Author: Vincent Diepeveen
Date: 07:03:26 03/24/02
Go up one level in this thread
On March 23, 2002 at 20:04:27, Sune Fischer wrote: >On March 23, 2002 at 19:34:11, Slater Wold wrote: > >>On March 23, 2002 at 17:39:21, Sune Fischer wrote: >> >>I believe Hyatt has done this test before. I *believe* anyway. > >OK, I just remember it differently, we'll probably see some comments from Hyatt >and co. soon ;) > >>Looking at a few articles, it seems that the speedup *might* be more than 15%, >>but not much. > >The jump from thunderbird to palomino was about 10-15%, compared to that the >jump from 32 to 64 _bit_ will be a true revolution in speed, or at least I >expect it to be :) there was no jump at all. the palomino is not faster than the tbird. If you look well the only thing that speeded up are the compilers. the alpha compiler was real good a few years ago compared to the pc compilers back then. that is no longer the case now. they handled 64 bits code in a 32 bits processor very poor. that has been greatly improved and is the main 'speedup'. 64 bits cpu's we won't see soon. a) they are hell slow because they are clocked low b) it is very expensive Even the mckinley (if you can pay it) will be only available at 1Ghz initially is what intel says now. latest MP will outgun it of course. No question. 2.6Ghz K7 versus 1Ghz Mckinley. Who wins the battle? For sure mckinley will have a hard time here! Of course you can't compare this with P4, which simply has lobotomized its L1 data cache somehow and also RDRAM runs at a slow 100Mhz with 15T latency. Mckinley is a very serious processor in that respect looking at designs. But when mckinely will be out, we won't be running on 1.67Ghz MP 2000. We will more likely be running on a very fast new intel 32 bits processor or a very fast new AMD 32 bits processor. the jump from 16 bits to 32 bits was important. The jump from 32 bits to 64 bits will be less important. Neglectible IMHO. The only INTERESTING thing is the new design of the L2 and L3 cache of the mckinley, now *that* might give a boost. megabytes of L2/L3 cache *onchip*. That is potentially kicking butt in combination with more instructions a second, because in the end the only thing that counts is how many instructions in each second the processor can execute. A clever design is #parallelinstructions x processorspeed. If you realize that Mckinley is x instructions a second times 1Ghz then that x must be real high compared to near to 3Ghz K7 versus 3 instructions a clock. http://www.specbench.org/osg/cpu2000/results/res2002q1/cpu2000-20020114-01202.html compare that with the alpha. you'll see. no speedup from being 64 bits. >I also think a lot of chess codes will be rewritten to take better advantage of >the 64 bit variables. >10-15% may be about right for ordinary programs that can't utilize the native 64 >bit dword size very well, but chess progs are different in that they are >inherently *born* 64 bit. > >>Besides, no one is going to be able to afford these chips. How many people do >>you know who have Itanuim's? > >Maybe not in the beginning, but AMD has the ability to mass produce their chips, >this should help bring down costs quickly. > >>> >>>>> >>>>>>-Tom
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.