Speed of cusolverSPDcsrlsvlu is very slow when solving large sparse matrix linear equations

Hello, is the function cusolverSPDcsrlsvlu executed on GPU or CPU? I found cusolverSPDcsrlsvlu could move the input data (csrValA, csrRowPtrA, and csrColIndA with host attribute) to the device but the GPU occupancy was only a few percent when solving the equations, not calling most GPU cores. Especially in solving the large sparse matrix, the program would be stuck in cusolverSPDcsrlsvlu for a long time, maybe a few minutes, maybe a dozen minutes, the GPU occupancy was less than five percent and GPU memory was less than 500MB (my device is RTX2080 with 8GB memory), no matter how large the matrix was, although the memory of my matrix may be larger than 500MB. So this is not normal. Are there any settings or memory management that need to be changed when solving a large sparse matrix linear equations using this function?

Thank you.
-Yu

Hello, is the function cusolverSPDcsrlsvlu executed on GPU or CPU? I found cusolverSPDcsrlsvlu could move the input data (csrValA, csrRowPtrA, and csrColIndA with host attribute) to the device but the GPU occupancy was only a few percent when solving the equations, not calling most GPU cores. Especially in solving the large sparse matrix, the program would be stuck in cusolverSPDcsrlsvlu for a long time, maybe a few minutes, maybe a dozen minutes, the GPU occupancy was less than five percent and GPU memory was less than 500MB (my device is RTX2080 with 8GB memory), no matter how large the matrix was, although the memory of my matrix may be larger than 500MB. So this is not normal. Are there any settings or memory management that need to be changed when solving a large sparse matrix linear equations using this function?

Thank you.
-Yu