2 ms·
Fast low latency interconnects, and scale up versions of the resources (eg lots of gpus per box) to minimise communication overhead.
by justincormack 9y ago
Fast low latency interconnects, and scale up versions of the resources (eg lots of gpus per box) to minimise communication overhead.