A Little Interesting 3D Mark 06 Test for You

[SKYMTL]

All right, I have been wondering this out loud for some time but I may as well post it in words right here.

Those of you who have switched from Nvidia to ATI GPUs: have you seen your 3DMark06 CPU score INCREASE?

The same goes for those of you who have gone from Red to Green. Have you seen your CPU score DECREASE?

I can't seem to find many threads about this so I thought I would open a discussion right here about it.

This comes directly from some of my GPU reviews (overclocking section) over the last while from a number of different systems and detail the CPU scores ONLY:


i7 920 @ 4.01Ghz w/HT

HD 4770: 6941
HD 4890: 6962

GTX 275: 6744
GTS 250 1GB: 6790 (upcoming review)


QX9770 @ 3.85Ghz

GTX 285 OCX: 5569
GTX 285 SSC: 5533
GTX 260 216: 5526

HD 4870 Sonic: 5724
HD 4870 512MB: 5686


I know this isn't a scientific test or anything but a difference of a pretty consistent 200 points is something to chew on.

Just wanted to bring it up as food for thought. Let's call it SKYMTL's Midnight Ramblings.

[cadaveca]

the nV tech you listed is four seperate processing cores, while ATI is wholely unified, so nV takes a bit more @ driver-level. It's even more obvious in true DX10, iMHO. Although the cpu tests are graphics drawn on the cpu, they still require the vga to display them, and hence the difference.

Even down to memory control, nV's current gpu line-up requires more driver-level management, and thier performance crown only speaks volumes of driver development, and 3rd-party developer interactions.

As I've said before, under deep analysis of the gpu business, nV lags behind in tech, but bolsters that with support that currently cannot be matched, and really, with all this renaming, places higher value on thier intellectual property, as they continue to strive to make it relevant.

[Squeetard]

Make it Scientific. Start with a fresh build, nvidia card, bench it and then swap to ati. Reverse it, fresh build, start with ati, bench then swap to nvidia. Post results.

[SKYMTL]

All of our GPU testing is done on completely fresh installs of Vista every single time.

Cads, that totally makes sense especially when you see the driver rather than hardware level performance improvements Nvidia has been able to eke out with its current generation.

[cadaveca]

Yeah, well, we both know that '06 cpu scores change from month to month as drivers change, and that the '06 cpu test was suspect when '06 was first released, and the cpu tests were capped at 3FPS. That cap has since been removed, but it speaks volumes as to how vga-dependant the cpu tests are.

'06 was relevant when it came out, but with every successive release of a new gpu, it's importance becomes more and more in question. Really, i think when we saw 10,000-point scores, it became fairly useless, and almost anything over 17,500k is cpu-limted.

Vantage is a bit better in this regard, giving consistent results regardless of config, but it's still a bit suspect itself, as the test scores are easily skewed by system bandwidth bottlenecks in a few tests.


I don't really like nV, because I love innovation, but I've realized that thier innovation lies on the programming front, while ATI lands squarely in the hardware innovation stage. these seperate dierctions is what sets the two companies apart, and makes them completely different. There's something to say about making your products relevant to the marketplace, vs making the marketplace relevant to your tech...



Now, we can dig further into this, and cover OS installs that are NOTfresh, and how this affects performance, but the fact you do a fresh install hints that you've seen the sae issues that many report in this instance as well.

[SKYMTL]

It isn't necessarily issues with performance. I do it to avoid the problems that sometimes rear their heads after installing and uninstalling 4-8 GPUs over the course of a few days.

[cadaveca]

Same difference, to me at least. I'm not saying there's any intentional gremlins here, just that installing multiple parts can cause issues...not just vgas. You always get optimal performance on a fresh install.

In the end, the difference reported in '06 cpu score...does it really affect anything?


I think that as reviewers, and having to keep current, we can be less exposed to issue that may surface from things like this...for example, is this difference due to how much cpu cache the driver uses, or is it due to cpu-vga communication? If it is due to cache use, those with lesser cpus are going to see a larger impact than those with "performance" parts.


Also, how does this change when multiple cards are installed, ie, Crossfire and Sli?

[Shadowmeph]

HD 4770: 6941

isn't that a vidcard that is around 150.00?

[Killswitch]

Originally Posted by Shadowmeph

isn't that a vidcard that is around 150.00?

Yes but that score is based on the system's CPU. Video card power has nothing to do with that number. At least it's not supposed to.

[crazyhorsejohnny]

Very good read.