-
Notifications
You must be signed in to change notification settings - Fork 38
Description
The Landau Damping + FEM solver test case in Alpine is failing to produce correct results in the multi-GPU case. The correct results should produce data with Ex field energy in the order of O(10), but it is O(10^4) or even larger.
This is found to be due to load balancing, as when I change the load balancing threshold to 1.0 (i.e. load balancing off), the simulation runs correctly (Ex field energy is in O(10) again).
Problem sizes which I checked:
./LandauDamping 17 17 17 100000000 10 FEM 0.01 LeapFrog --overallocate 2.0 --info 5
./LandauDamping 33 33 33 83886080 10 FEM 0.01 LeapFrog --overallocate 2.0 --info 5
What works (changing load balancing threshold from 0.01 to 1.0):
./LandauDamping 17 17 17 100000000 10 FEM 1.0 LeapFrog --overallocate 2.0 --info 5
./LandauDamping 33 33 33 83886080 10 FEM 1.0 LeapFrog --overallocate 2.0 --info 5