2 ms·
It's worth pointing out that those algorithms are also generally the ones that most benefit from being ported to a GPU, which removes the need for a shared-memo
by wtallis 14y ago
It's worth pointing out that those algorithms are also generally the ones that most benefit from being ported to a GPU, which removes the need for a shared-memory mechanism in the host-side language.
- scott_s 14y agoYes, if the latency of transferring the data and results to and back from GPU memory is low compared to the speedup.