GPU Passthrough on Linux and Docker for AI, ML, and Plex

Подписаться 235 тыс.

Просмотров 46 тыс.

50% 1

We've already figured out how to pass through a GPU to Windows machine but why let Windows have all the fun? Today, we do it on an Ubuntu headless server that's virtualized, run some AI and Deep Learning workloads, then turn up the transcoding on Plex to 11.
Video Notes: technotim.live/posts/gpu-pass...
📦 Cards I mention that don't require external power
GTX 1650 amzn.to/3dnG31y
GTX 1050 amzn.to/2GQlOgO
(Affiliate links may be included in this description. I may receive a small commission at no cost to you.)
Video mentioned:
• Remote Gaming! (and Video Encoding using Proxmox and GPU Passthrough)
• Remote Gaming! (and Vi...
• 4 Ways to Install Plex (one is unexpected)
• 4 Ways to Install Plex...
Support me on Patreon: / technotim
Sponsor me on GitHub: github.com/sponsors/timothyst...
Subscribe on Twitch: / technotim
Become a RU-vid member: / @technotim
Merch Shop: l.technotim.live/shop
Gear Recommendations: l.technotim.live/gear
Get Help in Our Discord Community: l.technotim.live/discord
2nd channel: / @technotimtalks
(Affiliate links may be included in this description. I may receive a small commission at no cost to you.)
00:00 - Intro
00:47 - Why use a GPU with Docker & Kubernetes?
01:43 - What are we going to do today?
02:17 - Passing through an NVIDIA card with Proxmox
02:55 - Add video card and modify config
03:36 - Install NVIDIA drivers on an Ubuntu headless server
04:25 - Install Docker support for NVIDIA
05:19 - Check NVIDIA driver to make sure it is exposed to Docker
06:09 - Install nvtop to measure and monitor our GPU
06:27 - Launch Deep learning TensorFlow workload
07:30 - Set up Plex using Kubernetes, Docker, Rancher
07:43 - NVIDIA with Rancher and Kubernetes
08:51 - Plex Hardware Accelerated Transcoding in Docker
09:13 - Transcode 4k video with NVENC in Docker
11:00 - Would you ever use this?
11:31 - Stream Highlight -The most unexpected follow yet!
#Homelab #Docker #GPUPassthrough #Kubernetes #Rancher #TechnoTim #Proxmox #Virtualization
#Plex
"Big Buck Bunny" which appears in this video is licensed under the Creative Commons Attribution 3.0
peach.blender.org/about/
Thank you for watching!

Наука

Опубликовано:

31 июл 2024

Ссылка:

Скачать:

Готовим ссылку...

Добавить в:

Мой плейлист

Посмотреть позже

Комментарии : 168

@TechnoTim 3 года назад

Have you tried passing though a video card yet to your Linux server? btw, if you're new here welcome! Be sure to subscribe for more content like this!

@boxedowl 3 года назад

Yes and it worked beautifully after your tutorial made me realize I had 1 tiny part of my config file wrong despite browsing most of the interwebz. Also are you psychic? It's almost like you knew I was trying to use Ansible and Docker for GPU loads right now.

@jesuskingofkings 3 года назад

Hi Tim. Do you have the Link to download the kubernets P4000 Nvidia Quadro drivers!? Thank you!

@jamesm5192 3 года назад

Any ideas why we can't have a DaVince Resolve container for any Linux distro that supports AMD GPU acceleration? Using a special distro just to use DaVince pretty much kills the magic of OSS.

@ml.2770 4 месяца назад

3 yrs later. Is this all still valid? I've noticed Portainer adding gpu support?

@Mikesco3 3 года назад

I love how he presents information: concise, well structured and almost predicts your questions. Turning really a complicated head-banger subject into making it really simple distillation of research.

@TechnoTim 3 года назад

Thank you so much!

@unstoppablemango2370 3 года назад

I've had this on my backlog to learn for a while, and I've admittedly been putting it off because it's not a quick "sit down and pickup" kinda thing. This checks all the boxes for me! Thanks for such an amazing easily digestible tutorial! Also, this may be just my opinion but I really appreciate you don't hold the viewers hand through the entire thing. So many tutorials feel the need to stop and explain what "ssh into the machine" means and suddenly we're a quarter of the way through the video and haven't even touched on the content yet. There is a skill level that exists between beginner and advanced and they often go neglected.

