Author: David Franklin
Date: 19:10:17 06/14/00
Go up one level in this thread
On June 14, 2000 at 09:08:52, Mogens Larsen wrote: >On June 14, 2000 at 07:02:47, Adrien Regimbald wrote: > >>I'm overjoyed to hear this! ;) I have been getting sick of the strategy of >>making constant tiny improvements to the existing architectures - a move to a >>completely new architecture would be tremendous! >> >>From my classes on architecture, there were a lot of _very_ interesting >>optimizations done to CPU architecture and strategies at the start of the PC >>boom. Lately though, it has just been a mantra of "smaller smaller smaller, >>more more more!" with no interesting advances made - I can't wait to see what >>they've done for the new architecture :) > >I just wonder about one thing. If the architecture is so advanced, why would it >be necessary to start at 1.4GHz? Presumeably such an advanced product would also >outperform Pentium III's at current speeds. Why *shouldn't* it start at 1.4Ghz? There are always tradeoffs to be made; one possible tradeoff is to accept a low number of instructions/clock, if that allows you to have a very high clock rate. Of course, it seems Intel have gone a *long* way down that route with Willamette - some parts are actually double pumped (effectively 2.8Ghz), but the tradeoff is a *really* deep pipeline, and pretty high latencies for a lot of instructions. It doesn't look like a winning strategy to me, but only time will tell. Architecturally, the one thing Intel have done is basically provide a completely new FP instruction set, which should get round the whole FPU stack nonsense we've been stuck with so far. I'm not sure how much it will help the SpecFP figures, but a fair bit, I suspect.
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.