Nvidia Tesla GPUs stop working at high particle numbers
Hello,
I already contacted the developers of DualSPHysics, but maybe someone of the other users here encountered the same problem and knows a solution.
I am trying to simulate a comlex floating body, which works very well. Unfortunately the comparison with experimental data shows a big difference in the pitch angles, so we need to do the simulation with a much finer resolution.
We use a Nvidia Tesla K80 and a K40, but the limit of simulation seems to be at ~7 million particles. If we want to get finer, the simulation stops without giving any reason. GenCase creates the initial setup, but the DualSPHysics code doesn't start or stops directly after it started. This error occurs on both cards.
This is how the Run.out file looks and the output file of the server looks as well:
--------------------------------
GenCase4 v4.0.027 (11-05-2016)
================================
...
Allocated memory in CPU: 1051300270 (1002.60 MB)
Allocated memory in GPU: 2066597824 (1970.86 MB)
Part_0000 11110779 particles successfully stored
[Initialising simulation (on3cg6z4) 11-11-2016 22:59:04]
PART PartTime TotalSteps Steps Time/Sec Finish time
========= ============ ============ ======= =========
--------------------------------
So as you can see, no error message or anything.
According to the developers, we did nothing wrong with the setup of the simulation, so maybe we have to change some settings of the GPUs to reach their computing limit?
Regards
Jannik
I already contacted the developers of DualSPHysics, but maybe someone of the other users here encountered the same problem and knows a solution.
I am trying to simulate a comlex floating body, which works very well. Unfortunately the comparison with experimental data shows a big difference in the pitch angles, so we need to do the simulation with a much finer resolution.
We use a Nvidia Tesla K80 and a K40, but the limit of simulation seems to be at ~7 million particles. If we want to get finer, the simulation stops without giving any reason. GenCase creates the initial setup, but the DualSPHysics code doesn't start or stops directly after it started. This error occurs on both cards.
This is how the Run.out file looks and the output file of the server looks as well:
--------------------------------
GenCase4 v4.0.027 (11-05-2016)
================================
...
Allocated memory in CPU: 1051300270 (1002.60 MB)
Allocated memory in GPU: 2066597824 (1970.86 MB)
Part_0000 11110779 particles successfully stored
[Initialising simulation (on3cg6z4) 11-11-2016 22:59:04]
PART PartTime TotalSteps Steps Time/Sec Finish time
========= ============ ============ ======= =========
--------------------------------
So as you can see, no error message or anything.
According to the developers, we did nothing wrong with the setup of the simulation, so maybe we have to change some settings of the GPUs to reach their computing limit?
Regards
Jannik
Comments
So no problem with DualSPHysics and no problem with the Tesla GPU.
Regards
Jannik
Happy to read that the problem was finally solved, Jannik
Regards
Multiprocessors 15 (2880 cores)
Memory global 11519 MB
What limit for this device, I do't inspect