Aspects

Cublas

The NVIDIA CUDA Basic Linear Algebra Subroutines (cuBLAS) library is a GPU-accelerated version of the complete standard BLAS library for use with CUDA capable GPUs.

Magma

MAGMA (Matrix Algebra on GPU and Multicore Architectures) is a collection of linear algebra libraries for heterogeneous processor architectures.

Others

Quality | Example |
---|---|

Higher performance |
"Magma routine magma_gemm has higher performance than cublas in some cases" from question "Tips for optimizing X_transpose*X CUDA kernel" |