I have a very large computation graph with approximately 70,000 vertices, with both GPU and CPU workloads, which I’m considering to execute with CUDA Graph.
I’m curious about CUDA Graph’s capability of processing graphs of this scale, and I hope someone can enlighten me with this so that I can make the design decision to use CUDA Graph or not.
我有一张很大的计算图考虑用CUDA Graph执行，它包含大约7万个顶点，每个顶点可能是GPU或者CPU任务。
我想知道CUDA Graph处理这种规模的图的能力如何，我希望能有人能够帮助我了解这点，这样我可以决定是否要用CUDA Graph实现我的需求。