NVIDIA CUDA and OpenCL Roadmaps – CUDA PTX Bytecode


IN this CUDA roadmap, we can see that a version 3.0 is planned at the end of 2009. And here is the OpenCL roadmap. The first release of OpenCL is planned for July 2009:


OpenCL does not mean the end of CUDA. CUDA is a proprietary API that allows NVIDIA to update it quickly and frequently to add new GPU features without unveil the architecture of NVIDIA new GPUs. So CUDA will be always ahead of OpenCL in terms of features.

CUDA and OpenCL compilers will use the same byte code called: CUDA PTX This byte code is used to create the low level code executed by the GPU:

NVIDIA CUDA / OpenCL compilers

This technique is not new on NVIDIA graphics cards. The GLSL (the OpenGL shading Language) is compiled via the Cg compiler. There are many common things between CUDA and OpenCL then it’s almost logical that NVIDIA uses its work on CUDA to offer OpenCL. And I think DirectX11 Compute shader will follow the same path.



  • saew

    But why’re continuing to release CUDA SDKs if they’re going to support OpenCL?
    Why not to make OpenCL extensions to support specific NVhardware features? Is a non-sense for me…

  • Denis

    Well, there are things supported in CUDA that are not in OpenCL 1.0
    It is not at all about hardware features, it is also things like template support. Also on the CUDA roadmap is C++ and fortran support, which is not (yet) in OpenCL.

  • cgtext

    After all CUDA will be optimized by its NVIDIA GPU.

  • developer

    NVIDIA is putting money on both horses.

    CUDA is of course by far more developed but OpenCL has in theory great promises.

    But OCL has basically been tailored around CUDA, NVIDIA has had a hughe say in the workings of it.