AMD: R7xx Speculation

Status
Not open for further replies.
But RV670 uses the first 55nm TSMC node and there is a better one. RV770 should take the newer one.
If you could please explain in more detail how this 'better' 55nm process increases the amount of logic/mm2 by 20%? The semiconductor industry would be most interested.
 
Wrong math. 34.7% would be 897 million. :LOL:

It's really just silly trying to get that accurate with a transistor number when the area is basically an estimate.

So, RV670 666Million; if RV770 is truly about 897Million - then thats put about ~231Million more transistors.

Now the question remain is to split those transistors into 3 category?

A: partially for texture units.
B. partially for stream processors.
C. now the question remain - will ATI increases ROP's count or not.


Edit: I asked this question before and nobody seems able to calculate approx how much transistor count eats-up for additional texture units and additional increase stream processors and now possibly ROP's.

Thanks,

Shtal :)
 
Last edited by a moderator:
If you could please explain in more detail how this 'better' 55nm process increases the amount of logic/mm2 by 20%? The semiconductor industry would be most interested.

Didn´t said that increased logic by 1% or 20%.
Just that there is a new node at tsmc.
You can get a chip between 900-1000M. 1B is a more round number, just that and they like to markting big numbers.
 
C. now the question remain - will ATI increases ROP's count or not.
They haven't done this since... May 2004, when Ati introduced its X800-series, but ever increasing each ROPs capabilities.

So, one could argue, we're seeing a trend here, while on the same basis, one could say, it'd be about time to increase that number.

However, even at 2.560x1.600, 16 ROPs at 850 MHz (that's what i at least expect) would yield about 3320 Fps - enough even for Q3.

If ROPs would be doing everything single-cycle, not stalling etc.pp. - ideal circumstances, if you will.
 
If you could please explain in more detail how this 'better' 55nm process increases the amount of logic/mm2 by 20%? The semiconductor industry would be most interested.

Oh puhleeze, like you didn't know about the fairies and their inexhaustible supply of pixie dust.
 
40-50% improvement can put it in the way of 9800GTX. With good pricing strategy it can be amazing to recover ATI.

...

Nope, and I bet many other enthusiasts out there will see it in the same light.

RV770/R700 needs to exchange blows with GT200/G100/D10U, otherwise we'll be in the exact same scenario as we are now--NVIDIA dominating the high end and thus drawing in "sheep customers" to performance and mainstream parts.

By exchange blows I mean winning at least 2 out of every 5 benchmarks (not counting synthetics).

A brand new high end architecture that only manages to compete with last year's competition is a failure, IMO. But then again, this chip is called RV770...

I'm hoping that R700 doesn't turn out to be some lousy "CrossFire on a Stick" card.
 
Last edited by a moderator:
About those reported clockspeeds of 900 and up... I'm hearing something in the range of 7xx for the RV770XT.
 
About those reported clockspeeds of 900 and up... I'm hearing something in the range of 7xx for the RV770XT.

Such low? I thought to improve clock-speed was one goal of RV770 and the last number I heard was 875MHz for shipping.
 
Oh, quite on the contrary. What else might lower clocks indicate? ;)
Higher complexity => greater power draw
Higher functional unit utilization => greater power draw

Power draw can then be alleviated with lower voltages and clocks, allowing for less complex cooling and possibly higher yield.

Of course, lower clocks can be used to lessen yield problems (more sellable product per wafer), and to lower power draw for its own sake, which could be popular when negotiating with larger OEMs who then can avoid noisy cooling in their boxes or added expense for heftier power supplies.

So actually, there are quite a few good reasons to not push the clocks to the bleeding edge.
 
Oh gosh, no. I wasn't implying ATI was underdelivering on the clocks; I was rather suspecting that this was, in fact, the original design target. However, I will gladly admit that I don't actually know whether it was indeed the original target...
 
Status
Not open for further replies.
Back
Top