Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Some benchmarks...

Author: Aaron Gordon

Date: 07:53:14 04/30/03

Go up one level in this thread


On April 30, 2003 at 09:56:43, Johan Hutting wrote:

>On April 29, 2003 at 18:04:21, Aaron Gordon wrote:
>
>>On April 29, 2003 at 17:51:19, Keith Evans wrote:
>>
>>>On April 29, 2003 at 17:45:25, Aaron Gordon wrote:
>>>
>>>>On April 29, 2003 at 15:00:51, Robert Hyatt wrote:
>>>>
>>>>>On April 29, 2003 at 14:36:39, Aaron Gordon wrote:
>>>>>
>>>>>>On April 29, 2003 at 14:20:08, Tom Kerrigan wrote:
>>>>>>
>>>>>>>On April 29, 2003 at 10:48:24, Aaron Gordon wrote:
>>>>>>>
>>>>>>>>On April 29, 2003 at 02:38:17, Tom Kerrigan wrote:
>>>>>>>>
>>>>>>>>>On April 27, 2003 at 16:32:10, Aaron Gordon wrote:
>>>>>>>>>
>>>>>>>>>>On April 27, 2003 at 14:50:27, Tom Kerrigan wrote:
>>>>>>>>>>
>>>>>>>>>>>On April 26, 2003 at 22:25:47, Aaron Gordon wrote:
>>>>>>>>>>>
>>>>>>>>>>>>On April 26, 2003 at 21:11:59, Tom Kerrigan wrote:
>>>>>>>>>>>>
>>>>>>>>>>>>>I checked Aaron's story with his contact at AMD. The guy said that AMD didn't
>>>>>>>>>>>>>allow performance testing with the memory _overclocked_, but it certainly isn't
>>>>>>>>>>>>>underclocked. This makes perfect sense to me. (If you allow overclocking memory,
>>>>>>>>>>>>>why wouldn't you also overclock the processor? Then all your benchmarks are
>>>>>>>>>>>>>worthless.)
>>>>>>>>>>>>>
>>>>>>>>>>>>>So SPEC is comparing non-overclocked Intel to non-overclocked AMD and Intel
>>>>>>>>>>>>>wins.
>>>>>>>>>>>>>
>>>>>>>>>>>>>-Tom
>>>>>>>>>>>>
>>>>>>>>>>>>When I ran the tests I recalled seeing some information where the P4 was running
>>>>>>>>>>>>CAS2 and the like. The settings I was told to use put me at CAS 2.5.
>>>>>>>>>>>
>>>>>>>>>>>It sounds like you don't really know what configs Intel uses for SPEC testing.
>>>>>>>>>>>
>>>>>>>>>>>>How would this be 'fair'? Same thing happens on some review pages, but to a much
>>>>>>>>>>>>larger degree. As I have proven in the past tomshardware has actually run the
>>>>>>>>>>>>memory lower than the bus on the athlons tested, put the AGP to 1x, etc.
>>>>>>>>>>>
>>>>>>>>>>>I think we can all agree that review pages may be biased. My point was that SPEC
>>>>>>>>>>>is not biased, because the vendors are submitting their own scores.
>>>>>>>>>>
>>>>>>>>>>I've said this many, many times already. AMD told me to run CL2.5. I've seen
>>>>>>>>>>them do the same thing for the SPEC benchmark. Try reading the lawsuit message I
>>>>>>>>>>posted here again. I'm sure they'd run the fastest timings in the bios if they
>>>>>>>>>>could. I can, and have, and don't have anything to fear from Intel.
>>>>>>>>>>
>>>>>>>>>>>>slow. I went and 'rented' one myself. I compared a few clock speeds, I'll post
>>>>>>>>>>>>what I have so far but the most for now will be just the max both systems could
>>>>>>>>>>>>do.
>>>>>>>>>>>>GCC (Linux kernel compile times)
>>>>>>>>>>>>XP-2.50GHz: 119.5 seconds
>>>>>>>>>>>>P4-3.32GHz: 126.87 seconds
>>>>>>>>>>>>Gzip:
>>>>>>>>>>>>P4-3.32GHz: 25.340 seconds
>>>>>>>>>>>>XP-2.50GHz: 26.060 seconds
>>>>>>>>>>>
>>>>>>>>>>>etc. Your gcc test shows a 41% improvement in IPC for the Athlon, vs. the 9%
>>>>>>>>>>>improvement in official SPEC submissions. You get a 29% improvement in Gzip vs.
>>>>>>>>>>>a 22% improvement. How do you explain this? You're obviously a big AMD fan, why
>>>>>>>>>>>should I think your results are somehow more accurate than results from the
>>>>>>>>>>>companies themselves?
>>>>>>>>>>>
>>>>>>>>>>>-Tom
>>>>>>>>>>
>>>>>>>>>>I'm only a fan of whats fastest. Also, if I see a good product getting reviewed
>>>>>>>>>>or tested poorly I'm going to make a comment. AMD, Intel, Cyrix/VIA, doesn't
>>>>>>>>>>matter.
>>>>>>>>>>
>>>>>>>>>>First of all, I used the fastest timings on both systems. I didn't run CL2.5 as
>>>>>>>>>>some of the SPEC systems run. I used the fastest drivers I could find on both
>>>>>>>>>>systems. The point is.. when both systems are configured so they just can't
