GK110 = Kepler-based Tesla K20 GPU, with up to 2,880 SPs, 7.1 Billion transistors!

[iMacmatician]

I'm calling BS on those specs. K20 won't have a 320bit memory bus; NVIDIA needs every drop of memory bandwidth they can get for this market.

Apparently they're real (original).

Reil [CADnetworks' Managing Director Enrico Reil] emphasized in an interview that all the data comes directly from Nvidia and also be in writing.

 

[sontin]

As long as nVidia did not published the specs they are not "real".

But it think for the "get go" product it's alright. The K5000 has only 160GB/s and 2,1 TFLOPs SP performance with 122 Watt. This K20 is much faster and is only using 86% more power for 66% more SP performance.

 

[Lonbjerg]

CUDA 5.0 is out, with support for GK110:
http://developer.download.nvidia.com/GTC/PDF/GTC2012/PresentationPDF/S0641-GTC2012-CUDA-5-Beyond.pdf

Dynamic Parallelism
Spawn new parallel work from within GPU code on GK110