http://www.hexus.net/content/item.php?item=18488

That's similar to what I read somewhere in the Andantech forums. They may opt for the 2 GB config for the flagship, but that doesn't really matter now.

With these specs it's a real beast. And there's more: the shaders have been reworked for better flexibility, which could go a long way to extracting more real world performance from the theoretical peak performance. Considering G200 already brought massive "performance increases per GFLOP" for GPU-Grid with much smaller changes.. we might be in for a pleasant surprise. Double precision is also rumored to be considerably improved.

MrS
____________
Scanning for our furry friends since Jan 2002

We estimate a minimum of 2.5 times faster over a GTX285.

gdf

Marketing BS. The shaders get more flexible, which is definitely a welcome addition and might be a very clever implementation. The marketing guys use terms like "new architecture" and "radical" far too easily.
Just consider this: nVidia has put 4 years of development and an unheard of (for GPUs) amount of money into the development of G80. It's not yet time to throw that investment away, especially when you could instead polish that still somewhat raw diamond further.

MrS
____________
Scanning for our furry friends since Jan 2002

Sure they have different teams. But these can could also work on different iterations of the same underlying architecture, couldn't they? (if they're careful not to solve problems twice)

My point was that marketing guys would probably also call Phenom / K10 a "new architecture", whereas in reality it's a heavily polished K8, which itself is a heavily polished K7. K7, Pentium Pro or Pentium 4 - these were really new architectures.

MrS
____________
Scanning for our furry friends since Jan 2002