Hardware requirements/suggestions around dedicated GPU? (Nvidia Quadro P2000)

I read somewhere that both NVDEC and NVENC will be supported by Plex soon. Does that mean the whole transcoding pipeline can be handled by an Nvidia GPU or do I still need a CPU with Quick Sync support?

I basically want to create a headless PXE thin-client build around an Nvidia Quadro P2000 and it should be able to transcode multiple 4K 10bit/HDR streams. I am considering getting an i3-9100F or i5-9400F because of the single thread performance but I’m not sure what I need and I don’t want to introduce bottlenecks.

All the storage will be on a NAS and connect over 10 Gbit ethernet.

Suggestions?

nVidia codecs are only needed if you’re using the card.

QSV is already available in the Intel processors up through -8xxx today.

The -9xxx processor support is now in Forum Beta.

If you need strong subtitle support (burning) and/or audio, go with the higher thread speed CPUs.

Yes, but do I need QSV / integrated graphics after the NVDEC-NVENC feature goes public? I already have a P2000 card because I replaced it. So I want to create a Plex Media Server build around that, but I don’t want to introduce bottlenecks.

Maybe you misread the question but I’m talking about the NVDEC and NVENC encoding features on Nvidia cards, not the codecs.

^ I read this

The first thing I thought about is your audio. Video transcoding is happening in a lot of hardware, but who’s paying for the DTS and AC3 licenses to do it in hardware? I think Apple does in their new iPads, because they convert 4k entirely within themselves. Does NVIDIA plan to support that?

If not, like Chuck said, a higher speed CPU is always nice.

for linux transcoding, video ram is the most important thing.

Plex pays for the codec licensing. This is part of your base server.

Subtitle transcoding will be done in the CPU until such time as there is a way to burn the subtitles on the GPU (either QSV or nVidia GPU).

Audio will always be done in the CPU.

Not quite the most important. :wink:

Sufficient memory available on the device is important however having sufficient ALU’s in the device is actually the most important factor. These are what govern image quality and processing speed (how many streams it can handle concurrently).

ok, let me qualify that, for 4k transcoding on linux/nvidia gpu, video ram will run out before anything else.

@ ~1.4 gig vram per 4k transcode.

Care to share how you did that? One of us made a math error :wink:

Ah, okay. Well, it sounds like I will be pre-transcoding everything then… Not my favourite solution but like @TeknoJunky said

If you don’t have the storage space for a copy of both 4k and 1080/720, then perhaps you should not even be collecting 4k.

Is there any solid solution for converting 4k HDR to 1080p SDR?

finally see someone else research @ nVidia Hardware Transcoding Calculator for Plex Estimates

in particular the section about 4k transcoding vram requirements

for reference, all my own screenshots are from full 4k bluray rips made myself via mkv to max profile ios clients (typically iphone 7+ and ipad mini 4.

I can only assume that others that are getting ~1gig vram usage are doing 4k > 720p or lower bitrate 1080p (or the source 4k is not a full 4k bluray remux), or there is something specifically different between their system and my system which causes different vram usage.

Thank you for those charts. I will rerun the math but I do come out with higher memory requirements as I don’t consider transcoding to 720p 4Mbps a watchable size and rate (imho)

if 1GB for 720p target, 4GB for 1080p target will be required. (in my head)

the ~1.4g video ram I get is plex transcoding 4k hevc main 10 > 1080p

transcoding hevc 1080/720 to 1080/720p x264 will be much lesser vram use

Techno,
What are some of the bitrates (input / output). This will give everyone a sense of how much memory is used knowing higher bitrates need more memory to represent all the bits (less macro blocking)

This topic was automatically closed 90 days after the last reply. New replies are no longer allowed.