Apple Neural Engine Internal: From ML Algorithm to HW Registers

Re:Apple Neural Engine Internal: From ML Algorithm to HW Reg... by dannyw on Friday December 10, 2021 @12:27AM (#29506727)

Dedicated neural net silicon is faster than GPU, at the cost of being less general purpose.

NVIDIA GPUs for example dedicate a modest fraction of their die to Tensor cores, at the cost of less general-purpose CUDA cores.

[ Parent | on HN ]

Re:Apple Neural Engine Internal: From ML Algorithm to HW Reg... by martinald on Friday December 10, 2021 @08:31AM (#29509589)

Thanks. I didn't realise how much faster. Probably you can do 10x the FLOPs in similar die space on dedicated silicion vs GPU. I thought it would be faster, but not an order of magnitude more.

[ Parent | on HN ]