XP, 754 and 939 CPU benchmark comparisons

[deeppow]

Have been accumulating testing results for my rigs over the last few years and put together a comparison just for the fun of it. Plus I was curious.

For those that are curious about such a comparison too, here is a link to the html page containing the results summary. I realize that this isn't a perfect study but hopefully provides a flavor of the differences.

[largon]

What OS' were used?

[deeppow]

WinXP

[largon]

The performance gain that is practically possible with a change from a 32-bit system to a 64-bit system can be reasonably significant.

This may lead to a misinterpretation:
One might think A64 acts always as a 64 bit processor just because it's capable of 64 bit calculations. But ofcourse, a 64 bit OS (which normal WinXP isn't) is needed to take advantage of 64 bit a proc.

[don_vercetti]

hmm might i suggest you clock them all at the same speed? That would highlight architectural differences, at least. If they could all have the same RAM timings as well, that would also be good.

Comparing a 2.2ghz 1mb cpu to a 1.8ghz 512k cpu (3700 vs. 2500) is never going to be totally conclusive, really. But a 1.8ghz barton vs. 1.8ghz sandy vs. 1.8ghz newcastle would be far more interesting.

[deeppow]

This may lead to a misinterpretation:
One might think A64 acts always as a 64 bit processor just because it's capable of 64 bit calculations. But ofcourse, a 64 bit OS (which normal WinXP isn't) is needed to take advantage of 64 bit a proc.

No debate there. However the objective was to test within as similar conditions as possible and I can't test using 64-bit OS/benchmark on a 32-bit processor. Thus I ran WinXP on all platforms. I would also content that this is also a common OS for many folks today and a good configuration to test. As more 64-bit software becomes available in the future that will of course change.

You do raise an interest point and while I have only very preliminary results to date using the 64-bit version of Prime95 on Win64Pro I find that that benchmark produces the same results (within the uncertainty of both sets of results). This conclusion will be application dependent and a result of the talent of the programers and needs of the particular calculations. For example, highly nonlinear scientific calculations typically needing double precision on a 32-bit OS will benefit significantly from the 64-bit hardware and software.

Thanks for your comment.

Comparing a 2.2ghz 1mb cpu to a 1.8ghz 512k cpu (3700 vs. 2500) is never going to be totally conclusive, really. But a 1.8ghz barton vs. 1.8ghz sandy vs. 1.8ghz newcastle would be far more interesting.

I'm sorry but you completely lost me there don_vercetti. The abscissa (x-axis) of the plots allows you to do just what you asked I think. Pick a fixed CPU speed and read the corresponding values for the respect CPU above and off the y-axis.

If this isn't the case maybe you can clarify.