Nvidia Releases Final Version of CUDA 5

The platform is available as free download from its developer website.

According to Nvidia, CUDA 5 adds dynamic parallelism, which allows the GPU to create new threads, as well as GPU-callable libraries and GPUDirect Support for RDMA to enable direct communication between GPUs and other PCI-E devices. Also new is the Nsight Eclipse Edition which can be employed to develop, debug and profile GPU applications within the Eclipse-based integrated development environment on Linux and Mac OS X.

Now in its fifth generation, CUDA was first released for the GeForce 8800 series of graphics cards in 2007. CUDA 5 runs with all CUDA capable GPUs, on Windows XP 32-/64-bit and up, Linux (SUSE Server 11 SP1 and up, Open SUSE 12.1, Ubuntu 10.04 and 11.10, RHEL 5.0 and up, as well as Fedora 16), and Mac OS X. Developer will also need Microsoft Visual Studio 2008 or 2010, or the respective version of Microsoft Visual C++ Express.

Contact Us for News Tips, Corrections and Feedback

  • teaser
  • Cryio
    No Visual 2012?
    I wonder would that improve poor GTX 6xx series performance on Cuda accelerated apps like Adobe Premiere Pro CS6. At the moment GTX 660 Ti runs on par with GTX 560 Ti that is half the price..
  • Cy-Kill
    I wonder if Stream will ever be as good as CUDA, of course ATi -- or was it AMD by then -- was late to the game with Stream, so AMD is playing catch up now.
  • calmstateofmind
    This article is a disaster. No structure, abrupt dialogue, random facts that don't flow with overall paragraph, etc. If I wanted to read something like this I'd just check the release notes.

    CUDA 5 is cool and all but what are you thinking?!?

    Get it together...
  • raytseng
    and the title says "final" yet there's no blurb about exactly about that finality
  • danwat1234
    Final as in there will never be a 6.0?
  • I would say Final Means its No longer in Beta
  • dasper
    The final version or a finalized version (or finalised if you are so inclined to spell it that way)?
  • heffeque
    Less CUDA and more OpenCL, please. Stop this nonsense, nVidia.