Author: Robert Hyatt
Date: 08:48:11 12/03/02
Go up one level in this thread
On December 02, 2002 at 23:31:10, Matt Taylor wrote: >On December 02, 2002 at 22:46:25, Robert Hyatt wrote: > >>On December 02, 2002 at 20:07:15, Aaron Gordon wrote: >> >>>I'm actually hoping Hyperthreading takes off. What I've seen of it so far hasn't >>>been too impressive though. Perhaps applications need to be recompiled to better >>>support it. Anyway, if it's able to take off most likely much more applications >>>will start having SMP support. When this happens there will be all the more >>>reason for AMD to introduce the dual-cored Sledgehammer systems. One can only >>>hope. :) >> >> >>Eugene also reported that running two instances of the tablebase compression >>code >>on a single hyper-threaded cpu runs almost exactly twice as fast. _that_ is >>impressive performance... > >Interesting. I mentioned in another post that the P4 could potentially execute >almost 2 times faster due to the fact that it has the ability to execute 5 ALU >ops/cycle while the trace cache can supply at most 3 ops/cycle. It also has a >slight advantage over Athlon in that the L1 data cache is 1 cycle faster (2 >instead of 3) and would handle the extra strain of another virtual processor >much better. > >I wish Itanium were affordable. I think that architecture would -kill- IA-32. I >think it has something like 4 KB of register file for applications alone. >Athlon64 looks promising, though. They've got 16 general registers now to reduce >the memory contention. At least it will be affordable... Itanium2 looks very good. But a bit pricey so far. Hopefully this will drop when AMD introduces something competitive... > >Is the algorithm in Crafty at all suitable for running in a distributed system? >(I would be merrily amused if a small collection of PCs could outperform the ~1 >billion moves/sec that Deep Blue is capable of evaluating.) Not yet, but one day, yes. The message-passing is the bottleneck there and at the moment, crafty depends on shared memory... There are ways to "simulate" shared memory over a network, but the latency doesn't go away and it hurts badly.
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.