Actually most reviewers use 3DMark less and less due to the obvious problems with it, and they certainly don't bother with all of the individual synthetic results like TR does. We've already talked about a few glaring examples where the results directly contradict your arguments in real-world games, like multi-texture fillrate with G80 and G92, yet the G80 still matches G92 in games. Or G92 vs. G94, where 9600GT still matches G92 in games. Or better yet, how the R600 from ATI wins in 3DMark but still can't beat any G80/G92 parts in actual games.
I think you need to review 9600GT reviews. It keeps up with 8800GT (and even G80/9800) up until higher resolutions or with AA enabled. Bandwidth isn't a factor compared to 8800GT because it has the same 256-bit bus and similarly clocked RAM at 900-1000MHz.....and it manages all this despite having half the SP and Texturing ability of 8800GT. Wonder why?
I think you said it because you have no clue what you're arguing about. Once again, do you think number of memory controllers/bus width has an impact on performance when total bandwidth is not an issue? Yes or no.
Rofl ya, except the other cards are running the same settings and in the case of QW, effects were disabled specifically because Radeon HD parts couldn't use them. But that doesn't change the fact you can't read benchmarks or make sense of anything more than synthetic 3DMark results. Also, the effects turned off in FEAR and QW, like soft shadows and particles have nothing to do with your arguments about performance with texturing and bandwidth, as those features in games stress the ROPs and shaders more than anything. They also happen to be some of the most expensive features to enable in games, with modern GPUs handling large textures and filtering gracefully. But if you actually spent time playing games rather than talking about them, you'd know turning up texture and filtering quality is much less performance expensive than any shadowing, particle or post-processing effects.
Uh, so what's a 9800GTX? Its an overclocked 8800GTS with a few power tweaks and slightly faster RAM. In this case the Gainward uses the same .8ns Samsung so for all intents and purposes the cards are identical. And no its not any real surprise the Gainward G92 wins many tests as its only clocked 155MHz faster and closer to that 756MHz number I quoted you earlier. Oh ya, that's where its fillrate would begin to match the 33% difference in ROPs.
Uber high resolutions like Crysis at 1280 with Medium settings? Weren't you knocking another review site for using similar setttings? But I guess its OK when TR does it right, just as long as they include detailed 3DMark results. :laugh:
The Ultra is an overclocked 8800GTX, plain and simple. It has an updated cooler and a few power tweaks but any of the OC 8800GTX will perform identically to it clock for clock, as shown in numerous reviews. G80 is clearly the faster chip clock for clock, so I really have no clue what you're talking about. It'll be pretty obvious once GT200 rolls out with similar clock speeds to G92, only with more ROPs to give it the boost in performance lacking with G92.
It beats 8800GTX in some benchmarks and loses in others. That's with a 100MHz core speed increase to help close the gap in fillrate. G92 certainly benefits from some of its other enhancements, but overall it still can't beat the Ultra where it only has a 50MHz lead. So once again, clock for clock, G80 is clearly far superior to G92 and that's due to its ROPs more than anything else.
LMAO! More proof you can't read benchmarks, or comprehend anything more than what you see fit.
Of course texturing ability would be included with increase to core clocks, but that's not an issue when comparing to G92 since it already has the advantage over G80 with improved 1:1 TMUs. But we see with G92 that even with improved texturing units, it still can't surpass G80 without extreme core clock increases, and certainly doesn't scale nearly as well as G80. Why? Because it has 1/3rd fewer ROPs. Also notice his results mirror what I've said earlier...that performance with G80 scales linearly at nearly 1:1. You can't say the same for G92 because again, you need much faster core clocks to make up for the 1/3rd fewer ROPs which have the biggest impact on performance.
No need to talk about my fantasies, I'd be content if you properly comprehended the benchmarks and arguments you present. As for the last comment about GDDR5, that was the point in showing 9800GTX compared to G92 8800GTS, that a 270MHz increase in bandwidth isn't what G92 needs the most. Don't believe me? Go check out some of the 9800GTX OC Edition Reviews (756-770MHz core, 2300MHz memory). You'll find once again what I've been saying to be true. G92 needs fillrate more than it needs bandwidth (or anything else), which it gets with extreme core clock increases.