"Two samples from the vertical and two samples from the horizontal = 4 samples."
ok... first off, I'd like to know where they got those V5 6000 scores.. was it Voodoo extreme? if so, did they chose the exact same quake 3 options? according to Anandtech, the Ultra scores 33fps with 1024X768X32 (don't know which Quake 3 setting it is), compared to V5 6K hitting 35.7 fps using V5 6K benchmarks.
their benchmarks might have been taken off of other sites.. my questions are such: what settings were used to test the NV20?
according to these benches, though, without HSR, it hits about 75 fps, at 800X600X32 with 2x2 FSAA, compared to about 55 FPS for the GTSU and about 57 FPS for the V56K (running at 4x FSAA).
and at 1024X768X32 with 2x2 FSAA the NV20 hits ~45 fps, the GTSU hits ~33, and the V56K with FSAA hits about 36 or so FPS.
EDIT: after going through the numbers of other benches, I'm wondering if I estimated these numbers accurately.. please post a response if you think I did!
the performance improvement over the GTS can thus be broken down to 136% performance of the GTSU at 1024X768X32, and 136% in 800X600X32.
in fact, the percentages I just found are EXACTLY the same. in both cases I get 1.36363636363636363636363636363636 as my number (it represents the amount you multiply your FPS number on a GTSU to find what the NV20 should hit, but only should work at extremely fillrate limited resolutions), when dividing the numbers that I mentioned above..
the chances of this happening in real life are slim to none. thus, either I have been inaccurate in reading the graph, or the values have only been calculated using a percentage value, rather then really benchmarked.
AND, when looking at Anand's GTSU review when he overclocked it, he didn't do FSAA benches, but he DID include some resolutions that CAN be compared (1600X1200X32 for example). the GTSU clocked at 285/500 core/memory hits 64 fps at 1600X1200X32, compared to the 75 accomplished by the NV20 at 800X600 with 2x2 FSAA. however these benches probably don't have much meaning.. only that even without HSR, if the NV20 DOES accomplish those scores, it means that either the RAM is running at over 500mhz, or that the core is doing more then HSR.
to add to that, when you compare the V5 6K (from Voodoo extreme benches) with 1600X1200X16 compared to 800X600X16 with 4X FSAA, you get a score that is within 2 FPS from each other. this adds to the above statement.
I'm inclined to believe however, that neither of these is true, becuase the chances of me getting that number above (136%) are so low, that I couldn't believe it at first either.. it's
conclusion? these numbers are BS. I'd like to wait for some REAL benches, before I draw any conclusions..