Repository navigation
Replies: 3 comments 2 replies
|
PyMAPDL executes MAPDL commands on the backend. Based on the #757 the best is to use # Slow
for i in range(1000):
mapdl.n(i, x, y, z)
# Very fast
with mapdl.non_interactive:
for i in range(1000):
mapdl.n(i, x, y, z)
# The fastest
with mapdl.non_interactive:
mapdl.run("*do,i,0,1000)
mapdl.n(i, x, y, z)
mapdl.run("*enddo")Alternatively, you can use other MAPDL commands that perform actions in batch, for instance If you provide an example I can give you better advice. |
|
Hi @germa89, thanks a lot for the detailed answer. You mentioned NBLOCK as a batch‑oriented option. Do you have any pointers to docs or examples on how to use it for node updates? That would be very helpful. If we end up with a concrete PyRBF (Radial Basis Functions mesh morphing) use case example that could benefit from further optimization, I’ll definitely follow up here with more details. Thanks again for the guidance! |
|
Hi @germa89 @dpat-ansys Would the direct MAPDL database access be faster than N or NBLOCK? I think we tested it but quite a while ago... |
Uh oh!
There was an error while loading. Please reload this page.
From internal user:
All reactions