The Intel 486SX 25MHz processor has been given the iCOMP rating of 100. All other processors have been assigned a number which is their performance relative to this base processor.
<hr></blockquote>
So basically the majority of this benchmark consists of the synthetic Ziff Davis and an old, outdated version of Spec.
It leads me to believe the binaries for this thing are very old.
Any amount of accuracy that this thing achieves may be completely incidental. There are a lot better benchmarks out there than this IMO: Arstechnica's Testbench, QuakeIII, Photoshop, Seti, Lightwave, ect.
<strong>I don't particularily care for CPU scorecard as it utilizes the iComp benchmark suite:
So basically the majority of this benchmark consists of the synthetic Ziff Davis and an old, outdated version of Spec.
It leads me to believe the binaries for this thing are very old.
Any amount of accuracy that this thing achieves may be completely incidental. There are a lot better benchmarks out there than this IMO: Arstechnica's Testbench, QuakeIII, Photoshop, Seti, Lightwave, ect.</strong><hr></blockquote>
You seem to be under the impression that there should be one tell-all benchmark to gauge processor performance. This is what synthetic benchmarks try to achieve and most of them fail for a simple reason.
The simple fact is that different applications exibit a different code behavior. If your looking for a processor that excells in server applications, why judge it based on CAD scores?
(note: I didn't say I actually liked all those benchmarks I mentioned, simply that they were better than the iComp)
When you run a benchmark, you must know what the limits and caveats are in order to draw pertainent conclusions.
Here's what I think about the various benchmarks out there:
QuakeIII:
Used to be a real bad crossplatform benchmard before OSX as OPengl on system 9.x.x wasn't that great. Now I think it's a decent measure of FPU/System bandwidth capability (at low res). The only problem I find is that it may well stress Graphic Accelerator driver quality even down at low resolutions.
Photoshop:
This used to be a good Int-scalar benchmark back in the day, but with SSE/2 and Altivec extentions it's more of a measure of SIMD-Int capability. It also includes filters that stress the FPU and FPU-SIMD (like the lighting effects and such). But being that it includes a large number of filters that behave differently, many different conclusions have been drawn with this application. You pretty much have to draw conclusions from each different filter individually as it's easy to be selecive.
Seti/RC5:
I really don't like these too much. IIRC they use fairly esoteric bitfield manipulations that don't come up in critical situations in most other applications. The PPC does well in them due to special hardware implementation to speed this up like a barrel shifter. The P4, for example lacks a barrel shifter.
Lightwave/Cinbench:
These tend to flex Scalar FPU/ Multiprocessing capabilities: the Cinbench raytrace is a nice little app you can download from <a href="http://www.maxon.de"; target="_blank">www.maxon.de</a> and run. It tends not to flex memory or system bandwith much, but rather makes a nice FPU test. Lightwave raytrace is the same way and you can find results at <a href="http://www.blanos.com"; target="_blank">www.blanos.com</a> . More complex scenery will tend start to stress memory bandwith more.
Arstestbench:
It's a little benchmark that was put together over in the Ars battlefront. It was spearheaded as a bipartisan effort and I think it is usefull for telling a few things.
personally I think it shows off the ability of a processor to reorder and schedule common quality code (VC++ 6 and Codewarrior binaires without a lot of optomizations) in addition to showing off it's cache, ALU, and FPU prowess.
Comments
<a href="http://www.cpuscorecard.com/"; target="_blank">http://www.cpuscorecard.com/</a>;
[quote]The iCOMP value is the weighted geometric mean of four benhcmarks;
•ZD(Ziff-Davis) Bench - 68% •16-bit Whetstone - 2% •SPECint92 - 25% •SPECfp92 - 5%
The Intel 486SX 25MHz processor has been given the iCOMP rating of 100. All other processors have been assigned a number which is their performance relative to this base processor.
<hr></blockquote>
So basically the majority of this benchmark consists of the synthetic Ziff Davis and an old, outdated version of Spec.
It leads me to believe the binaries for this thing are very old.
Any amount of accuracy that this thing achieves may be completely incidental. There are a lot better benchmarks out there than this IMO: Arstechnica's Testbench, QuakeIII, Photoshop, Seti, Lightwave, ect.
<strong>I don't particularily care for CPU scorecard as it utilizes the iComp benchmark suite:
So basically the majority of this benchmark consists of the synthetic Ziff Davis and an old, outdated version of Spec.
It leads me to believe the binaries for this thing are very old.
Any amount of accuracy that this thing achieves may be completely incidental. There are a lot better benchmarks out there than this IMO: Arstechnica's Testbench, QuakeIII, Photoshop, Seti, Lightwave, ect.</strong><hr></blockquote>
how are those accurate benchmarks?
Quake III= PC lead
Photoshop= Mac lead
SETI= Mac lead
Lightwave= PC lead
that's accurate?
<hr></blockquote>
You seem to be under the impression that there should be one tell-all benchmark to gauge processor performance. This is what synthetic benchmarks try to achieve and most of them fail for a simple reason.
The simple fact is that different applications exibit a different code behavior. If your looking for a processor that excells in server applications, why judge it based on CAD scores?
(note: I didn't say I actually liked all those benchmarks I mentioned, simply that they were better than the iComp)
When you run a benchmark, you must know what the limits and caveats are in order to draw pertainent conclusions.
Here's what I think about the various benchmarks out there:
QuakeIII:
Used to be a real bad crossplatform benchmard before OSX as OPengl on system 9.x.x wasn't that great. Now I think it's a decent measure of FPU/System bandwidth capability (at low res). The only problem I find is that it may well stress Graphic Accelerator driver quality even down at low resolutions.
Photoshop:
This used to be a good Int-scalar benchmark back in the day, but with SSE/2 and Altivec extentions it's more of a measure of SIMD-Int capability. It also includes filters that stress the FPU and FPU-SIMD (like the lighting effects and such). But being that it includes a large number of filters that behave differently, many different conclusions have been drawn with this application. You pretty much have to draw conclusions from each different filter individually as it's easy to be selecive.
Seti/RC5:
I really don't like these too much. IIRC they use fairly esoteric bitfield manipulations that don't come up in critical situations in most other applications. The PPC does well in them due to special hardware implementation to speed this up like a barrel shifter. The P4, for example lacks a barrel shifter.
Lightwave/Cinbench:
These tend to flex Scalar FPU/ Multiprocessing capabilities: the Cinbench raytrace is a nice little app you can download from <a href="http://www.maxon.de"; target="_blank">www.maxon.de</a> and run. It tends not to flex memory or system bandwith much, but rather makes a nice FPU test. Lightwave raytrace is the same way and you can find results at <a href="http://www.blanos.com"; target="_blank">www.blanos.com</a> . More complex scenery will tend start to stress memory bandwith more.
Arstestbench:
It's a little benchmark that was put together over in the Ars battlefront. It was spearheaded as a bipartisan effort and I think it is usefull for telling a few things.
personally I think it shows off the ability of a processor to reorder and schedule common quality code (VC++ 6 and Codewarrior binaires without a lot of optomizations) in addition to showing off it's cache, ALU, and FPU prowess.
[ 12-15-2001: Message edited by: JFW ]</p>