GPU Acceleration on a RTX 2060

    • DevmalyaB

      Hi all,

      I While Ansys mentions that GPGPU cards compatible for acceleration are only Teslas & Quadros, I have seen people accelerating Fluent Solvers with RTX cards. I have a RTX 2060 Max-Q with 6GB of VRAM and I have set an system environment variable of ANSGPU_OVERRIDE=1, which enables non-compatible cards to be used for acceleration. I also enabled the gpu accel by writing in the console (and checked that my GPU is available):


      It does show this text when i start the solver, which means that the GPU should be used for accelration... :

      AMG on GPGPU

      AMGX version 1.0.0-build105

      Built on Sep 18 2014, 20:05:05

      Compiled with CUDA Runtime 6.0, using CUDA driver 11.3

      But nope!... My GPU usage was 0 😔 All cores of my CPU were being used but not gpu. I am modelling a object with a mesh size of 1 million cells (read that 1M cells need approx 4GB of VRAM) so I did not think it would be a problem. When it did not work, I even tried with a 60k cell mesh, but nope!

      Can anyone tell me what else do I need to do to get the GPU to accelerate my solver?

    • DrAmine
      Ansys Employee
      Ansys Staff won't comment on the usage of non-compatible GPU Cards.
      Perhaps you better estimate if your case will take advantage from GPU: Check the LE Timer after running of couple of iterations. If that time is more than 80% then GPU is helpful
    • DevmalyaB
      Yes I will surely do that. That was the main objective, to check whetherthis GPU is any good. But I am not able to do that since there is no GPU usage in the first place! If only there was any script or any commands other than the ones I've already used that would let me accelerate using my RTX then it would be really helpful! Any clue?
    • Rob
      Ansys Employee
      Fluent will use resource as it needs it, if it doesn't want/need the gpu nothing will happen.
Viewing 3 reply threads
  • You must be logged in to reply to this topic.