What do you expect for R650

What do you expect for HD Radeon X2950XTX

  • Faster then G80Ultra about 25-35% percent overall

    Votes: 23 16.4%
  • Faster then G80Ultra about 15-20% percent overall

    Votes: 18 12.9%
  • Faster then G80Ultra about 5-10% percent overall

    Votes: 18 12.9%
  • About same as G80Ultra

    Votes: 16 11.4%
  • Slower thenll G80Ultra about 5-10% percent overall

    Votes: 10 7.1%
  • Slower then G80Ultra about 15-25% percent overall

    Votes: 9 6.4%
  • I cannot guess right now

    Votes: 46 32.9%

  • Total voters
    140

Shtal

Veteran
What do you expect for R650 ---->R680

What and how could ATI/AMD improve R650 ----->R680 core, since it will be shrink at 65nm ----> 55nm tech, which means it will have more space to boost performance.

A. higher clock frequency for Stream Processing Units "example 900MHz"
B. higher clock frequency for GPU itself "example 900MHz"
C. Increase texture units count "example 32TMU's"
D. Increase ROP count maybe to 24ROP "depends by how much it will reflect performance"

What ATI/AMD can do @ 65nm ---->55nm shrink: What do you expect out of ATI HD Radeon 2950XTX.

Edit: Increase Stream Processing Units count maybe depends by how much space it could fit over for 65nm ---->55nm shrink"
 
Last edited by a moderator:
TEXTURES TEXTURES TEXTURES

Oh and more shaders if possible. Gotta kreep up with the Joneses.

When is R650 due? And will it be just a clock bump or possible actual redesign (lets hope the latter, since R600..erm, is not the greatest).
 
TEXTURES TEXTURES TEXTURES

Oh and more shaders if possible. Gotta kreep up with the Joneses.

When is R650 due? And will it be just a clock bump or possible actual redesign (lets hope the latter, since R600..erm, is not the greatest).

Ditto to all of that.

Though I would like to think that if you only compare R600 to the 8800GTS, it's actually pretty 'great' considering the launch price and availability (for the first time EVER, I can actually order one from a nearby store. That's never happened for me). I've had a good think about it and i'm going to just get one R600 for now...
 
TMU's and clock speed (which, of course, affects both shaders and the rest of the core unless they implimented clock domains, which won't have happened for a refresh). I'd say fat chance on the first, but the second is pretty much a given.
 
I didn't vote because I don't think the vote options are particularly useful.

And unless we have quantifiable answers for A, B, C & D in the OP, they're not useful options either.

IOW, I honestly don't know what to make of this thread or have any idea what this thread really serves.

Or if I agree with this thread being in this 3DT&A forum at all.

:devilish: :smile:
 
TMU's and clock speed (which, of course, affects both shaders and the rest of the core unless they implimented clock domains, which won't have happened for a refresh). I'd say fat chance on the first, but the second is pretty much a given.

They did rejigger R520>R580 though, which is rare.
 
You forgot one option: "there will be no R650", which is what I expect. I expect it skipped, they should go for the R700 right away or they'll fall further behind.
 
You forgot one option: "there will be no R650", which is what I expect. I expect it skipped, they should go for the R700 right away or they'll fall further behind.

Considering that realistic timeframe for R700 is fall of 2008, how can they skip the refresh?
 
You forgot one option: "there will be no R650", which is what I expect. I expect it skipped, they should go for the R700 right away or they'll fall further behind.
Or perhaps simply an R600 with twice the TMUs/ROPs count and a double pumped ALUs core?

As it is now, R600 isn't that bad and that would double raw power while not increasing transistor count too much.

We seriously need to wait until the end of June in order to see the first acceptable drivers, then we will be able to see what's really wrong.


Considering that realistic timeframe for R700 is fall of 2008, how can they skip the refresh?
By investing more in that project?
 
Doubling the texture sampling capacity would be good enough, for the AF performance sake. No need to touch the # of TMU at all.
As for the ROPs (eg. RBE) - going any different odd count will "destroy" the pipeline tiling orthogonality of the current marchitecture, thus involving much kind of redesigning large parts of the chip. Maybe putting them in a differential clock domain will do the job -- plenty of bandwidth there, anyway.
Double-clocking such a beefy chunk as the R600 shader core, IMHO, isn't going to be the same thing as for G80 skinny SPs. My be extra buffering, if the shrinking process allows it as a free die space bonus.
 
I expect a simple clock speed boost because of the reduced leakage by moving to 65nm, perhaps allied to GDDR4 too
 
Back
Top