>>>>>>>>>>possibly go ANY faster this is what you get. Believe what you want, doesn't
>>>>>>>>>>matter to me either way. I'm just reporting my test results.
>>>>>>>>>
>>>>>>>>>Can you run the same tests with slower memory settings? Do you see a 30%
>>>>>>>>>difference?
>>>>>>>>>
>>>>>>>>>-Tom
>>>>>>>>
>>>>>>>>When I was doing the Quake3 benchmarks for AMD I saw a little over 20% drop in
>>>>>>>>FPS from running the slow memory timings. This is why I was wanting them to use
>>>>>>>>the CAS-2.0, 4-bank interleave, etc settings.. because it beat the crap out of
>>>>>>>>the P4-2GHz they were testing again. With the timings at the slowest settings
>>>>>>>>the 1900+/1.6GHz lost by a few fps.
>>>>>>>>
>>>>>>>>I didn't try slower timings in the other benchmarks. I'm only interested in what
>>>>>>>>the systems could at their peak.
>>>>>>>
>>>>>>>Interested or not, this indicates that your memory timing explanation probably
>>>>>>>doesn't entirely explain the differences between your benchmarking and official
>>>>>>>SPEC submissions.
>>>>>>>
>>>>>>>-Tom
>>>>>>
>>>>>>The bit of testing I did in the past with crapped out memory timings did prove
>>>>>>that the memory settings helped. I only said I RECENTLY tested Quake3, that
>>>>>>doesn't mean I didn't do any testing at all. If I hadn't I wouldn't be making
>>>>>>such a fuss over this stuff. The ram settings DO help a lot. As I said before,
>>>>>>you're welcome to telnet into my machine and run the tests yourself.
>>>>>
>>>>>\
>>>>>However, in the case of Intel or AMD, I'd suspect that if they want to test
>>>>>using cas 2.5 memory at 133mhz, then they would be hesitant to run that memory
>>>>>at cas 2.0 even if it would run.  Since it is outside the spec provided by the
>>>>>manufacturer of the memory itself.
>>>>>
>>>>>I can't imagine a vendor wanting to publish SPEC numbers, and then have a huge
>>>>>press release 6 months later saying "vendor used unsafe memory timing to produce
>>>>>a lead in SPEC numbers..." when that unsafe timing fails for someone else.
>>>>>
>>>>>specifications are specifications.  Going beyond them invites trouble.  I'm
>>>>>running a Merc V6 outboard and I have taken the compression to extreme levels,
>>>>>as well as RPM.  And I have had to rebuild the thing at _my_ expense when it
>>>>>comes apart due to my exceeding the specs, even if it were still in the warranty
>>>>>period.  I accept that without a hassle.
>>>>>
>>>>>Just because John Tiger can run his stock merc at 7500 against an advertised
>>>>>peak rpm of 6500 does _not_ mean that someone else is going to be able to do
>>>>>it with success...
>>>>>
>>>>>Same for memory, or anything else.  I trust the manufacturers to test and decide
>>>>>on what the upper bounds are, and I live within those if I want reliability.
>>>>>For racing someone up and down the river here, I want horsepower, with
>>>>>reliability a distant second place overall.  But the average mom and pop that
>>>>>take their runabout out for a Sunday afternoon of skiing or pulling a tube
>>>>>around the lake want _reliability_.  And that is what Mercury/Evinrude/Intel/AMD
>>>>>want to provide...
>>>>>
>>>>>Us "hot rodders" want something different, but we don't necessarily have to push
>>>>>_our_ wants down "mom and pops" throats...
>>>>
>>>>I completely understand, but I'm curious as to the rating Corsair put on the ram
>>>>itself. I emailed them about it and will post the response here if I get one.
>>>
>>>Also if this is why the manager wouldn't let you run with CL=2, then I would ask
>>>him why he didn't just let you stuff in a faster DIMM. I don't think that
>>>Corsair even sells this one any more. (I may have missed it.)
>>
>>Yeah, they don't unfortunately. At the time though I didn't have any faster
>>dimms, the Corsair PC2400XMS CL2 was the fastest stuff out.
>
>http://www.specbench.org/osg/cpu2000/results/res2003q1/cpu2000-20030224-01964.html
>
>==> CL2

What about this?



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.