Author: Ricardo Gibert
Date: 15:55:10 01/10/02
Go up one level in this thread
On January 10, 2002 at 17:15:11, Tom Kerrigan wrote: >On January 10, 2002 at 05:52:34, Ricardo Gibert wrote: > >>You already mentioned SMT. This is one way of making a chip perform better >>without clocking it higher, so there is no reason to accept your "_only_" as >>true. > >I meant single-thread performance, sorry if I didn't make that clear. >Single-thread performance is ultimately the most important thing in computer >performance, because you can always parallelize chips later. I'd much rather >have one single Pentium 4 than a couple hundred 486s in parallel. > >>I would accept that whether a higher IPC or higher clockability is better is >>*not* a settled question, so saying that the P4 is inferior, because it employs >>an inferior strategy is unwarranted. > >I believe it is a settled question. Microprocessor people call this the brainiac >vs. speed demon contest, and it was won by speed demons. There is no question in >my mind that Intel is doing the right thing by going after higher clock speeds. > >http://www.mdronline.com/mpr_public/editorials/edit13_17.html > >-Tom It fails to explain why the P4s clock twice as fast as the alpha (the big star of the article), but is still only about as fast at SPECint2000. See http://www.aceshardware.com/SPECmine/ where alpha has slightly higher peak and slightly lower base score at SPECint2000. At SPECfp2000, the alpha is clearly faster despite having half the clock spped of the P4) Higher clockability doesn't score a decisive victory there, yes? It does mention that bandwidth is also important in addition to IPC and clock, but qualifies this with, "For applications larger than the meager SPECint95, designers must supply adequate memory bandwidth to keep up with the fast CPU core." It seems the real picture, is not so simple. It isn't just a higher IPC vs higher clock issue as evidenced by the alpha vs the P4. CPUs are too complex for such a simple explanation. I think the article suffers by trying to make itself comprehensible to lay persons like me. Still, the article is quite interesting. Thanks for the link. You are are most probably correct that overall the evidence is generally leaning very much toward yours and the articles interpretation, but as far this particular article goes, I don't see it as settled...yet. Give it another week maybe ;-) BTW#1, the alpha crushes the P4 at the crafty benchmark, but the Athlon annihilates the alpha even though the alpha is a 64 bit CPU running a bitboard program. Only a small, but interesting part of SPECint2000. BTW#2, The POWER4 cpu at "only" 1300mhz outshines everybody at SPECint2000 *and* SPECfp2000. Is it like the POWER3, the "brainiac" approach according to the article or more like the P4, the "speed demon" approach. Hmmm.
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.