BASEMENTBasic Simulation Environment for computation of environmental flow and natural hazard simulationLaboratory of Hydraulics, Hydrology and Glaciology (VAW)ETH Zurich | ![]() |
You are not logged in.
Pages: 1
It happend to me a couple of times, that with very cumbersome simulations (very large domains +200k cells), run with the cudaO backend, the run would start using all the allocated resources and at a certain point it drops and started using half the allocated resources.
If I stop, close basement, reboot and restart the simulation from where I stopped it, the problem persists. If I start the simulation anew it uses the allocated resources and at the same point as before performances drop.
I tested it on different computers both having
windows 10
basement 3.1.1
NVIDIA quadro P620
but it happend also with basement 3.1
I have the feeling it might have something to do with basement because I tried to copy the sim to a second workstation (results.h5 included) and make it start from where I stopped it on the first workstation and it started using 5 cores instead of the 9 I allocated to the process.
I am still googling to understand why, I fear it might be due to windows limiting the use of resources if the temperature cpu temperature is too high, or something like that.
EDIT: if I use the omp backend it works perfectly using all the allocated threads.
EDIT 2: other strange thing happening to me right now is a simulation running with peaking cpu % (see image below)
EDIT 3: it happens also on another machine with a NVIDIA quadro P2000. Allocated CPUs = 15, used CPUs = 12.
Last edited by Matteo Facchini (2021-03-31 08:14:04)
Offline
This occurs to me also with BMv4_BASEHPC_cudaO_single.exe
I still cannot figure out why.
Offline
Hi Matteo,
Do I understand correctly that this occurs only with cudaO backends?
Best regards,
Matthias
Offline
Pages: 1