So I looked on a desktop browser and sure enough no black boxes… Weird.
Anyway, yes, you’re probably right. My internet connection is my bottleneck with only 24Mbps up.
My server is a 28k+ Passmark dual Xeon E5-2690 v4. So if I can get the Quadro to decode and encode then I should be good for at least 30 1080p simultaneous transcoding sessions.
So just to be clear, does PMS for Windows support Decode on nVidia cards or is it just encode? I’m hearing from other sources that decode is only supported on Linux at this time.
It’s the exact opposite. Decode on nvidia cards is only officially supported on Windows at this time and not Linux. For my setup I am using a NVIDIA P2000 but to get decode to work I have to use a dongle that simulates a monitor being connected.
Take a look at this support article from Plex.
From the article:
On Linux, hardware-accelerated decoding is not supported on NVIDIA GPUs. Intel Quick Sync is required for hardware-accelerated decoding.
Here’s a screenshot from my setup. It’s a Windows VM with a NVIDIA P2000 in passthrough mode using VMware vSphere. You can see that it’s using hardware decode and hardware encode.
Oh! Ok! Awesome! I sent my Tesla P4 back and a RTX 4000 is arriving tomorrow. But you’re saying I have to have a monitor (or something that looks like a monitor) connected for NVDEC to work?
I hope I don’t run into trouble because I have a Supermicro server that has a built in Aspeed 2400 VGA graphics chip that works in conjunction with the IPMI for remote KVM administration. I don’t want to disable that chip or I’ll loose the KVM functionality. Do you think these two devices are going to play nice or is there gonna be a conflict?
I think they’ll work fine together. For my setup I am using this DisplayPort Dongle but an HDMI one would work just as well. For some reason Plex seems to want a monitor to output video to. I was able to get Plex to use hardware decode without the dongle after a fresh install but after the first reboot it stopped working until I attached the dongle.
I want to see actual GPU use screenshots from the driver, Windows Plex can only decode with DX2A or software in windows. DX2A will show as (hw) in the Plex dash. That isn’t proof of NVDEC…
I assure you Windows PMS is indeed capable of hardware decode on NVIDIA GPUs. I have Xeon E5 v2 processors that do not support Intel Quick Sync or DirectX (DXVA2). Besides, back on Feb 6th in this very thread @tobiashieta pointed out that the transcoder team fixed a bug specifically for Nvidia decode. I’m not sure why they haven’t updated the bottom of the support article to mention Nvidia decode but it does work. The support article also suggests that hardware acceleration is not possible inside a VM but they do not cover hardware passthrough which is what I’m using.
When 13 streams were being played on my server tonight the decode queue was hovering around 40% to 50% and the encode queue around 20% according to task manager. The 10 CPU cores were around 30% used because audio and one unsupported video format still required software transcoding.
actually plex does not support NVDEC on windows
it supports DX2A which still uses nvidia for hardware decoding.
semantics sure.
either way, windows is currently the only supported way to get decode and encode for nvidia.
it is stated @ https://support.plex.tv/articles/115002178853-using-hardware-accelerated-streaming/ in the tech spec section at the bottom of the page.
since it doesn’t format well when pasted here, you will just have to follow the link and scroll down.
as for the dummy plug/monitor, I believe this is required because without a monitor the hardware goes into a low powered sleep state that apparently does not expose or wake up for decode/encode function.
I just installed the RTX 4000 and as stated, without the dummy plug, no decode.
Which confirms what @barnitos said, this isn’t NVDEC. A monitor (or dummy monitor) wouldn’t be needed for NVDEC. I wonder if Plex is working on this? Or did they make an executive decision to use direct X instead because there might be a bit more control over the video rendering quality via Direct X?
dummy plug is definitely needed, I have a p400 on linux which doesn’t work without one.
I don’t think it has anything to do with OS or with nvdec/nvdec vs dx2a.
the OS needs an active display for it to render/decode/encode.
you can continue to fight it or be mad, but if you want it to work then you will need to get a hdmi or displayport dummy plug (or monitor) like everyone else.
I’ve got two on order. I just didn’t know I needed it until 10pm last night. So it’ll be a couple of days before it gets here.
But I’m pretty confident that it’s only needed for DX2A. The Tesla cards support NVDEC/NVENC and they have zero monitor outputs.
I would think your experience so far would prove you wrong, at least as far as plex is concerned.
And my experience tells me that I needed one.
So, you are free to continue to experiment and try to get or keep it working without a dongle, maybe you will figure something out that the rest of us have not.
Uh… what? I’m confused how your confused about what I said…
Maybe you should re-read my post.
now I am too.
you said you ordered some dongles, ok good, you need them.
then you said you were confident it (which I can only assume means the dongles you previously referred to) is only needed on dx2a (windows).
which doesn’t seem to be the case, as dongles or monitors are needed for both windows and linux (which uses nvdec/nvenc instead of dx2a).
and since tesla cards don’t have a video out, and a monitor/dongle appears to be required for both window plex and linux plex, it seems like tesla cards simply won’t be suitable for plex usage, regardless of their support for nvdec/nvenc.
Ok. Well I wasn’t aware that the Linux users had to use them as well. I’ve only talked here to Windows users.
Then it doesn’t make sense why nVidia would market the Tesla cards for this use knowing they don’t have any monitor outputs and the chipset requires a monitor to function.
because tesla cards aren’t really designed for video decoding/encoding, and more intended for scaling out compute resources for AI and other high end software that does not need a video interface ?
It would be awesome if gpu acceleration worked without dongles/monitors, or even without non-video-out hardware such as tesla cards, and maybe someday that will be fixed or figured out.
I really don’t think very many people even have access to, or want to spend the money on, tesla cards for use with plex to even experiment much.
as mentioned previously and many other places, a p2000 quadro (or 10 series gtx with modded drivers) will solve most problems for most people for generally significantly less money and effort.
Yeah. In 99% of cases a Pascal card is adequate. I intentionally went with the best Turing card for this operation just because I buy equipment based on a 10 year service life. Obviously it probably won’t last that long but that’s always the goal.
According to nVidia, the Turing architecture encodes at a higher quality and with a 15% savings in bandwidth. Plus it supports more future codecs. Not that I’m going to use anything other than 1080 H.264.
I don’t believe it until I see documentation. Any GPU that supports nvdec will do dx
