P
Pr()ZaC
Guest
http://www.gamersdepot.com/hardware/video_cards/ati_vs_nvidia/dx9_desktop/001.htm
Gamers Depot:
I don’t know how anyone could objectively look at the performance we’ve seen in these benchmarks and conclude that a 5900 Ultra is a smarter buying decision than a 9800 Pro – even the 128MB 9800 Pro (as used in the tests here) trumps the lofty 256MB 5900 Ultra. If you’re still “stuck in the past” and think that ATI is plagued with driver issues, than go ahead and keep your head stuck in the sand like an ostrich, buy a 5900 Ultra and then start crying when your pals are smoking your ass in games like Half Life 2 and Halo because they’re running ATI hardware.
And ATi's Patti Mikula, Public Relations Manger, Desktop Products wrote:
"The R300 architecture was built from the ground up for performance in DirectX 9. DX9 instructions map naturally to our hardware, without any tweaking or driver optimizations. This is important, because the vast majority of games out there won't have the benefit of driver optimizations from anyone (unlike certain game benchmarks), because no-one has the engineering resources to spare. Whether you look at brute force (our 8 pipes versus the competitor's 4) or elegance (we can run many shader operations in parallel that our competitor can't) we have a fundamental advantage with our hardware. With more and more DirectX 9 games coming onto the market, the battle will be all about who can run shaders faster and more efficiently. And our shader performance is hugely better. ShaderMark and other tests show shader performance that is three to six times better on ATI's hardware than Nvidia's. This architectural advantage is evident in shipping games like Tomb Raider: Angel of Darkness. You will also see this in Half-Life 2 and every DX9 game coming out before the holiday season."
And John Carmack (Doom3)
"Yes. NV30 class hardware can run the ARB2 path that uses ARB_fragment_program, but it is very slow, which is why I have a separate NV30 back end that uses NV_fragment_program to specify most of the operations as 12 or 16 bit instead of 32 bit."