GPU Computing: Past, Present and Future


Tianhe-1: China's Supercomputer
Tianhe-1, Chinese National University of Defense Technology (NUDT) Supercomputer: 2560 Radeon HD 4870 X2 : 4’096’000 shaders processors (1600 sp * 2560 nodes



Micheal Chu (Product Manager, ATI Stream Computing Software) has published a 45-page PDF that relates the history of GPU computing via ATI Stream and Radeon HD 4000 (the past), OpenCL and Radeon HD 5000 (the present) and OpenCL and AMD Fusion (the future – beginning of the era of heterogeneous systems – APU: accelerated Processing Unit).

OpenCL vs CUDA
OpenCL vs CUDA


The PDF is available HERE.

Related links:


6 thoughts on “GPU Computing: Past, Present and Future”

  1. idinev

    :(
    Math precision: undefined, cpu-support: no.

    I wonder what those dozens of pages of precision and etc in the cuda manuals are, then …

  2. fff

    AMD has to learn a lot about CUDA and NVIDIA yet:

    1. NVIDIA’s OpenCL is much much much faster currently, even using bank conflicts !

    2. AMD’s OpenCL does not support images currently, which is ** very ** unfortunate.

    3. ATI’s OpenGL interop is really bugged and incomplete. Their OpenGL 3.2 drivers are a huge mess currently.

    4. The NVIDIA CUDA/CL dev program is absolutely superior, not only in the
    number of aplication but also in the quality of the support ( for example, using the TWIMTBP program, Ventures, University programs, etc… )
    4.

  3. Emil_vivi

    There is always positives and negatives for both Nvidia and ATi.
    Why don’t we just have a peace without ‘bullying’ the other?
    I myself is a Nvidia fanatics 😉
    Without disrespecting ATi, but i think without ‘bullying’, Nvidia still going to be popular, and vice versa.

    Viva Nvidia with CUDA, PhisX, and hail GeForce!

  4. DrBalthar

    NVidia is an ill company that is why the always try to go for closed standard and even try to lock out competition. They are the Microsoft in the hardware world (and yes worse than Intel)

Comments are closed.