Author: Robert Hyatt
Date: 08:15:15 04/13/03
Go up one level in this thread
On April 13, 2003 at 02:38:00, Tom Kerrigan wrote: >On April 13, 2003 at 01:03:17, Robert Hyatt wrote: > >>>You have been criticizing people for "bad math" this entire thread. You rejected >>>the notion of a 50%-50% division: >> >>I didn't "reject" anything. I simply said it didn't match _my_ results and >>it doesn't. I've already posted some results with wild variance. Separate > >Nonsense. Just read the quote I posted and you seemingly ignored: > >"But I don't buy the 50% stuff, the cpu is not that simple internally. One >thread will run at nearly full speed and the other gets slipped into the gaps" And? That is based on _my_ testing with Crafty, as I said. IE it doesn't happen that way for my _threaded_ application. I agreed that it _did_ happen for totally independent processes, after I tested, because I don't normally run that kind of test... But it doesn't happen for the particular program that was the subject of this thread, namely does SMT produce any _real_ speedup in Crafty. Data said yes. from 0-20% on four positions... with only one zero and the rest pretty reasonable. > >You go on to say that your assertion is supported by your results, but there is >a big difference between saying "this is how things work and my results agree" >and "my results indicate this is how things work." Again, look at the context of the entire thread. Not just the post you read. It was about Crafty and SMT. That's why I ran the tests and posted the results. > >>>"If your NPS goes up by 10%, then with a 1.7x multiplier on two real cpus, the >>>program should run 1.07X faster using SMT." >> >>This "gem" is trivially supportable by experimentation. I posted the speedup >>numbers and this happened _perfectly_. If you don't understand the math, that's >>not my problem... But it is _definitely_ correct. > >Trivially supportable by running an experiment using Crafty that coincidentally >gives you the numbers you want. Well, I don't even know that, because I don't >have the patience to find the post with your numbers. How do you explain other >programs not supporting your "math"? Absolutely no idea. Eugene posted that he has an application that speeds up by exactly 2.0 when turning SMT on. I have gotten 20-30% improvements in NPS, and 0-20% improvements in raw time-to-solution. I posted the files needed to run the test yourself. I ran them a total of six times, and things were pretty consistent. > >http://www.talkchess.com/forums/1/message.html?292584 > >>Except they don't, as I showed. Two separate chess programs is a different >>thing from one chess program using both logical processors... >> >>Why the odd behavior I don't know, yet, however. > >Right, see my reply to your other post. > >-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.