I have been working to solve 3104 sets of Ax=B when the "A"s are Hermitian 186x186 matrices and the "B"s are vectors with lengths 186.
I am able to use cuSolverDn to solve a single Ax=B but I need to scale it to 3104 matrices.
Is there a batched library to can solve a Dense Hermitian Ax=B?
Do you have 3104 linear systems (i.e. 3104 A matrices and 3104 b vectors), or just 3104 right hand sides (i.e. 3104 b vectors)?
If you have 3104 linear systems, I would move them all at once into big 1D arrays on the device then do a for loop and pass the starting pointers for each matrix and vector to the routine. This could be slow due to overhead.
If you have 3104 b vectors, then you can batch solve. If it requires a sparse matrix, you can just convert A to sparse first or use a GPU routine. (dense2csr)