------------------------------- (End of quotes)
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.