Nvidia Releases OpenCL Driver, SDK



+ Add a Comment

Keith E. Whisman

Way to go Will they listened to you. You go girl.



It's my understanding that OpenCL is a lower level language that interfaces more directly with the GPU and CUDA is a bit higher up.  The real issue is that parallelization of code is pretty tricky especaily with the complex memory allocation that has to happen on a smaller scale than with RAM on a CPU.  You can't currently take existing code in C or Fortran or whatever that has been parallelized with MPI and compile it for CUDA, which really kills the appeal for 90% of people.  At least in teh scientific community we are all waiting until there is a very solid foundation and standards before we try to port all of our code to a GPU.


digital demon

I've been wondering this since I first read Will's column: if CUDA is an architecture and OpenCL will run on top of it, then why kill CUDA? Aren't they going to work hand-in-hand?



...Processing powerhouses.

Imagine dual-GPU cards in Tri-SLI, with current cards already with multiple shader units, each unit, a separate, thread for programs!

Imagine video encoding with that!

Log in to MaximumPC directly or log in using Facebook

Forgot your username or password?
Click here for help.

Login with Facebook
Log in using Facebook to share comments and articles easily with your Facebook feed.