What is causing error 50193 for RLGO with GPU utilization activated?
Hello,
I am attaching the .cfm file and a screenshot of the error I receive when trying to run the simulation on my end. Could you advise on how to resolve the error, such that I can run RLGO on the GPU? It might be that it's incompatible with solely a plane wave source. Ultimately trying to do a radar cross section problem.
Thanks for your help,
Luke
Answers
-
Hi Luke,
As the error message says, curvi-linear mesh with GPU is not supported. Curvi-linear mesh is default in RL-GO, but you can also disable it and use flat triangles. Please note that in most cases multi-core is faster than GPU. You can give it a try.
Best regards,
Torben0 -
Torben Voigt_20420 said:
Hi Luke,
As the error message says, curvi-linear mesh with GPU is not supported. Curvi-linear mesh is default in RL-GO, but you can also disable it and use flat triangles. Please note that in most cases multi-core is faster than GPU. You can give it a try.
Best regards,
TorbenHi Torben, thanks for your reply. The same error message appears when curvi-linear meshing is disabled. However, if you suggest multi-core CPU processing is still faster than GPU usage I would be interested. Do you happen to know for what cases is GPU faster than multi-core? If any?
Thanks for your help,
Luke
0 -
Luke_21978 said:
Hi Torben, thanks for your reply. The same error message appears when curvi-linear meshing is disabled. However, if you suggest multi-core CPU processing is still faster than GPU usage I would be interested. Do you happen to know for what cases is GPU faster than multi-core? If any?
Thanks for your help,
Luke
Hi Lukas,
Please attach your model (.cfx) so I can check why trhe error message still comes up with flat triangles.
GPU acceleration is ideal for models using the FDTD solver. Here I would say that GPU in general clearly dominates multi-core.
Best regards,
Torben0 -
Torben Voigt_20420 said:
Hi Lukas,
Please attach your model (.cfx) so I can check why trhe error message still comes up with flat triangles.
GPU acceleration is ideal for models using the FDTD solver. Here I would say that GPU in general clearly dominates multi-core.
Best regards,
TorbenHi Torben,
Apologies for the delay. With the triangular mesh (not curvi-linear) I still get the error. I attached a .cfm feko file in the original post. Are you able to access that? I'm curious if it generates the error on your end.
I don't have immediate access to that computer to copy and send the .cfx file today, but can do so tomorrow. Just let me know if the .cfm does not work for you.
Thanks again for your help,
Luke
0 -
Luke_21978 said:
Hi Torben,
Apologies for the delay. With the triangular mesh (not curvi-linear) I still get the error. I attached a .cfm feko file in the original post. Are you able to access that? I'm curious if it generates the error on your end.
I don't have immediate access to that computer to copy and send the .cfx file today, but can do so tomorrow. Just let me know if the .cfm does not work for you.
Thanks again for your help,
Luke
Hi Luke,
Unfortunately, the .cfm file (CADFEKO session) requires the .cfx file (CADFEKO model).
Best regards,
Torben0 -
Torben Voigt_20420 said:
Hi Luke,
Unfortunately, the .cfm file (CADFEKO session) requires the .cfx file (CADFEKO model).
Best regards,
TorbenHi Torben,
Gotcha, attached is the .cfx file. Thanks for your help. Please let me know if you get an error when using a GPU to solve.
Thanks,
Luke
0 -
Hi Luke,
I didn't realize it because I only use GPU acceleration extremely rarely (parallel computing is typically faster, except for FDTD), but even the "Adaptive" setting is not supported with GPU:
This is actually a very useful feature, which automatically determines where many ray tubes need to be shot and where not. With GPU you are unfortunately forced to use "Fixed grid increments", where you may have to successively reduce the increment to make sure it is fine enough.
The attached model works with GPU, but I think it might be faster even with only 4 parallel processes.
Best regards,
Torben0