cuCtxCreate Problem on P5000

Hi threre~

I met a problem in creating enc/dec contexts for multiple encoding/decoding tasks. On M5000, I can create more than 100 contexts while on P5000, I can only create 20+ contexts, using the same version of the driver. Does anyboday have any idea about it?

Tons of thx in advance~

I think I am seeing the same problem comparing a Tesla P4 to an M4. Creating one encoder context on the M4 consumes 38 MB, but on the P4 it goes up to 113 MB. I measured this using the NvEncoder_2013 sample app and NVSMI monitor.

Any insight about this increased GPU memory usage would be greatly welcomed!