I am currently evaluating the costs of transcoding several livestreams in EC2. As far as I can tell only the recent g2.2xlarge instance type supports hardware accelerated transcoding (using NVENC). Intel Quick Sync and CUDA (on Linux) are not supported as far as I know. Has anyone tested how many parallel streams can be handled (when transcoding to 2-3 lower bitrates) by a single instance? I am looking at streams with a resolution of 640x480 with about 1.2MBps. Also are there any numbers on the more recent EC2 types c3.large, c3.xlarge, c3.2xlarge, etc. (the benchmarks provided on the transcoder forum are quite dated)?
As a practical matter, I've found that it's possible to do about 1 incoming transcode with 1 or maybe 2 output levels on c3.large. Not bad if you don't have a heavy transcode load or are only doing audio, but don't go too crazy with it.