AMD: R9xx Speculation

[rpg.314]

So much for Fuads 'biggest GPU ATI ever made'. This thing will be below 380mm2 and performance close to GTX580.

The thing with trying to hold rumor sites to their word is that they always have the option of bending semantics, eg biggest gpu ever can also mean most transistors ever.

 

[Lightman]

The thing with trying to hold rumor sites to their word is that they always have the option of bending semantics, eg biggest gpu ever can also mean most transistors ever.

Or biggest AMD GPU ever You can only compare it to Barts really ...

 

[ECH]

Or maybe its Cayman Pro with only 24 of 30 SIMDs active.

I can't see why that's not plausible. Just because someone posts a pic doesn't mean it's at standard clocks.

 

[GZ007]

So much for Fuads 'biggest GPU ATI ever made'.

And then in a next article they state that antilles will pack 2 caymans.
A 200W card in dual configuration will be already a cut down or clocked down card.

 

[digitalwanderer]

I can't see why that's not plausible. Just because someone posts a pic doesn't mean it's at standard clocks.

But I saw it on the internets, it MUST be true!

 

[DarthShader]

I have the gpu!

And you are open to talk about it on the internet? Ok, I get it.

 

[ECH]

What is AMD's position on Tessellation? I read some snippets when it came to the Hawx 2 demo but not enough to get an understanding if AMD/Nvidia are on the same page or not.

 

[Arnold Beckenbauer]

My crazy idea: 1920 SPs (VLIW5) = 1536 SPs (VLIW4). Better said: 1536 SPs (VLIW4) are as good as 1920 SPs (VLIW5).
Not technical, but it's good for some fun.

 

[fehu]

So the family is called Northern Islands, but the chips are named after islands in the south of the Northern Hemisphere, and individual products are named after… random birds?

What the hell?

AMD guy: "nVidians'll never know what hit them"

 

[I.S.T.]

Fud is pro-NV in general, SA is pro-ATI, and the third, whatever the site was called, is just "don't even look there"

I think all three sites qualify as "don't even look there."

 

[Forrest]

And you are open to talk about it on the internet? Ok, I get it.

Hi, my name is Forrest and everyone knows who I am.

 

[chavvdarrr]

Why do you think that Cayman has to fight that thingie? Have you already forgotten that Antilles will do the job right?

That "thingie" will be the fastest single GPU card on the market.
Unless something unexpected is coming ANtilles will be hotter dual-GPU AFR-abomination.

 

[UniversalTruth]

I don't know what you are smoking but people expect Cayman to be the fastest GPU on the market. I think you have to re-think and re-read everything.

 

[Shtal]

I think AMD (Former ATI graphics) will be the faster single GPU for Q4-2010 until Nvidia responds next year 2011 with GF110 bug fix Fermi second edition; but then again we don't know how fast cayman will be and maybe GTX580 will be to slow against Cayman.


EDIT: I really can't stand paper-launch GPU's and that is what Nvidia doing for 2010 with GTX580 (Proper name should have been GTX490)......

 

[CarstenS]

Sure that the reduction can't be asymmetrical?

This time I think it actually could, contrary to what we were seeing with Cypress.


WRT to clocks exposed to OpenCL: Nvidia had 810 MHz standing there for a long time. Some intermediary state or whatever. See my confusion over it here: http://forum.beyond3d.com/showthread.php?t=58935

 

[ZerazaX]

I don't know what you are smoking but people expect Cayman to be the fastest GPU on the market. I think you have to re-think and re-read everything.

Seeing as how every source out three is talking about AMD readying its return to the high end GPU game... going to go ahead and save this quote

 

[A.L.M.]

However , we still need to make up for the lost efficiency of processing transcendental operations , so 1536 ALU might be a little insufficient for that , and the need for higher ALUs count would be direr .

Maybe I didn't pay attention to the whole discussion... But who did say that the unit that has to go is the "fatty" one?
I mean, to me it doesn't make sense at all to eliminate the only unit capable of doing transcendental operations... It would mean to rethink all units (making them all special operations - capable), while they could make it easier and just trim one out of the four "simple" units...

EDIT: obviously, this is true unless AMD deliberately wants to have spec ops done much slower than in past...

 

[Gipsel]

Maybe I didn't pay attention to the whole discussion... But who did say that the unit that has to go is the "fatty" one?

Just click on my signature for a starting point.

 

[Robin B]

Some rumored specs from Fuad. If true, how in world could this card be much faster than the 5870?

1536 stream processors
We managed to get some specs of the soon-to-launch, well at least on paper, AMD Radeon HD 6970 card.

What we can share with you is that Cayman packs 1536 stream processors, 32ROPs, 96 Texture units and 2GB of GDDR5 memory.

http://www.fudzilla.com/graphics/item/20759-some-cayman-radeo-hd6970-specs

 

[Picao84]

Some rumored specs from Fuad. If true, how in world could this card be much faster than the 5870?



http://www.fudzilla.com/graphics/item/20759-some-cayman-radeo-hd6970-specs

Yup, it doesnt make sense. Charlie Demerjian went on record saying that 4 Cayman shaders represent roughly 98% of the Cypress 4+1 combination.
1536 SP is rather close to 98%, being 96%. If that is case, Cayman performance will depend a lot on uncore parts of the chip, probably, because shader power will stay the same. I have a hard time coping with that.