Author: Robert Hyatt
Date: 13:30:14 02/19/03
Go up one level in this thread
On February 19, 2003 at 14:42:34, Aaron Gordon wrote: >On February 19, 2003 at 12:58:10, Robert Hyatt wrote: > >>On February 19, 2003 at 11:31:49, Wayne Lowrance wrote: >> >>>On February 19, 2003 at 00:19:41, Robert Hyatt wrote: >>> >>>>On February 19, 2003 at 00:11:08, Aaron Gordon wrote: >>>> >>>>>I just downloaded Crafty 16.19 and ran a bench for you guys. No single cpu Intel >>>>>box could ever touch this without sub-zero cooling. Just plain not going to >>>>>happen. >>>>> >>>>>Crafty v16.19 >>>>> >>>>>White(1): bench >>>>>Running benchmark. . . >>>>>...... >>>>>Total nodes: 67136136 >>>>>Raw nodes per second: 1766740 >>>>>Total elapsed time: 38 >>>>>SMP time-to-ply measurement: 16.842105 >>>> >>>> >>>>This is not a great test since that is a very old version. I'm not sure how >>>>1.7M compares to version 19.3 in nps... >>>> >>>>However, while on the question, what is an XP 2.44ghz machine, since I am not >>>>an AMD expert. Overclocked? If so, I consider that a worthless number, because >>>>of obvious reasons... >>> >>>Just wondering Bob, why do you feel that way ? stability ?, ask to find out if >>>his overclocking is rock solid. I would be interested too. If not then I agree >>>it is worthless. >> >> >>Yes, stability. The problem is that something can _appear_ to be rock solid, >>but not be. >> >>For example, suppose an integer add is done wrong, once every one billion times? >> My >>hash collision experiment suggests that this might not be noticable at all, yet >>the machine >>is _still_ producing wrong answers from time to time. And logic says that this >>will >>eventually "come home to roost". >> >>Think about AMD. They are competing with Intel. Why would they sell a >>processor >>to run at XXXX mhz, when it is capable of running at XXXX+N mhz? The answer is >>that they would not. They would rather run faster and increase their >>performance edge >>over their competitor. But the engineers have found that XXXX works in 100% of >>the >>cases, where anything greater slowly increases the chance for errors. >> >>Overclocking is like juggling hand grenades with loose pins. Eventually >>something bad >>is going to happen, although it might take a long time. >> >>When you overclock, you reduce the ability of the processor to operate correctly >>if there >>is any variance in the supply voltage, any variance in the temperature, any >>variance in the >>outside EMF interference, etc. > >There are tests you can run to guarantee stability in overclocked situations. Sure there are. And they take _years_ to run. You have to run all four billion possible values thru every possible instruction, in every possible sequence, with every possible pipeline delay, at every possible temperature, with every possible voltage variance with... You get the idea... >Two main tests are memtest86 which is excellent. Makes sure your memory, L1 & L2 >caches aren't producing errors. Also Prime95 is extremely (probably an >understatement) sensitive. If you have any errors or any remote possibility of >an error then yes, it'll show it. > >Here is one example.. my old Celeron-2 566 @ 1202MHz errored instantly in >prime95. This shows there WAS a problem. I never crashed in anything else or was >unstable at all.. but it did show there was a problem. Now, with my XP I can run >2466MHz 100% stable without any errors of any kind at 1.85v. > >Also Hyatt, AMD can run a little bit faster. Right now they're producing >2.3-2.6Ghz AthlonXP chips. If you get the latest AIUHB stepping (anything from >november 2002 on up to present) you will get at least 2.3ghz. This means they're >consistantly producing those 2.3GHz chips. I figure they can continue to sit >back and take it easy while they produce the barton chips, though. They can tack >on a higher "PR" rating due to the extra cache.. better for them to not have to >crank the clock speed up as much. They could produce a 3200/3300+ Barton running >2.25-2.3GHz, but why would they want to? Answer: they want _any_ kind of performance edge they can get. But, the caveat is, they want _reliability_ as well... > It competes just fine at 2.16ghz/512k >L2/166fsb(333DDR). The more chips they release means they've got to lower the >price on the other chips and take a hit in profits which they would otherwise >have kept if they would have kept the slower chips out for a longer period of >time. > >I'm sure Intel could have sold the P4-3.06 Xeon 2 months ago.. why didn't they? >So people like you would buy them all up. Now that they've got plenty of money >from the 2.8's, here comes the 3.06's :) Possibly, but that's "business". But they weren't producing 2.8's that could run reliably at 3.06, which is the topic here...
This page took 0.03 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.