My case is M60 chip, HEVC, 30 fps video and the low-latency high quality preset.
According to Table 3 the encoding frame rate is 200 (HD, 8-bpp), thus i can simultaneously encode at most 6 streams per NVENC (i am aware that GPU and other factors would decrease that number).
On the other hand with a single NVENC (M60), i measured that the encoding intervals per HD frame are in the range of 10ms-12ms. So, roughly speaking, the encoding frame rate is 100 fps. i excluded the impact of clocks by disabling auto boosting by setting GPU and memory access clocks to maximum:
.\nvidia-smi -ac “2505,1177”
The discrepancy of the factor x2 is annoying to me. Perhaps, in order to get 200 fps NVIDIA QA team used a restricted search area size, low bitrate, single reference frame, simple video content etc.