News for hackers. Stuff that matters.
Re:Dissecting the Apple M1 GPU, part I by kllrnohj on Thursday January 07, 2021 @04:26PM (#25677741)
> Desktop Nvidia simply converts them to 32-bit floats up front so the performance is kept but memory is wasted

Pascal has native FP16 operations and can execute 2 FP16's at once ( https://docs.nvidia.com/cuda/pascal-tuning-guide/index.html#... )

BUT, and this is where things get fucked up, Nvidia then neutered that in the GeForce lineup because market segmentation. In fact, it's slower than FP32 operations: "GTX 1080’s FP16 instruction rate is 1/128th its FP32 instruction rate" https://www.anandtech.com/show/10325/the-nvidia-geforce-gtx-...

Re:Dissecting the Apple M1 GPU, part I by AzN1337c0d3r on Thursday January 07, 2021 @04:57PM (#25678140)
TU102 and GA102 are back up to double-rate FP16.
Re:Dissecting the Apple M1 GPU, part I by my123 on Thursday January 07, 2021 @05:07PM (#25678278)
GA102 is 1:1 FP16/FP32 rate, but with a ton of FP32 (30 Tflops for the RTX 3080).