Author: Peter McKenzie
Date: 15:43:30 04/08/99
Go up one level in this thread
On April 08, 1999 at 13:27:53, Dave Gomboc wrote: >On April 08, 1999 at 13:15:37, David Eppstein wrote: > >>On April 08, 1999 at 00:50:38, Dave Gomboc wrote: >>>My understanding is that the performance of JIT compilers on cpu-intensive >>>code is approximately 40x slower than C++. (Does that figure sound realistic?) >> >>No, I think the factor is more like 2x-3x if even that. I can get 200K nodes per >>second for the game program I wrote (not for chess, the game is simpler, but >>even so...) using Apple's JIT on a 233Mhz G3. > >Hmmm. It's hard to beat real evidence. :) The same fellow who quoted me "40x" >loss (actually 20x to 80x) also told me it's hard to tell for sure. He >specifically mentioned that Apple's latest JIT was "5x" better than the previous >iteration on some of his code, and that in general the technology was still >changing rapidly. But 2x sounds quite bearable, for research, anyway. > >Dave Gomboc I think 2x-3x is much closer to the truth, although I've never tested it - perhaps its even better than 2x. I'm basing my opinion on the performance of the java chess program Grok on ICC (by Peter Kappler who would probably partake in this thread if he were around). I doubt Grok would be able to achieve its 2300+ rating with a 40x speed handicap!! Peter
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.