2 Replies Latest reply on Nov 8, 2011 11:15 AM by N3KO

    AMD OpenCL under VMWare ESX

      Is AMD OpenCL supported under WMWare ESX & how do you set it up?

      I read this recent blog posting from a person from VMWare on how he got VMWare's ESX hypervisor working with AMD's OpenCL GPU implementation & also with nVidia's CUDA:


      The AMD CPU stuff would obviously work, but I'm skeptical that the GPU works.

      I'm interested in doing this myself using WMWare Workstation v8 under Windows 7 and LINUX.

      Can AMD:


      • Confirm that they have also tested and support this?
      • That the GPU accleration & all features of OpenCL work with no problems?
      • Tell us where the performance hit will be & how much it would be?
      • How I would need to set this up to make it work?
      • Whether it is supported under both Windows 7 & LINUX?
      • What AMD SDK it was tested with, which GPU cards & which AMD drivers?
      Thanks in advance for your assistance in this!

        • AMD OpenCL under VMWare ESX


          I'm also interested in this, but since it looks like you're new around here I'll give you a heads up.  AMD is 1/30th of the size of Intel and doesn't really have the resources to test everything that you have in that list.  It's amazing they can actually compete.  I wouldn't expect any corporate assistance on this one unless you represent a very large customer like Amazon and you want to use GPUs in your next cluster.

          Good luck and I'm interested in hearing if you have any success.

            • AMD OpenCL under VMWare ESX


              I think that the VMWare community might have better answers to your questions:


              VMWare Workstation/Server and VMWare ESX are different things. VMWare workstation sits on top of your host OS, so your virtual OS has to go through the host before accessing hardware. In contrast, ESX is sitting just in top of the hardware (so it has direct access to it).

              About GPU code performance, once the code and data is on the GPU it runs independently from the host so I would expect no overheads due to virtualization.