这个问题是 Jason R 的后续 comment罗伯特·克罗韦拉斯回答 this original question (“一个设备的多个 CUDA 上下文 - 有什么意义吗？”):
When you say that multiple contexts cannot run concurrently, is this limited to kernel launches only, or does it refer to memory transfers as well? I have been considering a multiprocess design all on the same GPU that uses the IPC API to transfer buffers from process to process. Does this mean that effectively, only one process at a time has exclusive access to the entire GPU (not just particular SMs)? [...] How does that interplay with asynchronously-queued kernels/copies on streams in each process as far as scheduling goes?
Robert Crovella 建议在一个新问题中提出这个问题，但它从未发生过，所以让我在这里做这个。
最佳答案
Multi-Process Service是 Nvidia 的替代 CUDA 实现，它使多个进程使用相同的上下文。这例如允许来自多个进程的内核并行运行，如果它们中的每一个都不能自己填满整个 GPU。
关于cuda - 多个进程可以共享一个 CUDA 上下文吗？，我们在Stack Overflow上找到一个类似的问题： https://stackoverflow.com/questions/58747321/