CUDA 10.2 Graphs

Accelerated Computing CUDA CUDA Programming and Performance
1

Hello,

I am currently working on CUDA version 10.2 optimising a closed loop deep learning library. In doing this I have noticed that the method of launching new kernels for every layer.

I am thinking of implementing a graph to reduce the launch overhead of this however I cannot find any working examples of this.

Is there anyone who has a working example of graph implementation on cuda 10.2 ?

Thanks in advance.
Luca

2

There are CUDA graph sample codes. Here is one example, published with/for CUDA 10.2.

3

perfect, thank you

4

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.