NERSC supports a wide range of math libraries. Many are available as modules.
Perlmutter has additional libraries that can be used on CPUs and GPUs:
- AOCL (AMD Optimizing CPU Libraries) (installation expected in near future)
- Math libraries for NVIDIA GPUs: cuBLAS, cuSOLVER, cuSPARSE, cuFFT, cuFFTW, etc.
For information on the libraries, check the Perlmutter Readiness page's Libraries section.
Documentation is also available at NVIDIA's CUDA libraries page. Some CUDA libraries relevant to HPC are:
- cuBLAS: CUDA Basic Linear Algebra Subroutine (BLAS) library
- cuSOLVER: Library for decompositions and linear system solutions for both dense and sparse matrices
- cuSPARSE: CUDA sparse matrix library
- cuFFT: CUDA Fast Fourier Transform
- cuRAND: CUDA Random number generation library
- cuTENSOR: A High-Performance CUDA Library For Tensor Primitives
- cuSPARSELt: A High-Performance CUDA Library for Sparse Matrix-Matrix Multiplication