Can I share cuda GPU device memory between host processes?

Question

Is it possible to have two or more linux host processes that can access the same device memory? I have two processes streaming high data rate between them and I don't want to bring the data back out of the GPU to the host in process A just to pass it to process B who will memcpy h2d back into the GPU.

Combining the multiple processes into a single process is not an option.

score 3 · Accepted Answer · answered Jul 22 '10 at 06:10

3

My understanding of the CUDA APIs is that this cannot be done. The device pointers are relative to a given CUDA context, and there's no way to share those between processes.

answered Jul 22 '10 at 06:10

Edric

23,676
2
38
40

moreover, cuda context is contained to individual threads. So you cannot even share GPU data between cpu threads. this is as far as I know – Anycorn Jul 24 '10 at 06:34
Not the answer I was hoping for, but the answer nonetheless – Mark Borgerding Aug 06 '10 at 16:32
@BrianTheLion, that forum post is about sharing *host* memory, not *device* memory. – Mark Borgerding Jun 01 '18 at 12:11
It looks like the answer may have changed in recent years according to https://stackoverflow.com/questions/19819216/interprocess-communication-cuda/19821387#19821387 – Mark Borgerding Jun 01 '18 at 12:15

Can I share cuda GPU device memory between host processes?

1 Answers1