Originally posted by: dreddfunk
Honestly, I think the mistake here is to even use Oblivion in a set of early-released benchmarks. Certainly it's a hot game--and the one in which I'm most interested--but it's also about as un-benchmark-friendly as any game I can recall. There isn't a benchmark suite and the differences in FPS can swing much more wildly than in any game that I can recall. If you bench just running around the territory outside of the starting sewer exit, you'll get pretty high numbers. Heck, my x850xt doesn't even break a sweat there. As you get deeper into the forest with all of its foliage, next to an Oblivion gate with all of its effects, then framerates seriously, seriously plummet. AT's original article has a good page on the difficulties in benching Oblivion:
http://www.anandtech.com/video/showdoc.aspx?i=2746
If I recall correctly, many of AT's early Oblivion Benchmarks used a special save game that allowed them to run through a particular Oblivion gate in the middle of a high-foliage forest. Most of the AT benches I've seen are under these conditions. Indeed, under these conditions, the x1950xt was only pulling in the lower twenties on 19x12, I think, and the GTS was in the mid thirties, while the GTX was in the upper forties.
Here's a link:
http://www.anandtech.com/video/showdoc.aspx?i=2926&p=5 (it's the GTS 320MB review).
In that review, the GTX was pulling 50+fps at 16x12. I can see it pulling 60+ easily at 12x10. But it's unclear whether or not that was using any type of AA. I sincerely doubt it, as most of AT's Oblivion benchmarks eschewed AA because of difficulties with all of the tested parts doing AA and HDR.
Now, I can actually see how a rushed benchmark using the area right around the sewer exit would be drastically different than these (and other) benchmarks. That just means, however, that, IMHO, they never should have used Oblivion in such an early release of performance numbers--it just adds too much uncertainty to an already inconclusive process.
Thus, I'd make two suggestions to DT/AT:
1) When releasing early numbers, please stick to games that are established as being dependable when benchmarking!
2) Make sure that all relevant settings are immediately available so that the apples to oranges comparisons are obvious, even if they can't be directly compared.
That said, I'm still disappointed with the results. If R600 turns out to be a fantastic DX10 part, then at least ATI made a bad bet on when DX10 games would be available in relation to their R600 development cycle.
If this isn't the case, then they've had their own FX-style fiasco.
Now if nVidia would just release an 8800GS already I'd open my wallet faster than Doc Holiday drew his gun!