ππ£π¨ ππ ππ£π¨
CPUs time-share thousands of times a second. GPUs can't: a running kernel holds the card until it's done.
That's why Kubernetes still hands out whole GPUs. Daniele Polencic & Mesut Oezdil explain. Full clip: youtu.be/uGuPkCuTqFc
#GPU #Kubernetes
CPUs time-share thousands of times a second. GPUs can't: a running kernel holds the card until it's done.
That's why Kubernetes still hands out whole GPUs. Daniele Polencic & Mesut Oezdil explain. Full clip: youtu.be/uGuPkCuTqFc
#GPU #Kubernetes
1:14