Author: Robert Hyatt
Date: 09:29:16 02/15/00
Go up one level in this thread
On February 15, 2000 at 12:15:14, Peter Kasinski wrote: >On February 15, 2000 at 09:53:37, Robert Hyatt wrote: > >>On February 15, 2000 at 07:12:17, Peter Kasinski wrote: >> >>>On February 15, 2000 at 01:25:41, Tom Kerrigan wrote: >>> >>>>On February 14, 2000 at 23:49:03, Peter Kasinski wrote: >>>> >>>>>The notion that they all do something ?absolutely stupid? by not yielding to >>>>>what often is nothing more than a marketing ploy (like releasing under-marked >>>>>chips) is boring. Most motherboard manufacturers today release boards supporting >>>>>bus speeds in 1 MHz increments. Why do you suppose they do that? And where >>>>>exactly do you buy memory chips afraid of 75 MHz anyway? >>>> >>>>Maybe your processor is part of the "marketing ploy." Maybe your video card is, >>>>too. >>>> >>>>But are you willing to bet that everything in your PC is underclocked? >>>> >>>>Do you think the entire computer industry is trying to trick you? >>>> >>>>IE, are you sure that your modem and sound card and network card and graphics >>>>card are so well designed that you can overclock them all by 14% and they will >>>>all work perfectly? >>>> >>>>If you believe this, good luck. >>>> >>>>-Tom >>> >>> >>>It's called progress. Computer manufacturers don't build to specs from 8 years >>>ago. Like your memory chips that fail at 75 MHz most of these problems don't >>>exist in today's world. >>> >>>The entire world is not trying to trick me. Don't be do condescending. >>>And marketing ploys do, occasionally, exist. Finally, it's quite ridiculous that >>>you sound willing to bet against the stability of my computer. I guess lack of >>>_any data whatsoever_ doesn't stop you. >>> >>>PK >> >> >>I hate to tell you this, but Tom is right. If you increase the bus speed, >>you _greatly_ increase the odds of very difficult-to-find system problems. IE >>SCSI or IDE controllers are very complex. And just because they run at 100mhz, >>does _not_ mean that they will run at 112.5, for example. And when they fail, >>they generally fail 'softly'... and just produce bad data. Just visit the >>linux mailing lists to see the horror stories. If every device in your >>computer has a 90% chance of running correctly overclocked, and you have 10 >>such devices, the probability of _everything_ running correctly is almost >>zero (try .9^10 for a horrifyingly clear data point.) >> >>Memory is very sticky. If you are running a FSB at 66mhz, and want to overclock >>to 75mhz, take your memory to a good local repair shop and let them run it on >>their tester at 75mhz. You might be surprised that it doesn't certify. Memory >>is already pushed to the edge.. > > >I agree with the premise. But they don't sell 66MHz memory any more. Not in >Toronto at least. They carry PC/100 and PC/133 these days. When they did _I did >witness such tests_ and in the last two years it invariably passed the 100 MHz >test. > >I could go through similar examples for video cards, IDE controllers, etc. You >can hardly push new hardware these days by running the bus at 75MHz. Vendors >know that. That's why Abit, ASUS, etc. make motherboards with 1MHz bus speed >increments. > >The one point tying all this to computer chess (and one I made persistently), is >that reportedly Shredder ran on an overclocked Celeron in Paderborn. Which >reminds me of a saying that a person who says that something can't be done >should at the very least not be interrupting the people who are doing it. > >PK That is true. But I can also refer you to a local reseller with a very good tech support group, which sold us some PC100 memory that later turned out to not be reliable past 77mhz (verified on their memory tester). I have seen 100 that runs at 133 just fine. I have seen 133 that won't run at 100. there is a lot of luck when you step over the threshold stamped on the chip. There is some luck even if you are under the threshold, it seems. The thing you are overlooking is that for every success story on overclocking, there are 10 failures. It is easy to overlook that. But it is reality... And the _bad_ thing is that 50% of the failures are _not_ known. People just get bad data. I can point to stories on the crafty mailing list where someone overclocked an AMD and it seemed to work fine, matched the right node counts and everything. Until playing several games in a row. Then it would go south and die. Seems that heat buildup would cause some instructions to fail which would cause crafty's engine to produce impossible (and non-reproducible) results. Cranking the clock back down to normal would fix the problem perfectly. That is ugly to think about... not knowing...
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.