NVIDIA Developer Zone

New CUDA Now Available

by Nadeem Mohammad, posted Dec 05 2011 at 08:00AM

This morning we posted the latest CUDA Toolkit release candidate which include a triple-pack of new features: .

  • Up to 10% performance improvement using the new LLVM-based CUDA compiler
  • Over 1000 new imaging and signal processing functions added to the NVIDIA Performance Primitives library.  If you do image processing, NPP has a GPU accelerated function for you.
  • New automated expert system in the completely re-designed Visual Profiler to give you step-by-step performance optimizations.

Learn more in this short video covering the highlights of this release.

Get the full details in the CUDA 4.1 Features Overview webinar recording, and join us in the new CUDA developer forums for a lively discussion about the best ways to use all the new features in this release!

The release candidate builds are available now at www.nvidia.com/getcuda. 

  • Login or register to post comments
  • Share this
COMMENTs
by homemade-jam
postd on Jan 07 2012 at 04:16AM

Am I being stupid? Where is the source for the LLVM compiler? When I click download I am taken to the RC 4.1 page which has all the usual downloads. Am I missing something? Is it somewhere in the toolkit once you have downloaded?

by Nadeem Mohammad
postd on Jan 09 2012 at 12:49PM

The source for the LLVM based compiler has not yet been posted.

If you are interested in the source - and want to ensure you are one of the first to be notifed when the source is released please register your interest at : http://developer.nvidia.com/content/cuda-platform-source-release

Sorry for any confusion.

by robgiorgi
postd on Dec 08 2011 at 02:45AM

...it runs also on Fedora-16: just change the /usr/local/cuda/include/host_config.h for getting rid of the message that says only GCC 4.5 is supported...

(I tested at least Ocean, Mandelbrot, MatrixMul of the SDK).