Has anybody experience with this chip? Can I use this only with trained neural networks? Or can I use it for general purpose SIMD computation? Can I address the chip conveniently from Rust or C++?
As someone who's hardware knowledge extends to "my Mac is fast enough at most things I do", could someone tell me if this would run Stable Diffusion / ComfyUI generations faster than say, a Mac Studio (ignoring all the CoreML SD stuff, just regular safetensors)?
