Before you run off and get Nvidia, take serious consideration of the Intel ARC line. They’re relatively cheap and have great transcoding performance. They’re supported by Linux out of the box, and I had no problem getting docker passthrough enabled. Unlike Nvidia the drivers don’t have built-in limits for how many simultaneous streams you can transcode.
I have an A380, but I bet an A310 would also do the job fine.
I’ve never actually tested the performance of simultaneous transcodes. However, my server generally sees 2-5 active users on a busy night, and nobody has complained about buffering so far.