Author: Dave Gomboc
Date: 16:05:19 09/13/99
Go up one level in this thread
On September 13, 1999 at 13:47:01, Robert Hyatt wrote: >On September 13, 1999 at 12:39:11, Thorsten Czub wrote: > >>On September 13, 1999 at 08:37:05, blass uri wrote: >> >>>On September 13, 1999 at 05:51:40, Thorsten Czub wrote: >>> >>><snipped> >>>>this is a shallow evaluation. In my games against Hiarcs7.32 CSTal >>>>gets 50%-. >>> >>>Maybe Hiarcs does not like your hardware. >>>I know that in one of your games Hiarcs7.32 did a stupid blunder that my >>>hiarcs7.32 could not reproduce and hiarcs can see from the first ply that it is >>>a blunder. >>> >>>I know that mark young claims that he did not find strange blunders in games >>>when Hiarcs used the pentium. >> >>my amd k6-400 gets normal benchmarks for hiarcs. >> >> >>>Uri > > >I am hardly an Intel bigot, as I happen to love alphas. :) However, here is >one horror story about a non-overclocked AMD. > >A student came in to my office about 2 years ago or so wanting to buy a good >computer that would have enough horsepower for his Ph.D. research in data >mining. I had heard good things about AMD and he ended up buying a K6. He >seemed happy at first, but came by one day with a problem where his program >was crashing unexpectedly. I gave him some suggestions on debugging, but none >helped. It ran _so_ long to crash, he asked if he could also use a couple of >the machines in one of the labs here so he could do different debug tests at the >same time, and I said "sure." He dropped this program into his AMD and two >Intel boxes, and suddenly they were running to completion on the intel boxes, >but still crashing on his AMD. He tried the same data exactly, and it would >crash on his AMD, but run on the Intel boxes. After a lot of searching, it >turned out that AMD had released a bad batch of cpus with a known math error >in them (does sound like the infamous intel floating point bug, to be sure) and >he had gotten one. After 3 months of work, he found that _all_ of his testing >had been for nothing... a big waste of time... as his CPU was bad. > >Others here have reported bogus results with crafty, and upon further >examination I have found that they are usually running on an AMD that is >overclocked. That seems to work for a while, but as the cpu heats up, it >starts producing some sort of funny math error that will screw things up in >ways that are totally unreproducible. > >For those reasons, I stay away from them myself, and will continue to do so >until I see 'saner' behavior from the processors. Finding that sort of problem >is nearly impossible without hardware logic analyzers and so forth, and at these >clock speeds, such analyzers are not cheap. > >YMMV of course... My understanding is that AMD doesn't leave as much tolerance as Intel does when they rate for speed. Consequently, it is difficult to safely overclock AMD chips. Dave
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.