NVIDIA GF100 & Friends speculation

Isn't 8192x8129x4bytes ? I think the current limitation of 256mb per buffer of ATI OpenCL comes from there.

 

Currently, yes. But numerical errors can creep into physics calculations in a number of different ways, and double-precision reduces numerical errors dramatically.

 

Do you have a source/link to confirm/reveal SP, TDP and price?

 

Which is of course why Ageia's original hardware used single precision...

DK

 

It's accurate enough for gaming purposes.

 

Price is not a spec of the card is it?
And we know Fermi has 512 cores for how long? The only debate was rumour it would have only 480 enabled. 448 cores for GTX470 was pretty much a given, because Tesla parts have 448. So it would make sence to be next downgrade step.

Really, this is not really new info, except for the price, which is NOT a spec.

What i mean is really that without the clocks you dont know THE final specs. Shader count and TDP are NOT enough to define card specs.

P.S. - LOL, I didnt even notice it was the same guy who posted the info in both forums xD

 

Since you are bent on buying the fastest single GPU, it may be the case that a 2GB 5870 eyefinity/oc may be faster overall and lower power to boot.

 

Will be faster in DX9 games which are fast enough even on RV770 but will fail miserably in all DX11 games where performance really matters?

 

Oh, so you have benchmarks of that card vs. Fermi?

 

Single gpu same power hog as dual gpu yet much slower?!

I find that very hard to believe.

 

I said may:

But I guess he does:

 

Are you guessing or do you know? So far the only DX11 'game' nV dared to show us was 60 second Unigine clip.

The rest of the leaks were not so promising but of course we can always argue leaks were fake or that AvP and Dirt 2 are not 'real' DX11 titles.

For me Metro is the test for GTX480. If it fails here it's an epic fail in my book.

 

Picture / source?

Between a GTX280 and a GTX260 192SP there's a smaller difference than between a GTX480 and GTX470, and still the two TDP differ of more than 50W....

 

Wow , that is such a clever use of the VLIW architecture !
Is this the case in Milyway@Home too ? I know it has amazing performance on ATi hardware too ..

 

Those should all be the same on rv7xx and rv8xx though IIRC, so wouldn't explain why rv8xx is faster.

 

There is a question mark at the end since it surely looks like you may provide us with an answer. So I'm waiting.

 

Thinking out loud, it seems Fermi engineers were distracted by LRB and forgot who their true competitor was.

 

Actually, Fermi was designed to face both competitors.

 

Yup.. All that geometry is aimed at tesselation, ATI's baby..

 

A war on two fronts, a bit like Germany in WWII, it was bound to end in tears.