Author: Vincent Diepeveen
Date: 14:29:09 02/09/04
Go up one level in this thread
On February 09, 2004 at 09:36:23, Aaron Gordon wrote: bandwidth doesn't count. it's random access latency that does count. did you try dieter's latency test? if so, show the times please. >On February 09, 2004 at 08:32:55, Vincent Diepeveen wrote: > >>On February 09, 2004 at 08:17:11, Aaron Gordon wrote: >> >>>On February 08, 2004 at 21:45:09, Vincent Diepeveen wrote: >>> >>>>On February 08, 2004 at 18:12:42, Aaron Gordon wrote: >>>> >>>>>On February 08, 2004 at 17:21:57, Ingo Bauer wrote: >>>>> >>>>>>Hi >>>>>> >>>>>>> >>>>>>>Shredder8Mark: >>>>>>> >>>>>>>Athlon XP 2.5GHz / 218fsb(436DDR): >>>>>>>64mb hash : 504kn/s - 3712 Shredder8Mark >>>>>>>409mb hash: 334kn/s - 2227 Shredder8Mark >>>>>>> >>>>>>>Athlon XP 2.5GHz / 200fsb(400DDR): >>>>>>>64mb hash : 503kn/s - 3712 Shredder8Mark >>>>>>>409mb hash: 309kn/s - 2227 Shredder8Mark >>>>>>> >>>>>>>Athlon XP 2.5GHz / 166fsb(333DDR): >>>>>>>64mb hash : 476kn/s - 3712 Shredder8Mark >>>>>>>409mb hash: 263kn/s - 1856 Shredder8Mark >>>>>>> >>>>>>>I'm impressed.. going from 166 to 218 resulted in a 27% increase in kn/s. Way >>>>>>>back in the day when I tested Crafty it showed no increase in kn/s from changes >>>>>>>in bus speeds (latency or memory bandwidth). Interesting... Looks like my next >>>>>>>system will be a freon cooled Athlon FX running over 3GHz and 300fsb >>>>>> >>>>>>Dont trust this Shreddermark! >>>>>> >>>>>>Check the same thing with a Fritzmark and/or Crafty. Somethings wrong weith that >>>>>>Shreddermark. >>>>>> >>>>>>Ingo >>>>> >>>>>I suspected the same.. so.. I did a few tests. The test was done using infinite >>>>>analysis from the start position. The ply next to the name of the engine is >>>>>where I took the total node count and divided it by the time to ply. >>>>>Here are the results: >>>>> >>>>>Athlon XP 2.5GHz and 384mb hash for all engines: >>>>> >>>>>Shredder 8 @ 18 ply: >>>>>218fsb: 409kn/s >>>>>166fsb: 409kn/s >>>>> >>>>>X3D Fritz @ 15 ply: >>>>>218fsb: 1116.9kn/s >>>>>166fsb: 1116.9kn/s >>>>> >>>>>Hiarcs 9 @ 13 ply: >>>>>218fsb: 275.35kn/s >>>>>166fsb: 269.23kn/s >>>>> >>>>>Junior 8 @ 17 ply: >>>>>218fsb: 1999.54kn/s >>>>>166fsb: 1987.98kn/s >>>>> >>>>>Deep Fritz 7 @ 15 ply: >>>>>218fsb: 1144.69kn/s >>>>>166fsb: 1129.83kn/s >>>>> >>>>>As you can see a higher fsb (and lower latency) did next to nothing. >>>>>ShredderMark definitely has some problems. >>>> >>>>Not at all. Shreddermark has NO problems. >>>> >>>>Shredder like DIEP just uses your RAM more efficient than Fritz&co, however >>>>unlike DIEP, shredder is doing it at a way higher nps than DIEP. >>>> >>>>That means that the number of random accesses to the RAM is really a lot bigger >>>>than it is for Fritz&co. >>>> >>>>I fully understand this from Shredder and i fear the day already that processors >>>>get a lot faster without having a L3 cache of say 64MB :) >>> >>>Do you not see that Shredder got absolutely *NO* increase in kn/s from a 30%+ >>>increase in bus speed? In the real world Shredder gets no increase.. in >>>ShredderMark it shows odd results and increases.. probably due to GUI overhead >>>(spending more time switching to various positions rather than searching a >>>position) and poor timer code. >> >>the short measuring time will have some influence. BUS speed is not holy. >>It *must* improve latency. >> >>Until you manage to proof that something improves latency, you won't find any >>speed diffs with shredder using the same cpu, i'm sure of that. >> >>A faster bus speed doesn't mean a faster latency to memory automatically. >> >>Only in general it means that. >> >>Also, you overclock your hardware just too much. >> >>I'm sure that the 1 week garantuee you give at it is sometimes not even covering >>your ****, as it will be broken after 6 days or so :) > >I already proved that when I increased fsb it decreased my latency by a >significant amount. If you recall I managed something like 100ns at 166fsb and >70ns at 218fsb (and 65ns @ 223 or so I think). This was tested with lmbench as >well as with sciencemark. Both showed the same result. So, during this test the >bandwidth went up over 30% and there was a BIG decrease in latency. No >improvements. > >As for the guarantee.. I have had NO cpus returned or had anyone tell me they >had any problems with it. They are OEM chips and AMD doesn't warranty those >chips anyway. They are lucky they get ANY warranty from me. If a chip fails I >replace it for free and take the loss. This has *NEVER* happened. > >My personal CPU has been running for over a year at 2.5GHz (from 1.73ghz) and >1.975v. No problems.. I also have a Celeron 566 running 1004MHz air-cooled. It >is my business system (accounting, filing, etc) and has been running 1.0 to >1.2ghz since I first got it a few years ago. It too has had *NO* problems. Nor >has the Athlon Thunderbird 1GHz AXIA running 1.5ghz in my fiances computer. It >too has been running many years without ANY problems. > >My server is a Duron 600 running 1GHz.. it had something like 300 days of uptime >before I took it down because the ball bearings on the fan were failing. This as >you know has nothing to do with overclocking. I just replaced the fan and the >system went back up as usual... no problems. Before that I use to run a dual >Celeron 400 @ 552MHz as the server. No problems with it, either. > >Don't blame me for you blowing up computers. If they explode for you apparently >you think no one can do it.. or no cpu will last more than 12 seconds >overclocked. Just because you cannot doesn't mean someone else cannot. If I >recall correctly you expressed the same attitude in the past in regards to chess >programming.. I'm sure Hyatt knows what I'm talking about. > >In short. If you're going to overclock.. do it properly and know you're taking >risks. If you blow up something don't go harassing the people that can >overclock. I know what I'm doing Vincent.. so I don't want to hear any comments >or opinions on the matter from you.
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.