@TechnoTim 3 года назад

UnstoppableMango thank you so much! It’s a balancing act!

@berndbrater9958 3 года назад

Thanks alot for that, was quiet struggeling with that. Keep up the good work ❤️

@TechnoTim 3 года назад

Bernd Bräter thank you!

@fanuchman 3 года назад

Second time I've come across your video. Second time I've found the contents hugely useful. Subbed!

@TechnoTim 3 года назад

Awesome, thank you!

@remythegreat11 3 года назад

broooooo!! i had done everything right except editing the deamon.json because i didnt know that needed to be done. now everything is working! thanks again fam!

@TechnoTim 3 года назад

Glad I could help

@DavidJeffreyOConnor 3 года назад

Your videos are fantastic, thank you. I'm still a newbie, but I'm learning more from you than I am in some of my college coursework...

@TechnoTim 3 года назад

David O thank you!

@dustinkrejci6142 3 года назад

👨‍🏫 professor Tim, class is in session.

@pabloescobar9337 2 года назад

For those looking for the tensorflow test script: python -m tensorflow.models.image.mnist.convolutional

@tomgrey3046 Год назад

Hi Tim, since this material is a bit old now, and you have shown us in the meantime your new 11th gen NUC with integrated GPU, maybe it would be a great opportunity to show how to passthrough integrated GPU (iGPU) to Proxmox LXC (and perhaps docker nested in LXC). ML/AI apps plex/jellyfin are even more popular nowadays, and even older NUC-like devices could perform such tasks pretty well (ie. 6-12th gen Intel). This differs from power-hungry servers with separate GPU, and refers to Intel specifically (however AMD users are interested in it, too). Keep up the good work!

@navinderdhaliwal9950 Год назад

Interested in this also! Hope to see a guide, any luck your end?

@AGG88100 3 года назад

This is awesome!! Thanks for sharing !

@TechnoTim 3 года назад

Agustin Gimeno thank you!

@yhua122 3 года назад

Nice video! your video keep me pushing my servers to their limit.

@TechnoTim 3 года назад

ken huang 🔥 😉

@jartran 2 года назад

I can't seem to get this to work on Portainer properly. Any tips?

@craven4800 3 года назад

Nice! I was thinking about looking into this myself

@TechnoTim 3 года назад

Craven awesome man! Let mw know how it goes!

@aleg_911 3 года назад

Nice video, really helpful.

@TechnoTim 3 года назад

AleG thank you!

@gaby1491 3 года назад

If you are having problems with the NVIDIA drivers on older cards pass the rom and check all functions on the pci prompt. This worked for me on a k2000 Quadro x4 setup for my wife. Also this is great video sir congrats on your almost 15k subscribers

@TechnoTim 3 года назад

gaby1491 thank you!

@nicholasboccio 3 года назад

How perfect... I have been struggling to use my gpu with my DeepQuestAi:Deepstack:Gpu workload. I see several things that you've shown that give me some new hope. Thank you

@TechnoTim 3 года назад

Nice! Thank you!

@ChrisHalden007 3 года назад

Great video. Thank you!

@TechnoTim 3 года назад

You are welcome!

@dkin7922 3 года назад

Tim, thank you for doing this video. I'm working on setting up some containers and would love to passthrough to two RTX 2080 supers. I know they're on the commercial side and I've had a heck of a time getting them to passthrough with the traditional esxi route. I hope this is going to allow me to utilize them. Again, thank you. Keep up the fantastic videos.

@TechnoTim 3 года назад

Thank you!

@ryanbell85 2 года назад

For Ubuntu 22.04, in the grub file, you will have to replace "video=efifb:eek:ff" with "initcall_blacklist=sysfb_init"

@DinkleDooper Год назад

Definitely wouldn’t mind an update. This gets me close but still not able to get it working. Appreciate the information

@syrrysaver2775 3 года назад

OMG yes. I should have seen this coming; it was the answer to a question I couldn't yet formulate ...

@TechnoTim 3 года назад

Glad it helped!

