3 ms·
This really isn't new. It looks like instead of treating the "Neural Engine" as an I/O device like the GPU, it's accessed as a coprocessor. It makes sense for t
by calo_star 6y ago
This really isn't new. It looks like instead of treating the "Neural Engine" as an I/O device like the GPU, it's accessed as a coprocessor. It makes sense for them to only expose their functionalities through libraries as the underlying hardware implementation will go through changes in the future. After all, does, say, Nvidia, document their GPU's ISA?
- mhh__ 6y agoYes. Not as much as the average CPU but still documented: https://developer.nvidia.com/blog/nvidia-ampere-architecture-in-depth/ https://developer.nvidia.com/blog/nvidia-ampere-architecture...
- mhh__ 6y agohttps://github.com/NVIDIA/open-gpu-doc https://github.com/NVIDIA/open-gpu-doc Also here as well
- calo_star 6y agoWhere is the native-to-hardware ISA?
- my123 6y agoInterestingly in Nvidia's case, the CUDA Toolkit ships with nvdisasm, which is a disassembler for the native hardware ISA. They do not ship an assembler though.
- userbinator 6y agonVidia is notoriously closed, a better example would be AMD/ATI --- which do have a lot of docs on their GPUs: https://developer.amd.com/resources/developer-guides-manuals/ https://developer.amd.com/resources/developer-guides-manuals...