Author: David Eppstein
Date: 11:28:43 04/24/99
Go up one level in this thread
On April 24, 1999 at 07:30:18, Ian Kennedy wrote: >On April 24, 1999 at 03:21:29, Bruce Moreland wrote: >>The 767 mhz Alpha was almost exactly 767/533 times faster than the 533 mhz >>Alpha. A single-processor 450 mhz P2 is about 20% faster than the Alpha 533. >>So it goes at maybe 640 "Alpha" mhz when running my program. >>I bet a 500 mhz P3 is just a tiny bit slower than the 767 mhz Alpha. > >If you look at the specint95 marks which I think are reasonably useful (as >benchmarks go!) for chess programs, they both (PIII/500, Kryotech 767) score >around 23. However don't forget the 21264 is a lot faster, 27 at 500Mhz and 30 >specint95 at the latest 575Mhz. They also have much more sensible on-chip cache >sizes of 2x64k instead of the diddly 8k primary cache of the 21164 which >preduces phenomenal cache miss rates on my program PSYCHO. There's a fascinating article about the 21264 processor design from the same issue of IEEE Micro as Hsu's blurb about Deep Blue: http://dlib.computer.org/mi/books/mi1999/pdf/m2024.pdf (you may need to subscribe to be able to download this; I am not a subscriber but I think my institution has a site license). The "simple" sequential RISC model presented to the programmer hides phenomenal intricacies of multiple parallel pipelines, out-of-order execution, speculative branch prediction, up to 80 instructions in progress at any time...I guess the same must be true for recent PPC and Pentium machines too. Relevant to chess programming, the 21264 has instructions for popcount, find first one, and find last one, however for some reason they are a little slower than other simple integer arithmetic operations (3 cycles instead of 1).
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.