Electronics

Electronics

Is EDT compatible with rCUDA acceleration? Is it planned?

    • AndyJP
      Subscriber
      Since it is hard purchasing a lot of top-class Quadro/Tesla boards per every workstation, and equivalent GTX boards are not supported, I was thinking about remote execution.nnThere is a thing called rCUDA, a CUDA server that transparently and dynamically allocates local NVIDIA CUDA boards to remote clients executing CUDA code.nThe only requirement is that the software should have CUDA library linked dynamically at runtime (for substituting the CUDA library with a client wrapper).... And obviously, it should not check the GPU version, since it may see something weird instead of Quadro.../Tesla...nnSo was it tried in ANSYS or by anyone?n
    • Randy Kosarik
      Ansys Employee
      Hi AndyJP, nrCUDA has not been tested with ANSYS products.nAdditionally, I have not located any references to other customer inquiries regarding rCUDA.nthanksnRandyn
    • AndyJP
      Subscriber
      Definitely not many RF customers know about rCUDA. It is a modern deep learning HPC thing. But you could offer its support as a valuable feature. Every developer searches for cost-down, and this would offer a serious cost down without ANSYS loosing anything. CUDA licenses are linked to HPC, so there will be no reduction in license purchase. Instead, funds can be rearranged for more ANSYS licenses.n
Viewing 2 reply threads
  • You must be logged in to reply to this topic.