Author: Robert Hyatt
Date: 12:19:04 08/31/99
Go up one level in this thread
On August 31, 1999 at 14:45:53, Marc van Hal wrote: >There comes a program with this card wich writes new codecs for the chessprogram >the way the original program's are written will be reprogrammed >so the prediction branch will be beter used so actualy a software change and not >a hardware change only the hardware will be beter used I doubt that this can be done without profiling and using a profiling optimizer. Current microprocessors are _very_ good at branch prediction. They assume forward branches are not taken unless they show up in the branch target buffer as being taken previously. They assume backward branches are always taken (loops) unless the branch target buffer suggests that was not taken two times in a row... IE _good_ compilers let you run with profiling turned on, they watch the code to see what branches are taken and which ones are not. Then you recompile using that profiling data and the optimizer uses that observed data to produce a better executable. But a 'static analyzer' isn't going to work. My own opinion is that this is just another "new spark plug" idea that people will pay for, but which produces no advantage at all. It is easy to test, of course, by running the same program, on the same position, 'before' and 'after' and seeing how much faster it really runs. I'd predict no change, personally... for compute-bound chess engines. The description you gave sounds like something quoted from a 'snake-oil' advertisement. The 'card' apparenly does nothing then, since it can't affect cpu performance anyway... without real data, I would avoid this like the plague. It sounds more like an algorithm to transfer money for your pocket to the seller's pocket, than it sounds like a product to make a chess engine run faster... caveat emptor certainly applies here...
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.