Fermi underpowered?

[monkey7]

And 400mm^2 is just 20*20mm, or below 1*1 inch for you imperials out there.

[Zoot]

Originally Posted by name='sammytomjohn'

think he means 40nm

Originally Posted by name='siravarice'

I thought he might ><

Originally Posted by name='monkey7'

And 400mm^2 is just 20*20mm, or below 1*1 inch for you imperials out there.

^^This is what I meant. 400mm^2 isn't very big physically but its enormous for a die size. They're asking a lot of TSMC on this one, I'm not surprised in the least that there's issues.

That said, Semiaccurate ain't the most credible of sources at times best to take it with a grain of salt none the less.

Originally Posted by name='Rastalovich'

I've highlighted the other error.

Error?

[sammytomjohn]

Originally Posted by name='Zoot'

Error?

where you said in the comment of yours he quoted you said fact! it is not fact just speculation and rumours!!

[Zoot]

Originally Posted by name='sammytomjohn'

where you said in the comment of yours he quoted you said fact! it is not fact just speculation and rumours!!

Well Fermi is on a 40nm process which thus far only has ATi and Nvidia's GPUs (I think).

And also, the process is relatively new; there's bound to be issues with it especially with a rather large die size.

It's the way the IC fabrication process works, die shrinks on the deep sub-micron level always present issues. The leakage current that intel encountered upon switching from 65nm to 45nm is one such example which they sorted with the use of High-K Gate Dielectrics is one such example.

(Look up intel's website if you want more info)

Also the shortages of the 5XXX series is another indication that the process is not completely solid yet. However we haven't seen irrefutable proof yet, but IMO you can bet there are big issues with a new state of the art process with a colossal die size.

[Rastalovich]

Zoot 40nm process has been used by other companies in the past, not for gpus tho. Ti are doing 2xnm processes with tsmc atm.

[siravarice]

Yeah Intel are now on 22nm with that ridiculous Cloud Processor with like, 48 cores.

[Zoot]

Originally Posted by name='Rastalovich'

Zoot 40nm process has been used by other companies in the past, not for gpus tho. Ti are doing 2xnm processes with tsmc atm.

Fair enough, that I didn't know. What sort of parts?

Also you can safely say a GPU is a more complex part than most of TI's parts. (DSP's, Microcontrollers etc.)

With a die size as big as Fermi there's bound to be issues. Granted though that 400mm^2 is hearsay at the moment but if the past GT200 series is any indication the die size is going to be rather large in comparison, yields can't be good in the beginning.

[Rastalovich]

That's the biggest problem, it's pretty much all hearsay.

Texas Instruments have chips everywhere, their mostly known these days for their firewire presence. But I'm sure if u sat down with a Ti rep, they could talk for days about their other interests. Mainly data conversion, comms and the like afaic. Don't forget the calculators

[Zoot]

They do indeed. Must do a bit of digging to see that they have on the lower processes.

I'm working with a 32bit Floating Point 225MHz DSP from them at the moment. Quite a nice part I must say, very easy to work with.