ATI Stream SDK v2.0 BETA Available With OpenCL

ATI Stream SDK

AMD has released a beta version of its ATI Stream SDK.

The ATI Stream Software Development Kit (SDK) is a complete development platform to allow you to quickly and easily develop applications accelerated by ATI Stream Technology. This version 2.0 has a new feature: you can now develop applications in OpenCL (Open Computing Language). OpenCL is the first truly open and royalty-free programming standard for general-purpose computations on heterogeneous systems.

This SDK contains the OpenCL.lib + OpenCL headers you need to compile OpenCL programs like this tutorial. Cool… GPU Caps Viewer will love that!

But AMD has only provided a VS2008 solution… not cool for all VS2005 developers!


[via] | [via]


  • Pingback: OpenCL Physics Simulation Across 6 AMD Opteron (24 Cores) | The Geeks Of 3D - 3D Tech News

  • Ziple

    Why just not use VC++ express?

  • verdy_p

    Why not VC++? simply because OpenCL is NOT C++ and NOT C++ but it CAN run on GPUs wiht exactly the same language for writing your kernels for various shaders, and that your OpenCL application will transaprently run on various GPUs or on small threads on your CPU, depending on available resources.
    This allows OpenCL to become a new virtual machine environment, that can be recompiled into binary form for each CPU or GPU, and that should run in safe mode (if the OpenCL compiler implemented in the device driver works correctly).
    OpenCL also features many streaming functions that can maximize the use of streaming instructions in your CPU. But even if the GPU is used for something else, the CPU can still be a valid alternate resource for running some kernels.
    Well, using 24 CPU cores is probably much more expensive than using some GPUs on relatively cheap graphics board that run MUCH faster.
    For about twice the price of an additional CPU, a single GPU will perform 10-20 times better for massively parallel operations. And GPU computing scales much better and more easily (it can even run and be connected via relatively cheap buses like a board on a PCI express slot, or connected via Firewire, when it will in fact be used for intensive parallel computing.
    nVidia and ATI are now building platforms for fast computing, and make their GPU not just for display, but for general computing (see the PhysX extension in nVIDIA which is just made for that, and not tuned only for computing pixel shaders, vertex shaders or geometry shaders: they are now build for unified shaders and generalized by supporting even more streaming instructions on GPUs that have many more parallel cores running in the same chip than CPU have in their chips).
    But also, being able to run a OpenCL kernel on the CPU allows easier debugging/tracing for your apps than on a GPU. So you can test your formulas, evaluate various precision limits, and the fine tune the OpenCL kernel that will finally be running on a GPU…