The GPU and ML libraries, frameworks, abstractions, and drivers have long since felt to me like the famed "Linux audio jungle" of old. At least I understand why ffmpeg, gstreamer, OSS, ALSA, PulseAudio, and PipeWire all exist.<p>CUDA, OpenCL, Vulkan, Metal, PyTorch, TensorFlow, I don't understand. They all build on CUDA and AMD / Intel / CPUs are treated as 2nd-class citizens, right?