3 ms·
> they use the hardware to its full potential If only! In fact it's a struggle to utilize a GPU to its full potential because the communication bottleneck make
by visarga 9y ago
> they use the hardware to its full potential
If only! In fact it's a struggle to utilize a GPU to its full potential because the communication bottleneck makes it infeasible. Compute is fast but data can't get there fast enough.
The authors of this paper were saying the same thing in the promo video, in fact, they were working on making GPU's more efficient. Why would they do that if GPU's are using their "full potential" already?