Running on GPU

GLASS is fully Python Array API compatible as of version v2026.2. Thus, it is possible to pass arrays from any Array API compatible backends. Therefore, running GLASS on GPU is as simple as selecting an array backend which is both Array API compatible and GPU enabled. For example, if JAX is installed with the relevant optional dependencies it will by default utilise any GPU devices it has available.

Unfortunately, we have found that this does not necessarily mean running GLASS with GPU enabled JAX will give an immediate performance boost. The performance gains, or as is often the case, degradations of running GLASS on GPU depends on many factors including the size of the problem and the accelerator architecture available.

For examples of running GLASS on GPU, check out our benchmarks.