Hello AmgX,
I'm writing my own very compute-intensive FEA software. It should handle as much as possible as fast as possible.
I'm afraid that AmgX wouldn't be able to run multiple solvers in paralel. Since I'm running my compute on (expensive) NVidia H100s, I expect that AmgX would be able to solve multiple (as much as possible) matrices in paralel, and ideally max out it's own usage.
So, can AmgX run mutliple solutions in paralel? Because if not, I'd better rewrite it in non-CUDA languages that can vector map and squeeze out usage out of every GPU every second of time.
Hello AmgX, I'm writing my own very compute-intensive FEA software. It should handle as much as possible as fast as possible.
I'm afraid that AmgX wouldn't be able to run multiple solvers in paralel. Since I'm running my compute on (expensive) NVidia H100s, I expect that AmgX would be able to solve multiple (as much as possible) matrices in paralel, and ideally max out it's own usage.
So, can AmgX run mutliple solutions in paralel? Because if not, I'd better rewrite it in non-CUDA languages that can vector map and squeeze out usage out of every GPU every second of time.
Looking forward, MRiabov