Author: Vincent Diepeveen
Date: 08:35:07 03/26/02
Go up one level in this thread
On March 26, 2002 at 10:51:53, Sune Fischer wrote: >On March 26, 2002 at 10:37:30, Vincent Diepeveen wrote: >> >>my argument is that the big and advanced branch prediction of the >>alpha and the 4 instructions versus 3 of the alpha and the huge >>L1 and L2 caches of it, add to that cheating on specbench; all that >>together brings 33% speedup compared to a 32 bits processor. >> >>I do not see the 32 to 64 bits speedup for the alpha at all. For sure >>not a factor 2 as claimed at some places. > >But you keep comparing a 1.6 GHz K7 to a 1 GHz Alpha, naturally that will even >things out. But the Hammer won't be running at 1 GHz. > >>If you claim 10-15% that's already a far smaller claim than other >>claims i saw here, majority is still claiming factor 2 to my amazement. >> >>Truth is that it is way harder to clock a 64 bits processor at 3Ghz >>than it is to clock a 32 bits processor at 3Ghz. > >Yep, that's where I have to agree with you, it might be that the Hammer will end >up running 500-1000 MHz slower than the cheaper 32-bit chips, at least in the >beginning. But 32-bit won't stay around forever. > >Actually the original post only said 10-15%, but compared to what? >I just assumed is was a K7 32-bit chip at the same clock speed. > >>We will see when we can afford processors that are 64 bits AND doing >>that faster than 32 bits equivalents. > >Most bang for the buck is a different discussion.... it is the only question. If intel comes out with a P5 that runs your program at 5Ghz and gets 2 million nodes a second, then you sure buy that one if the alternative is a hammer which is 64 bits and runs at 2Ghz and gets your program 1 million nodes a second. >-S.
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.