First post here, and I'm a fairly new user to DWSIM, but have been fooling around with a few things for now.
I'm using a modified UNIFAC (NIST) model to calculate phase properties for a gasoline-like mixture, but using a dynamic IronPython script to simulate the equilibirum distillation of the mixture. I.e. recurvsively passing the previous flowsheet's bottoms back to the inlet. Running into issues with computing time currently though as the number of compounds has increased a fair bit. Taking roughly 15 seconds for each flowsheet solution.
I've downloaded the necessary software and drivers from NVIDIA for CUDA use, but it doesn't seem to be pushing any of the calculations to the GPU when I enable that setting.
Any suggestions?
Is this a situation where CUDA parallel calculations would work? Or is the solution of the EOS something that can't be quickened via parallel processing?
Thanks,
Brian
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
If I remember well, GPU routines were done for PR, SRK and Lee-Kesler-Plöcker models. I've tried to do it for UNIFAC models but saw no speed advantage because the routines aren't "parallelizable" enough.
Regards
Daniel
If you would like to refer to this comment somewhere else in this project, copy and paste the following link:
Hi All,
First post here, and I'm a fairly new user to DWSIM, but have been fooling around with a few things for now.
I'm using a modified UNIFAC (NIST) model to calculate phase properties for a gasoline-like mixture, but using a dynamic IronPython script to simulate the equilibirum distillation of the mixture. I.e. recurvsively passing the previous flowsheet's bottoms back to the inlet. Running into issues with computing time currently though as the number of compounds has increased a fair bit. Taking roughly 15 seconds for each flowsheet solution.
I've downloaded the necessary software and drivers from NVIDIA for CUDA use, but it doesn't seem to be pushing any of the calculations to the GPU when I enable that setting.
Any suggestions?
Is this a situation where CUDA parallel calculations would work? Or is the solution of the EOS something that can't be quickened via parallel processing?
Thanks,
Brian
Hi Brian,
If I remember well, GPU routines were done for PR, SRK and Lee-Kesler-Plöcker models. I've tried to do it for UNIFAC models but saw no speed advantage because the routines aren't "parallelizable" enough.
Regards
Daniel
Hi Daniel,
Thanks for clarifying. Figured that may have been the case!
Thanks,
Brian