Author: Hristo
Date: 09:00:29 11/03/05
Go up one level in this thread
On November 03, 2005 at 11:32:54, Robert Hyatt wrote: >On November 02, 2005 at 22:48:53, Hristo wrote: > >>On November 02, 2005 at 18:02:59, Robert Hyatt wrote: >> >>>On November 02, 2005 at 16:23:36, James T. Walker wrote: >>> >>>>On November 02, 2005 at 16:19:03, John Dillard wrote: >>>> >>>>>On November 02, 2005 at 15:34:30, Joshua Shriver wrote: >>>>> >>>>>>http://www.apple.com/powermac/ >>>>>> >>>>>>nice :) would make a good quad system. >>>>>> >>>>>>-Josh >>>>> >>>>> >>>>>They're making a quad system. There's not other system on the market today, >>>>>super computer or otherwise, that can process as many gigaflops of info as the >>>>>dual core G5. I just wonder if any of the chess programs will benefit from this >>>>>power? >>>> >>>>I know really nothing about computers super or otherwise but I suspect that 76 >>>>Gigaflops on the quad core is not faster than the fastest supercomputer of >>>>today. >>>>Jim >>> >>>I'm not even convinced that 76 gigaflops is doable on any microcomputer today... >>> >>>Seems like a _BBBIIIIIGGGGGGG_ stretch... >>> >>>basically one floating point operation every 13 picoseconds or so... >> >>Bob, >>it actually works-out to be about "right" at that speed. The clincher here is >>the algorithm and the domain, i.e. required output. At work we (software) had a >>contest with the hardware dudes, who were using FPGAs (and what-have-you) ... >>and outcome was that in some of the cases (less than %20) the G5 won against the >>best FPGAs you can buy. >> >>Best Regards, >>Hristo >> >>p.s. >>The Apple claim is easily verifiable because you can download their example >>applications and test it for yourself; FIR, FFT work very well under Altivec. >>The problem, obviously. is that this speed doesn't apply to every problem >>(situation). > >I had not looked, but suspected that the altivec stuff was the key. It is >always possible to find some algorithm, and specific data set, where a >particular hardware instruction set really works well. Indeed. This is exactly what Apple is selling, without actually telling people that this type of performance cannot be achieved in %95+ of the algorithms. > My old "attack" code on >the Cray boxes was the only actual working piece of code Cray knew of that >executed from start to finish with zero wait states of any kind (waiting on >memory reads, waiting on the output of a previous instruction, etc.) It only >took a couple of years of twiddling to reach that point. For a piece of code >that represented less than 10% of the total execution time after it was >completed... :) Yep. Today most people prefer to achieve practical results rather than maximum theoretical performance of the given hardware; after all we live in the age of Java. :-) Best Regards, Hristo
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.