Page 11 of 18 FirstFirst ... 910111213 ... LastLast
Results 101 to 110 of 176

Thread: What does it take to code for a Xeon Phi?

  1. #101
    Administrator
    Join Date
    Feb 2009
    Posts
    7,533
    Quote Originally Posted by __shared__ View Post
    A more balanced article on the prospects:

    http://www.hpcwire.com/hpcwire/2012-...l?featured=top
    That article restates the various PR positions without bringing anything new to the table. That said, only one things to disagree with, the S10000 is not a GPGPU card, it is aimed at VDI.

    Back to the story at hand, found one thing that I can point to, Altair - these guys http://www.altairhyperworks.com/?Asp...ookieSupport=1 - and their porting to Phi. One slide from the presentation:


    They ported it, and that is what I would call a complex piece of software. Their software runs natively on Xeon, Phi, or hybrid, and is available for purchase. As a disclaimer, they were at the Phi press brief, but their experiences are consistent with the one I can't talk about.

    -Charlie

  2. #102
    Senior Member
    Join Date
    Feb 2011
    Posts
    1,821
    Altair also have used OpenCL and CUDA, might be interesting to get results
    Speaking for myself.

  3. #103
    Banned
    Join Date
    Jan 2012
    Posts
    185
    Thank you charlie for the interesting slide.

    To me what's going to be really interesting is:

    "We put in X amount of effort and got Y amount of performance."

    Above states that "we didn't have to put in a large effort", we have X but not Y in the equation.

  4. #104
    Senior Member
    Join Date
    Feb 2011
    Posts
    1,821
    Altair already had done the work for CUDA and OpenCL so the effort was obviously minimal.
    Speaking for myself.

  5. #105
    Banned
    Join Date
    Jan 2012
    Posts
    185
    Yes, and doesnt Phi run OpenCL anyhow?

  6. #106
    Senior Member
    Join Date
    Feb 2011
    Posts
    1,821
    It's supposed to. Which is an additional reason why the porting effort isn't indicative of Xeon Phi supposed benefit over CUDA or other existing solutions
    Speaking for myself.

  7. #107
    Administrator
    Join Date
    Feb 2009
    Posts
    7,533
    Quote Originally Posted by __shared__ View Post
    Thank you charlie for the interesting slide.

    To me what's going to be really interesting is:

    "We put in X amount of effort and got Y amount of performance."

    Above states that "we didn't have to put in a large effort", we have X but not Y in the equation.
    I can send you the entire deck if you want, email me, name @ site, my PM box is full.

    -Charlie

  8. #108
    Administrator
    Join Date
    Feb 2009
    Posts
    7,533
    Quote Originally Posted by laurent View Post
    Altair already had done the work for CUDA and OpenCL so the effort was obviously minimal.
    According to preso, they said that neither CUDA or OpenCL would work for what they were doing.

    -Charlie

  9. #109
    Administrator
    Join Date
    Feb 2009
    Posts
    7,533
    Quote Originally Posted by __shared__ View Post
    Yes, and doesnt Phi run OpenCL anyhow?
    It is supposed to, but see my previous comment. That said, not sure if the implementation is out yet, anyone know for sure?

    -Charlie

  10. #110
    Senior Member
    Join Date
    Feb 2011
    Posts
    1,821
    Quote Originally Posted by charlie View Post
    According to preso, they said that neither CUDA or OpenCL would work for what they were doing.
    Ha, so OpenCL and CUDA were used for some other product perhaps.
    Speaking for myself.

Posting Permissions

  • You may not post new threads
  • You may not post replies
  • You may not post attachments
  • You may not edit your posts
  •  
WordPress Appliance - Powered by TurnKey Linux