Fermi benchmarking controversy

[MrDeodorant]

A moderator on Overclock.net has run the Unigine benchmark at the settings used in nVidia's 'GTX480 vs 5870' slides on his 5850, and the results are surprising.

Originally Posted by IEATFISH

Uh-oh nvidia guys, my results are gonna probably make some of you flame me. I lined up the axis 0 an 80 and then stretched the sides until my peaks matches the ones on their chart (since fraps was started and ended at a different time than they did). My results show that they most likely benched a 5850 and not a 5870 as there was only a 1 or 2 FPS difference (and my system is far from optimized whereas theirs is probably a clean install). There is also some slight tilt in their chart still so that could account for the very slight differences. :eek:


edit: Or maybe my 5850 is a 5870 in disguise. :D

His results are in blue.

Source

[Jake_HT]

hmmmmmmmm

[MrDeodorant]

He's re-running it with his computer at 3.33 GHz instead of 4, and the differences are being blamed on everything from drivers to sunspots, but it's definitely thought-provoking.

[MrDeodorant]

Update:

Originally Posted by IEATFISH

Okie dokie, redid the test at 3.25 (just changed the multi, didn't want to mess with BCLK so I erred on the side of lower). The low FPS at the beginning is something I have noticed on multiple runs and not sure why that is, maybe because I it was a fresh boot and the bench wasn't in RAM yet. Anyway, the rest of it is pretty clear. I also tidied up my graph:

[Spaceman-Spiff]

Now, if only he has a 5870 too to compare.

I guess everything will be clearer by end of this month (hopefully).

[sswilson]

Yeah, I never put much faith into pre-release benches, and especially ignore them unless there are enough other sources to compare with.

[SKYMTL]

I've already got popcorn in the microwave. I'm going to sit back and watch this one play out without saying one.....bloody....thing.

[Dr_BenD_over]

Some people have way too much time on their hands. It will be out, eventually, and then the speculation can end.

[Bond007]

It must be nice to know the truth when nothing but rumours and bits and pieces of truth are floating around.

[sswilson]

Well, in fairness, the results being questioned aren't WRT the numbers being put out by the new vid card, they're about the reference #s being used. For this to get anywhere we'd need to see some #s comparing the 5850 to a 5870 using this benchmark... maybe for this particular benchmark those two cards perform similarly.