Number of concurrent transcode sessions on Quadro RTX4000 @1080p 30FPS


(Video Encode and Decode GPU Support Matrix [NEW] | NVIDIA Developer) indicates that unlimited NVENC sessions are possible on Quadro RTX4000. However, for any practical use, one should know the possible throughput (of course, under certain known conditions) that can be achieved even before we buy. We couldn’t find enough data on the performance. For example, if someone could point out to performance, e.g., how many parallel streams can be transcoded for streams of 1080p, 30FPS (or any similar type of video streaming application), it will be helpful.


Any inputs, NVIDIA?

RTX 4000 has an NVENC which is similar to that in T4. The benchmarks for T4 are published on NVIDIA Video Codec SDK developer web site. The quality of encoding will be similar to that on T4, but performance will depend on clocks, availability and speed of memory and efficiency of the application. The benchmarking assumptions used by NVIDIA are given on the above site.

Thanks! This is helpful.
I posted this to CUDA forum and I received another input:


Any idea on the amount of memory required to encode one 1080p stream (of high Quality H264, 5Mbits/sec) (@ 30FPS)?

Hi everyone,

The updated matrix can be found here:
TIP: Type “rtx” or “nvidia” in the search field above the table to find the new cards.

NOTE: Info on the the A100 will be added at a later date.