Author: Jeremiah Penery
Date: 15:31:50 03/26/02
Go up one level in this thread
On March 26, 2002 at 00:55:31, Sune Fischer wrote: >On March 25, 2002 at 18:15:54, Jeremiah Penery wrote: > >>On March 25, 2002 at 08:48:54, Sune Fischer wrote: >> >>>On March 25, 2002 at 08:00:18, Vincent Diepeveen wrote: >>> >>>>On March 25, 2002 at 07:52:12, Sune Fischer wrote: >>>> >>>>>On March 25, 2002 at 07:05:26, Vincent Diepeveen wrote: >>>>> >>>>>>The only INTERESTING thing is how fast a processor + compiler performs >>>>>>for a program. If you build a 1Ghz processor, then it gotta beat >>>>>>a 2Ghz K7 simply. If it doesn't, THEN YOU ARE SLOWER. >>>>> >>>>>Actually, it might beat it: >>>> >>>>>"In SPEC CFP2000 the Alpha 21264A running at 667MHz can outperform our beloved >>>>>AMD Athlon at over 2x the clock speed, not to mention that Intel's own Itanium >>>>>only runs at 800MHz while providing even higher scores." >>>> >>>>that's the floating point unit Sune. not a single chessprogram is >>>>using much floating point. also floating point isn't faster than >>>>integers (otherwise we could rewrite stuff to floating point). >>>> >>>>Just look how fast the best prepared alpha machine at specINT is >>>>completely outgunned by XP2000 (=mp2000). >>> >>>That is true when talking about the Alpha, it was probably not designed to be >>>very integer fast, I've even heard it couldn't do integers, that it would just >> >>The Alpha is just as much faster than other processors in Integer operations as >>it is in FP ones. > >Well lets not split hairs on this, I think the alpha was designed primarily to >do floating point ops. It is still pretty fast on the integers though: >http://www.redhill.net.au/hw-cpu-test-nonx86.html The primary design philosophy of all 64-bit RISC processors is pretty much straight FP performance. Some of the chips relatively suck at integer operations, Alpha doesn't. >>>cast from floats. Don't know if that is true or not, but why would the Hammer >>>have the same weakness? >>>Look at the specs for the Hammer, it looks as though it will be 2x faster at >>>64-bit int-operations. >> >>2x faster than what? Everything I've read indicates the Hammer will be about >>25% faster than the AthlonXP clock for clock in 32-bit mode, and going to 64-bit >>mode will give another 15-20% speed boost, mostly due to the extra GP registers. >> Maybe for a bitboard-based program like Crafty, it would get even more of a >>speed boost. > >The second part of thet doesn't really make any sense to me, "15-20% speed >boost" on 64 bit-operations???? Compile program "A" in 32-bit mode. Compile program "A" in 64-bit mode. It will run 15-20% faster in 64-bit mode. Most of this speedup is because in 64-bit mode, there are twice as many GP registers to work with. >First of all, why would Crafty then "get even more of a speed boost" and second Because Crafty is using 64-bit integers a lot, which aren't so hot on 32-bit processors. The 15-20% I said above was for the general case - the everyday, 32-bit programs, recompiled in 64-bit mode. >why not a clear factor 2 in speed (assuming everything runs in cache so not to >waste bandwith)? Crafty might get nearly a factor of 2 in speed. Nobody knows until we can see it run on that processor. Any "normal" program won't get nearly a factor of 2 in speed from being recompiled in x86-64 mode. >The registers are 64-bit, so that is twice the operational bit capacity clock >for clock over a 32 bit chip. Most programs are not using 64-bit integers, so this doesn't affect them much. The biggest speedup for them comes from having twice as many registers to work with.
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.