Hacker News new | ask | show | jobs
by deepnotderp 1739 days ago
V100 GPUs have non tensor core fp16 operations too I think
2 comments
woadwarrior01 1739 days ago
Yes. Non tensor core fp16 ops are the default. Tensor cores are essentially 4x4 fp16 mac units and there's a requirement that matrix dimensions are multiples of 8[1] that needs to be met for them to be used.

[1]: https://docs.nvidia.com/deeplearning/performance/mixed-preci...

link
ml_hardware 1739 days ago
That's true.. in fact, seeing V100 FP16 < T4 FP16 makes me believe you're right, the V100 should be much faster if the tensor cores were being used.
link