Is there any documentation available on how many simultaneous transcodes are possible when using Hardware Transcoding?
Historically, you could roughly work this out using CPU Passmark scores- but with Hardware Transcoding - it seems to be a bit of an unknown.
For example - can you get more hardware transcodes out of an i9-9900k than say a Nvidia P2000?
Also - does Plex take advantage of multiple GPU’s for Hardware Transcoding? For example- if I have 2 P2000’s running, will I got more hardware transcodes?
I feel this part of Plex documentation is severely lacking and any further insights into this would be useful for many users.
I don’t know. I don’t think plex explicitly supports either 2 or 3.
Might it work for some people, may be, maybe not.
Consider, there are people with p2000, which a lot of people love and have and works great for them, while others can’t get a p2k to work to save their life.
Whether its a motherboard compatibility problem or a user problem or they simply have a bad p2k, who knows.
Plex just did a major transcoder update recently, which finally allows HW decoding on linux/nvidia.
Whereas intel decoding/encoding has been supported on both linux and windows for a long time.
There are a large number of threads here with people who have problems with HW transcoding, and those who don’t.
HW transcoding by itself with FFMPEG may (or may not) be simple, but plex doesn’t use a plain FFMPEG and has their own customized implementation for streaming.
The Intel GPU’s will use up to half your system ram if available. You generically said transcode, h.264 and h.265 are worlds apart in computational requirements.
If the GPU is too busy it falls back to CPU, it does not cycle through available GPU’s, i.e. Nvidia > Intel > CPU. Since Local-Copy was just added to the transcoder it would be interesting to see if two Nvidia’s bridged together would be utilized.
Thanks @pl_5309 - do we know how much RAM a h264 and h265 transcode consumes?
It’s good to know hardware Transcoding uses system RAM, but without knowing how much RAM is used for each transcode, we still don’t know how many hardware transcodes can be done by an Intel Quick Sync CPU.
And is the RAM used dependent on the version of Quick Sync?
Anyone from Plex able to provide an official response to this query? With so much focus on Hardware Transcoding of late, the official info on its capability is quite scarce.
How much ram is dependant on the file being processed, 100-300 MB for h264 and 900-1200 MB for h.265. Network and Disk start to be a bottleneck 6-8 high bit rate h.265. chrisallen is the employee doing the transcoder work.
That would be extremely optimistic and ignore all onboard system bandwidth limitations. Not sure what you are after but I would be comfortable with 6-8 concurrent h.265 transcodes. If you want to go beyond that you would need to move to coreX with something like an RTX3000.
Sure I get it. I definitely have no need for 100 transcodes. But I think this discussion and the information you’ve provided does highlight a real gap in Plex documentation that I think should be explored
I have a different question than about number of streams.
I have an amd cpu, and rarely need more than 2-3 streams.
I also take most of my media from ripping blueray with MakeMKV. I then want convert to a smaller file size using xmedia or handbrake. This takes a long time. For one TV season, it took 34 hours.
Would a GPU like the p 2000 speed this up significantly?
Also, when I use live TV, the programs buffer a lot. Would a GPU at least reduce the buffering?
this only applies to intel, and probably only on windows. I don’t know whether the linux intel drivers also overflow to system ram or not, but linux and windows have very different system memory architectures.
to repeat myself, nvidia on linux uses video ram (only), and a 4k hevc > 1080p x264 conversion uses about ~1.4 gig of video ram.
