Server development: Currently we use DeepSteam6.1, which requires 8 T4 GPU cards for a single CPU; How to get the best performance from multiple Gpus?
- Solution 1: Run each DeepStream plug-in on different Gpus. Gpu-id =N needs to be modified in the configuration file for allocation.
This scheme needs to pay special attention to the shared access of memory.
- Scheme 2: Directly adopt multi-process scheme, one process corresponds to one GPU; Each process is independent.
Is plan 1 or 2 feasible? Or is there a better alternative?
How is the Scheme 2 implemented in your env?
Deepstream-app project source code independent management, and then independent compilation; Execute separate execution files in different directories.
There is no update from you for a period, assuming this is not an issue anymore.
Hence we are closing this topic. If need further support, please open a new one.
Ok, we suggest to use the second Scheme. Cause one process on the card, some specialized modules are used for inferfing, and some are used for codec. And there may be memory copy problem from differenet cards. So just choose Scheme 2.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.