No Shader 3.0 for next ATI chip

[Topha]

It is in German, so you may need a translator: here

Says ATI decided that Performance was more important than Shader 3.0 support, which no one is gonna need for the next few years anyway (except maybe 3dMark 04 ), so they improved that instead.

Pretty brave considering nVidia is definitley adding shader 3 to their product.

Lets see who will come up with the better chip

[borat]

smeels like a matrox esque decision, ohh wait there not quite as M decided that neither features or performance was important

[Chrono_Wanderer]

M believed banding quality is important

actually its not matrox, its dan woods

[Helevitia]

I wouldn't believe that article. ATI hasn't said anything yet. Time will tell.

Dave

[Indiana]

Originally posted by Helevitia
I wouldn't believe that article. ATI hasn't said anything yet. Time will tell.

Dave

I do believe the article, since most rumors say the R420 is an optimized and maximally streamlined R300/350 alike design. And it would be quite hard to add PS3.0 to ATIs R300 design without designing a GPU totally new from the scratch - the R500 will bring that.

(And btw. I'm quite sure that the R500 will be there before we see the first PS3.0 game title so I can understand ATIs decision. Still it likely will hurt them with so many THG readers out there...)

[Ali]

From what Ive picked up looking around, the R420 is probably going to be fully VS3.0 and only partially PS3.0. They will have the increased registers and instruction count, but wont have conditional branching. It turned out to require to many transistors for this generation, so they put it off till the R500.

Also remember the origonal R400 got bumped up to R500 and the R420 is a relatively new beast, being based on the R300 but with some of the old R400 (now R500) tech in it.

I think what we will see is a VERY fast DX9 part, but when V3.0 Pixel Shaders are called, the drivers will convert the branches into parallel instructions for each possible outcome. This will make PS3.0 very very slow, but as it probably wont be needed, then it doesnt realy matter at this stage.

ATI would be silly not to support V3.0 in any way shape or form, because then all developers who want to use it will have to use nVidia cards for development, and if in the future ATI release a 3.0 card, they will have no end of trouble getting all the apps patched to work correctly.

Then again, if ATI ignore V3.0 and go to V4.0 for DX10 in Longhorn, what developer in their right minds will design around V3.0 if only nVidia cards can run it, and only their top end cards can do it at a decent speed?

Personally I would like to see V3.0 supported, but would understand if it isnt.

Ali

[Marshmallowman]

Since MS is not going to be releasing a new directx for quite a while this is probably a good move by ATI.

It'll be a while before Game programmers max ouit PS 2.0 stuff.

[Chrono_Wanderer]

I'd think NV40 will have full PS3.0 and VS3.0 support, given they have to redesign the chip anyhow as NV3X is no good as a gaming architecture (too programmable, too complex).

NV40 probably won't be as programmable, but I'd think its very gaming performance-oriented.

I think NV4X will be a very strong part, after all, they are NVIDIA, they have a huge R&D team, and i think they learn a hard lesson from NV3X.

As for ATI, any chances they will adopt the FP32 format?

[Indiana]

Originally posted by Chrono_Wanderer

As for ATI, any chances they will adopt the FP32 format?

That seems quite sure from what I've heard..

[Ali]

It would seem a little silly. If ATI go FP32, then they would either have to drop their FP24 to FP16 to get any benefit from _pp hints, or have no way to use the FP32 from a programming point of view.

If you use no hint, then full percision is used, thats defined as FP24. If you add a _pp to your instruction, you get partial percision, which can be anything between 24 bit down to 16bit minimum.

Therefore, they either have two full percision modes, and have to decide themselves in their drivers what one to use for different situations, or they drop FP24 and go the nVidia way, which would seem odd.

I suppose they could have FP24 and FP32, but only use FP32 in their FireGL cards, but that would be a waste of transistors for their consumer cards.

There is no reason to go to FP32 in a consumer card at this stage, as you could never see any difference. Look at the quality of NV3x vs R3xx at the moment.

Ali

[leech]

I think the real question is not wether or not they'll support PS3.0, which as everyone says, won't be useful 'til a few years down the road..... What I want is a triple-head card with more performance, and better linux drivers. I love my Parhelia, but I'd love a faster, more feature filled card for both windows AND linux. Unfortunately I dont' think ATI or nVidia are going to be able to fulfill the wish for a triple-head card any time soon....

And I doubt Matrox will release another card that anyone who games will consider buying after their actions with the Parhelia. Nor will anyone who runs linux. I think they could save a lot of face if they opened up a lot of the specs, so at least there could be a mostly open source driver. I can understand them keeping somethings closed, like their FAA alglorithms....

Enough ranting...

Leech

[Chrono_Wanderer]

hmm... leech, don't count on ATI / NVIDIA to bring triple head within these 4 years. I'd think monitors aren't arn't cheap enough until some point in the future...

[knirfie]

i think it's also weird that the R420 will supposedly be ati's last AGP chip.

[Topha]

Originally posted by knirfie
i think it's also weird that the R420 will supposedly be ati's last AGP chip.

Individual manufacturers (ASUS, Sapphire) still have the option of adding an AGP bridge to their cards

[Byock]

.

Enough ranting...

Bah! hook one of the 21" monitors to a second machine, and get a 9800Pro. You know you want one!