NVIDIA's GeForce 6800 Ultra GPU
Quote:
"NVIDIA have surprised me with NV40. It's rediculously fast compared to R360 in some tests, if you avoid the 8X supersampling mode. But there lies the rub. Shouldn't we expect good high-end antialiasing performance with their latest generation GPU?
DX9 performance, at least theoretically, is very strong. The shading performance of NV40 is mighty in some respects. It's got all the features NV3x should have had (and R360 already does in some respects), like multiple render targets and floating point support for everything.
Overall, I'm appreciative of the features and the performance. It's nice to see NVIDIA get it more right than they have recently. Shader Model 3.0 performance will be nice to evaluate in the future.
A fast wee beasty with excellent features, we now wait for ATI and R420."
http://www.hexus.net/review.php?review=747
Resolution, AA setting choices very poor
I must say that I found the resolution and AA settings in the Hexus review to be very poor choices to demonstrate this card's true power.
First, the review itself notices how terribly inefficient the 8xFSAA setting is. As such, I find it unlikely many users will be willing to pay such a huge price for a relatively small improvement in image quality. Therefore, I think the 8xFSAA benches are of little practical value. I have no qualms about including 8xFSAA benches for comparison, but by all mean they shouldn't be the ONLY setting tested at any particular resolution.
The bigger problem is, once you discard those 8xFSAA benches, you are left with only 1024 x 768 resolution tests. This is ridiculous! Why not test in 16 color mode while you are at it? It would be fine to test 1024 x 768 all the way up to 1600 x 1200 with the SAME AA and AF settings... but to have ONLY 1024 x 768 tests at a practical AA setting is RIDICULOUS. Running the 6800 at 1024 x 768 is like taking the Concorde from Los Angeles to San Diego! You are NEVER going to stress this card in any meaningful way at this resolution.... and worse, the R360 isn't stressed either which means we are given no clear idea of the performance difference between them. Nobody is EVER going to run this card at 1024 x 768 in the real world. I really doubt 1280 x 1024 will even get much use in the real world on this card.
I know you've put a lot of hard work into this review, so I'm sorry to be so critical, but I just feel like all that had work was wasted by a couple poor choices on the test configuration. Test all resolutions at the same AA and AF settings. Or at the very least, if you are going to change the settings between resolutions, recognize settings that customers are going to find impracticle and don't make them primary data points in your graphs.
The review would make you think that...
Quote:
Originally Posted by Swafeman
its quite shocking, how in AF and AA, the 9800XT kicks its ass in most benchmarks, where as the 5950 did the same to the 9800XT
think this shows the 9800XT is better at high power than the 6800, but the 6800 really does have the crap behind it in low res no aa/af
I said it all above, but your comment highlights the poor choices in this review. The graphs in the review portray the 6800U performing badly with AA and/or high res. The fact of the matter is that the 6800U does extremely poorly with on one particular setting: 8xFSAA. That setting is an anomoly not the rule. With 8xFSAA, rather than using the typical (and efficient) multi-sample approach, the 8xFSAA does most of its work using the techincally superior but very costly super-sample techinque. In other words, at 1280 x 1024 the card is actually rendering the frame at 2560 x 2048 (!!!) then taking that frame and doing its normal 2xAA multisample AA on each pixel. Finally, it takes 4 of the pixels from the oversized scene and combines them into one pixel in the final 1280 x 1024 scene. As you might expect, though this technique produces excellent FSAA results, it is just not practical for gaming. My guess is that this techinque may have been included for 3D applications that do not depend so heavily on framerates.
The bottom line is, as the Hexus review will tell you, the 8xFSAA performance is an anomoly as far as performance. The review's own numbers show you that 4xFSAA runs 3.5 times faster (!!!!) than 8xFSAA. With that kind of performance delta, would you run your games with 8xFSAA?
The truth is, with 4xAA, high AF and high resolutions, this card smokes... often 100% or more faster instead of just 50% as indicated in this review.
Look at these benchmarks (and notice the biggest performance delta occurs at 1600x1200 which this review fails to consider, for some reason):
(3DMark03: up to 120% faster)
http://www.hardwareanalysis.com/content/article/1708.5/
(Tomb Raider: up to 112% faster)
http://www.beyond3d.com/previews/nvi...index.php?p=25
(Halo: up to 110% faster)
http://www.anandtech.com/video/showdoc.html?i=2023&p=13
(Far Cry: up to 105% faster)
http://www.tomshardware.com/graphic/...e_6800-42.html
(Splinter Cell: up to 90% faster)
http://www.firingsquad.com/hardware/...tra/page16.asp
(Serious Sam SE: up to 83% faster)
http://www.techreport.com/reviews/20.../index.x?pg=10
(Comanche 4: up to 62% faster)
http://www.hothardware.com/hh_files/S&V/nv40_debut(9).shtml