Author: Ralf Elvsén
Date: 04:54:21 04/10/02
Go up one level in this thread
On April 10, 2002 at 00:35:12, Robert Hyatt wrote: >On April 09, 2002 at 16:26:28, Ralf Elvsén wrote: > >>On April 08, 2002 at 23:25:16, Robert Hyatt wrote: >> >>>On April 08, 2002 at 06:57:56, Vincent Diepeveen wrote: >>> >>>>On April 07, 2002 at 12:09:12, Robert Hyatt wrote: >>>> >>>>>On April 07, 2002 at 01:00:55, Russell Reagan wrote: >>>>> >>>>>>Last night I was cruising the CCC archives reading over discussions on various >>>>>>data representation approaches in chess programs, and I came across this >>>>>>statement from 1999 in this post: >>>>>>http://www.it.ro/ccc_search/ccc.php?art_id=39708 >>>>>> >>>>>>"Just wait 5 years and see if you can find a 32 bit machine left." >>>>>> >>>>>>Well we are 2 years away from the 5 year deadline. Do you still agree with this? >>>>> >>>>>Yes. The end is "in sight". >>>>> >>>>> >>>>> >>>>> >>>>>>I'm not trying to throw old quotes back at you or anything. I'm really more >>>>>>curious about how much longer it will be until 64-bit machines are the norm. >>>>>> >>>>>>I'm also curious if when the time comes that 64-bit machines are the norm, if >>>>>>they will be on par with the Hz speeds of the 32-bit machines. For example, >>>>>>right now you can get a 2.1 GHZ Athlon or 2.4 GHz P4 without having to take out >>>>>>a loan. If you want a 64-bit Itanium, you're looking at $2,500 - $7,000 for a >>>>>>chip that runs at 733-800 MHz (www.pricewatch.com). So once 64-bit machines are >>>>>>practical from a price standpoint, will they still be at a third of the speed we >>>>>>can get from a 32-bit machine? >>>>> >>>>>64 bit machines can run at identical clock speeds. The issue is price. As >>>>>demand increases, price will go down driving performance up. >>>>> >>>>> >>>>> >>>>>> >>>>>>If my data from pricewatch isn't entirely accurate please correct me. E.g. if >>>>>>there are other 64-bit chips that are cheaper and faster than Itanium. Heck, how >>>>>>much would a good 64-bit system cost today? >>>>>> >>>>>>Thanks, >>>>>>Russell >>>>> >>>>> >>>>>There are plenty of chips cheaper than Itanium and far faster... but Intel >>>>>is going to set the standard. The other vendors (HP, DEC/COMPAQ/whatever they >>>>>are today, MIPS, IBM, have been doing 64 bit chips for years. Intel is _way_ >>>>>behind... >>>> >>>>I am not so sure about this. Did you already try mckinley? >>>>6 integer units or something? >>>> >>>>that'll kick butt of course. >>> >>> >>>I was just contacted by someone working on a SPEC test for a Mckinley. He >>>reported a speed of roughly clock speed / 1000 for Crafty on that machine. >>>Which is very good. The 21264 produced a clock speed / 750 which was a bit >>>better... Note that the above produce raw NPS numbers. clock speed is in >>>millions or billions of clock cycles per second. IE the 21264 we tested was >>>a 600mhz machine producing 800K nodes per second. 800K = 600000000 / 750. >>> >>>Eugene might be able to tweak this higher since he is in the compiler group >>>up at MS... and has access to more recent compiler versions... >> >>Will there be support for McKinleys PopCnt-instructions in the >>official Crafty? Isn't this required if they are to be used in >>SPEC? >> >>Ralf > >Actually, in SPEC no asm is allowed for the crafty part, as it is all C. It >must remain all C. The version distributed for SPEC did not have the x86.s >code included... because that would favor one particular architecture and >SPEC doesn't allow that for obvious reasons. > >What you _might_ see is some library routines to do the popcnt stuff and >modifying the source to use those is apparently O.K... Yes, that is what I guessed after thinking about it some more. I don't understand why there are no library routines that could do this stuff for us if we use a machine (e.g. x86) that have these nice instructions. But there is probably some reason for that. Maybe then the compiler could do it even better than inserting hand crafted assembly, but I don't know about that. Thanks for the answer Ralf
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.