Author: Vincent Diepeveen
Date: 15:51:34 09/24/03
Go up one level in this thread
On September 22, 2003 at 22:18:40, Dan Andersson wrote: > Mainly a matter of depth. They seem to follow a blueprint for reviews. I often >fail to see how their conclusions come from their results. And a certain lack of >depth and introspection in interpreting benchmark results. > Ace's is usually much more in depth and tech savvy. They also tend to throw in >the oddball benchmark to test when there is a discrepancy. > >MvH Dan Andersson I do not entirely agree. Intel testmachines provided to testers like aceshardware show simply a way better speedup for SMT for DIEP than when i let people months later test at home the same diep version. Even despite me improving diep's SMT behaviour, the claim is simply that it is 25% faster now. The only thing we can be sure of is that aceshardware *did* measure this. Also he repeats his tests several times to be sure there is no such thing as one time results possible. He simply ignores software that shows speed X at ply 13 and next test speed X+a big margin at ply 13. He tested till 13 ply each time which was a few minutes. Measurements in 1/100 of a second in nodes a second in the logfile, so it was very accurately measured. However i have put away this diepm.exe version now. saved it clearly to disk. "this version was so fast at the P4". That 100% same version i will try to test at the world champs when there is someone showing up with a P4 EE edition or P4 C 3.2Ghz machine. And i already bet now that i won't see a 25% speedup in SMT. Previous time testers reported nearly 20% speedup at 3.06Ghz P4 in SMT. Later some persons i know bought a 3.06Ghz P4 and their tests revealed *way* smaller speedups for SMT. Testers then said, after consulting intel that the testmachines had improved SMT. But those users had bought the chips months *later*. 10% versus 18% is like 8% difference. Now the claim is 25% speedup. What we know is that the tested P4s have a 800Mhz bus and we can read also what type of RAM was used. Yet we also know that there is no diference in speed between P4 EE and P4 C edition. That's tested carefully. Yet the 8% difference advantage the A64 has is based upon an overclocked A64, overclocked P4 EE/C editions (intel ships them overclocked if i understand well, i do not know what Johan means by that though) is IMHO not the real difference. If you buy such a processor in the shop with ram belonging to it, you'll find out it's *way more* than 8% difference advantage for the A64. So we will see within a few months whether intel did the same trick again. If so then i'm going to complain loudly publicly in all these forums. It happened once, let's not let it happen twice.
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.