3 ms·
Because there is latency and overhead involved when shuffling things to the GPU, and a lot of the focus on POWER these days is on OpenCAPI/NVLink. If there is a
by torpcoms 9y ago
Because there is latency and overhead involved when shuffling things to the GPU, and a lot of the focus on POWER these days is on OpenCAPI/NVLink. If there is an ARM chip fast enough to make OpenCAPI support worthwhile, I would be interested to hear about it.