3 ms·
Sad to see no fractional-GPU instances. A 4xlarge is massive overkill and unaffordable for our use case.
by bryanlarsen 9y ago
Sad to see no fractional-GPU instances. A 4xlarge is massive overkill and unaffordable for our use case.
- sp332 9y ago-
- moonbug22 9y agoSo?
- shaklee3 9y agoYou can't give fractional GPU instances with this card. The K80 had two logically separate chips that were separately-addressable over PCIe. This allowed them to send two different PCIe devices to different VMs. The M60 doesn't have this. The V100 is supposed to allow time slicing to do this kind of thing, but that's not out, nor do we know how well it'll work.