News for hackers. Stuff that matters.

A comment on YOLOv5 on CPUs: Sparsifying to Achieve GPU-Level Performance. Parent comment

Re:YOLOv5 on CPUs: Sparsifying to Achieve GPU-Level Performance by deepnotderp on Friday September 10, 2021 @05:27PM (#28487084)
V100 GPUs have non tensor core fp16 operations too I think
[ Parent | on HN ]
Re:YOLOv5 on CPUs: Sparsifying to Achieve GPU-Level Performance by woadwarrior01 on Friday September 10, 2021 @06:07PM (#28487447)
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...

[ Parent | on HN ]
Re:YOLOv5 on CPUs: Sparsifying to Achieve GPU-Level Performance by ml_hardware on Friday September 10, 2021 @05:48PM (#28487272)
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.
[ Parent | on HN ]