4 ms·
While i agree that it is more difficult to distribute computational load between cpu and gpu than it aught to be, Gpu is not necessary faster for all workloads
by fleischhauf 5y ago
While i agree that it is more difficult to distribute computational load between cpu and gpu than it aught to be, Gpu is not necessary faster for all workloads than cpu, it is very specific. Most of computations can probably be executex faster on cpu.
- ledauphin 5y agoagreed, but i take the GP's argument to be that the languages aren't sufficiently expressive to allow us to communicate generically about workloads in a way that allows a system to correctly determine whether they would be faster on a CPU or not.
- andrewmcwatters 5y agoCorrect. Another good example here is that to my knowledge, developers don't do things like write directly to L1, L2, L3... they tune so that those caches are used effectively.