
Solve a system you can check
inv(A)*b against A\b on a Vandermonde matrix.
Solve systems with backslash, then factor matrices and find eigenvalues and singular values with lu, qr, chol, eig, svd and the other matrix functions you already know.

Moving a factorization to the GPU usually means paying for a parallel computing toolbox. RunMat runs lu, qr, chol, eig and svd on NVIDIA, AMD, Intel and Apple GPUs from the same script, and the runtime is free for individuals, education, research and smaller companies.
GPUDense solves and thousands of small matrix products are often the slowest part of a model. Large arrays stay on the GPU between statements and pagemtimes runs batched products in one call, so far larger systems fit in the same run time with no GPU code to write.
GPUA dimension mismatch deep in a script fails the run after the slow setup has finished. Mismatched sizes in products and solves are flagged in the editor and by runmat check, with the sizes found, before anything runs.
How RunMat runs your codeWhen a result looks wrong a week later, the matrices that produced it are usually gone. Each run keeps its figures, variables and code, so they can be opened and checked.
Desktop app
inv(A)*b against A\b on a Vandermonde matrix.

Natural frequencies of a spring-mass chain from eig.

Keep five singular values of a noisy matrix.

Normal equations, QR and pinv on one fit.

Random vectors from a covariance matrix with chol.

The smallest modes of a 1-D Laplacian with eigs.