ATI R420: Two Times Faster Than The RADEON 9800 PRO?

[lukar]

r420 will have internal cashe mem
16x1 design
500 GPU
700 GDDR
Near 200mil transistors
Full DX 9.1
Full hardware supported Trueform 3.0
and so on

I posted this info a long time ago.

[prodikal]

Originally posted by lukar
r420 will have internal cashe mem
16x1 design
500 GPU
700 GDDR
Near 200mil transistors
Full DX 9.1
Full hardware supported Trueform 3.0
and so on

I posted this info a long time ago.


well **** if lukar says it, its got to be true. :rolleyes:

[StealthHawk]

Originally posted by Sazar
ati does not have the pressure on it we would have expected from nvidia and therefore is able to work its books into a state of balance and go into the black in terms of profit/loss...

Well, the perception exists for many people that NVIDIA is king again with NV35 by a good margin.

At the end of the day, the truth really is unimportant, it is all about image. It's one of the reasons we haven't seen ATI cut into NVIDIA's marketshare although they had a superior product out for some 8 months. NVIDIA has only recently got back into the game with a competitive product. Now that they actually have one, and one that is available, I question whether ATI will continue to have such "good" fortune. I also wonder how much profit ATI got from the GameCube deal, did their desktop Radeon graphics cards get ATI into the black by themselves?

[reever2]

Originally posted by Paul
The difference is that ATi have further to go than nV, in terms of setting that technology up.


Yeah, but Ati can only go half-way in setting up the new architecture, like NV did with the NV30, and can still hold the performance crown. Unlike Nvidia Ati doesnt HAVE to set up a completely new core, they still have 0.13 up their sleeve which can give them a large performance boost, and along with some minor tweaks, improved AA/AF methods, or some new technology like nvidias HCT and ultrashadow, can still hold the performance crown. Whereas Nvidia, needs to almost redesign their NV30 core to get out all of the nasty bugs and get all of the performance out of it and stop wasting silicon with stupid wastes of space like the partial precision

[Nutty]

Partial precision isn't a waste of space, its a bloody good idea. It gives you greatly increased precision over the GF3/4/8500 generation, at half the cost of FP32.

Why dont games programmer program all their graphics using 64bit doubles?? Because you dont need to, and its faster to use 32bit floats. Alot faster!

Same reason with regards partial precision in GPU's.

[Hanners]

Originally posted by Paul
And nVidia aren't going with 4x2 again. That much is for certain.

I wouldn't be so sure, from what I've heard nVidia may well be sticking with the same 8x1/4x2 design NV3x has.

[Uttar]

Nutty: Actually, there's a slight difference between FP64 in CPUs and FP32 in GPUs.

In the NV3x, it's really FP32 units which some MAJOR register usage problems. In CPUs, it's FP32 units with no register usage problems, and which can do 64-bit at half-speed.

FP32 isn't twice as slow on the NV3x - the hit greatly varies ( there are some detailed numbers at B3D, search for posts by thepkrl ) - I estimate it's 60% of the speed of FP16... Just an estimate of course.
And, well, that number can, and will, vary depending on driver releases. I've got good hopes in that area, too! I'm sure a lot of people would love to see the real potential of the NV3x :)

Hanners: Hmm, I doubt that. AFAIK, the NV40 is a "non-traditional 8x2". Note the 'non-traditional' part... Not like the NV3x was traditional anyway ;) And frankly, 8x2 or 4x4 is unlikely, hehe...

8x2/16x0 like the NV30 is 4x2/8x0 is not out of the question, but unlikely.


Uttar

[Paul]

[StealthHawk]

Originally posted by Paul


You don't say :p