How the nvenc's performance of Quadro P2000 compared to Quadro K2200?

We are using NVENC in a big realtime streaming solution where having more than two encoding sessions.
In https://devtalk.nvidia.com/default/topic/800942/gpu-accelerated-libraries/session-count-limitation-for-nvenc-no-maxwell-gpus-with-2-nevenc-sessions-/post/4562315/#4562315 , Eric_Young says: If you want Maxwell GPUs with the best NVENC performance, K2200 and M6000 are the recommended products.

M6000 is too expensive, the price of P2000 is about the same as K2200.

In https://developer.nvidia.com/nvidia-video-codec-sdk, the nvenc’s performacnce of P2000 is bether than K2200.

Does anyone verified in reality that Quadro K2200/P2000, which one should we buy?

I recommend P2000 (but check support issues like passthrough in virtualization environment (see https://devtalk.nvidia.com/default/topic/992447)). The Pascal generation (P2000) has also HEVC encoder (not supported by Maxwell Gen 1 (K2200)) (see https://developer.nvidia.com/video-encode-decode-gpu-support-matrix). P2000 has also more RAM (5GB vs. 4GB) (see memory allocation problems with multiple encoder streams https://gridforums.nvidia.com/default/topic/1310/#4789).
My K2200 has max core clock 1124 Mhz (reported by nvidia-smi) and my P2000 has max core clock 1721 (reported by nvidia-smi (after “nvidia-smi -ac 3504,1721”)) and if you read “NVENC_Application_Note.pdf” you get expected performance metrics (but reality depends on many factors):

Thanks very much. Do you have any idea to resolve the issue of Quadro P2000 passthrough in virtualization environment?

Hi mcerveny,
What tool did you use to verify the results from Nvidia (number of encoded frames)
I want to use it to benchmark M4000 and compare with P4000 with non 16/9 ratio streams.
Thanks in advance,