align-toparrow-leftarrow-rightbackbellblockcalendarcamerachatcheckchevron-downchevron-leftchevron-rightchevron-small-downchevron-small-leftchevron-small-rightchevron-small-upchevron-upcircle-with-crosscrosseditemptyheartfacebookfullheartglobegoogleimagesinstagramlocation-pinmagnifying-glassmailmoremuplabelShape 3 + Rectangle 1outlookpersonplusImported LayersImported LayersImported Layersshieldstartwitteryahoo

Maximizing Performance on AMD's Radeon Cards

AJ will provide an outline of the HD7970 AMD GPU architecture, and illustrate the keys to top-performance through a set of OpenCL benchmarks. You will learn how to use CodeXL to tune your kernels. AJ will also provide advice on how to achieve maximum performance based on his own experience. There will be two OpenCL lectures uploaded to YouTube that you should watch prior to this talk that will provide all required background (AJ expects to upload these videos by Friday November 1).

Join or login to comment.

  • AJ G.

    Hi everyone,

    For those of you who could not make it, this YouTube video covers most of the talk that I gave: http://youtu.be/1Yr-E9w4tGI It has been slightly modified from my original talk based on feedback.

    There is another segment that will be added to YouTube later on.

    AJ

    November 14, 2013

  • Greg

    There was question last night about OpenCL books.

    The 1st book published OpenCL Programming Guide, by Munshi, Gaster, and Mattson describes things reasonably clearly and demostrates concepts with sample code.

    The 2nd, Heterogeneous Computing with OpenCL, by Gaster, Howes, Kaeli, Mistry, and Schaa, is horrible. The code examples have mistakes, not just syntax but logic problems, missing steps, etc. The authors either have no idea how to implement anything in OpenCL, or they are unable to communicate how. The 3rd, OpenCL in Action by Scarpino is brillant. The explanations are suscinct, explain how things work, and illustrate how to do things with sample code. This sample code is robust boilerplate that can be modified or used to build applications. Buy this book.

    I am aware of a 4th book, Integrated Framework For Heterogeneous Embedded Platforms Using Opencl and a 5th book, OpenCL Parallel Programming Development Cookbook. Never seen these books, can't comment.

    2 · November 8, 2013

  • Chris S.

    AJ highlighted the issues critical to maximizing the performance of a GPU from the perspective of the OpenCL programming model. He covered the need to distinguish between compute and memory bound applications, addressed divergence and coalesced memory access with insightful comments about hardware, and discussed resource constraints. He concluded his presentation with a simple set of guidelines. Checkout his screen-cast, which he hopes to post shortly. Thank you AJ for your contribution. I was comforted to learn that I have been heading in a reasonable direction in my academic work on programming GPU accelerators.

    November 7, 2013

  • AJ G.

    Hi everyone,

    Here are the two videos that you should watch before you attend:

    High-Level OpenCL: http://youtu.be/8D6yhpiQVVI
    OpenCL C: http://youtu.be/RKyhHonQMbw

    If you are already very familiar with OpenCL, including the NDRange, asynchronous execution, vector types, and device model, then you don't need to watch them. I will start my talk by asking if there are any questions on this material.

    These videos are all part of a series that I have developed to provide a complete background in OpenCL and GPU programming to newcomers. This is still just background, I intend to provide details on how to tune your algorithms to GPUs in future lectures.

    For those of you who cannot attend, I will upload a screencast of this material on the weekend.

    I look forward to seeing everyone!

    November 3, 2013

17 went

Our Sponsors

People in this
Meetup are also in:

Sign up

Meetup members, Log in

By clicking "Sign up" or "Sign up using Facebook", you confirm that you accept our Terms of Service & Privacy Policy