Tracked topic
… PyTorch is probably the best demonstration of that. PyTorch 2.9 brought ROCm into its experimental wheel-variant work, making installation less awkward with compatible tooling, but the more important parts are what's happened since. …
… Intel's XPU backend has been upstreamed into mainline PyTorch, meaning any Arc GPU can run PyTorch workloads natively without a separate extension. vLLM also has an official XPU backend that supports Intel Arc GPUs, and the performance numbers can be genuinely good. …
… It's open source, built on Python and PyTorch, and was originally created in January 2023 by a developer called Comfyanonymous. …
… The practical result is that tools like Ollama, llama.cpp, and PyTorch can lean on the system's accelerators directly from inside Linux without the tax that made the WSL 2 path a compromise. …
… The install process also brings in PyTorch, which Whisper uses under the hood for running inference. …
… It is written directly against Apple’s Metal API and doesn’t depend on MLX, PyTorch, Core ML, or another intermediate framework. …
… PyTorch, TensorFlow, and a disproportionate amount of tools you'd find on GitHub assume CUDA as the backend, which affords anyone using an RTX card a relatively frictionless experience. …
… Everything that works today will continue to work, but for example, PyTorch moved compute capability 7.0 users onto the CUDA 12.6 wheel and has an open proposal to deprecate Volta outright. …
… PyTorch's default image doesn't have Blackwell sm 120 kernels built-in, so while it found my RTX 5090 fine, every CUDA call failed. …
… While ROCm has improved substantially with PyTorch 2.12 and AMD's HIP, rough edges still exist. …