Most GPU computation uses NVIDIA's CUDA framework. However, this is not ideal for a number of reasons:

• Only runs on NVIDIA hardware -- Mac laptops for example do not use NVIDIA anymore, and for local development and testing, it is ideal to be able to run on the laptop. Or even the phone in the future.
• Closed source.
• C-based -- integration with Go is not ideal.

Thus, in V2, we are now using the WebGPU based 'gpu' package in Cogent Core, which provides a gosl system that automatically translates Go into the wgsl shader language, creating "go as a shader language". This is used in axon -- see that for details.