[numerobis]

Big Kepler...

http://www.tomshardware.com/news/nvidia ... 15683.html

[dmeyer]

Big Kepler...

http://www.tomshardware.com/news/nvidia ... 15683.html

Now you're talking...

[Fernando Tella]

It's just for GPU rendering? I can't see any display connector.

[dmeyer]

It's just for GPU rendering? I can't see any display connector.

The final version will probably have one connector, the current Teslas do. But yes, they are engineered primarily for GPU computing.

Overall though my experience with different GPU enabled apps has been mixed. Its not the panacea some seem to think, at least not yet. Memory limitations, driver issues etc. You can see very big performance swings based on which nvidia/radeon driver you are using. If you are tuning a workstation for one specific purpose it can be great but getting one setup that will advance all applications is near impossible.

The next two years will be particularly interesting because of the more focused nature of CUDA with this K20 will go up against some of the first commercial implementation of the Intel MIC stuff which should run regular x86 code.

[dmeyer]

Or just wait for this, no special CUDA or OpenCL API.

http://lenzfire.com/2011/11/intel-says- ... ing-32261/