@kristoffseisler2163 2 года назад

how do you do it in normal docker for windows or at the command line for Linux though?

@techlordgyan9961 3 года назад

You make unique content kepp it up

@TechnoTim 3 года назад

Thank you so much!

@jarekmor 3 года назад

Hi Tim, another great video. Thank you very much. One question. Do you know why nvtop installation removes these following packages: libnvidia-compute-460 libnvidia-decode-460 libnvidia-encode-460 nvidia-compute-utils-460 nvidia-headless-460 nvidia-headless-no-dkms-460 nvidia-utils-460. After it all nvidia drivers are deleted and nvidia-smi does not work.

@rockercoolification Год назад

Hi Whats the command to run AI training script in tensorflow docker image?

@mrrogers4591 3 года назад

Another great video Tim. Every Saturday morning I look forward to what you put out to teach us something new. Which GPU to buy that is less than $200?

@LampJustin 3 года назад

Buy a Intel processor for your Nas and profit from QuickSync it's worth it, trust me!

@TechnoTim 3 года назад

Geoff Rogers thank you so much! See the description for a few budget ones! 1660s are great but a little over 200

@peny1981 2 года назад

What should be the newest line to install nvidia drivers for rtx 3050? I have errors for 450 and 510.

@ebiscaia 6 месяцев назад

Hi Tim, have ever had a look at nvidia gpu operator? Also, any thoughts of others kubernetes distros like k0s and microk8s?

@Shane-Singleton 3 года назад

Glad you were able to get this working and put together a video :) I don't have a use for this yet.. But It may be something fun to lab.

@TechnoTim 3 года назад

Shane Singleton thank you!

@plu2 8 месяцев назад

Hey Tim, love your content, it's so very helpfull. I just worked my through this video and there's quite a bit of outdated stuff in after three years. Maybe consider making an update?

@weizige8502 Год назад

What is the test graphics card command?

@jart122 2 года назад

I found the answer to my question on another one of your videos.

@ewenchan1239 5 месяцев назад

Three things: 1) I just recently figured out how to allow multiple LXC containers to be able to share a single RTX A2000 on my test system, so yay! (But sharing it between VMs and LXC containers -- that's a no go). 2) I've also recently shared a single RTX 3090 between 4 Windows 10 gaming systems and using Parsec instead of remote desktop to remote into said gaming VMs. (I tried remote desktop, Teamviewer, Sunshine/Moonlight, and Parsec and Parsec, so far, as been the most usable out of these four options. You can remote in with Remote Desktop and Teamviewer, but the inputs gets messed up when you're trying to play Minecraft or Halo Infinite.) 3) My own personal take is that I am sharing the GPU between LXC containers because it is easier to ssh into a LXC container than it is to ssh into a Docker container.

@itsathejoey Год назад

Ran the lshw -c video on my openmediavaultserver and it told me command not found.

@carsenik7143 2 года назад

My k8s cluster has multiple proxmox vms as nodes, so should i add the gpu to the main vm that has rancher/k8s cluster or add it to the vm worker node that has my plex container on it?

@TechnoTim 2 года назад

to the one where the gpu workload will run, e.g. plex

@maydonozluyumurta9420 3 года назад

Waiting for the nextcloud - rancher video impatiently 👀

@TechnoTim 3 года назад

You and me both!

@craven4800 3 года назад

So i noticed that at one point you are running 8 transcodes off a consumer gpu. It is my understanding that consumer cards are locked at 2 simultaneous transcodes. Did you have to patch the drivers for your card or how did you get around this?

@constantinamundsen2487 3 года назад

you can do this with docker. Normally, you can only do one workload at a time, but since its all running on docker, it appears as one workload to the gpu. Thats a really nice alternative to buying tegra or quadro cards.

@gaby1491 3 года назад

@@constantinamundsen2487 I have quadro cards but my setup preceeds docker lol it would have been nice to be able to use this a few years ago

@TechnoTim 3 года назад

Craven i’ll do some more checking and report back! I was able to transcode 4 just fine!

@eatinbass Месяц назад

If someone reads this in 2024 - will this way of setting up passthrough enable power management for the GPU? In Tim's nvidia-smi it's showing P0..

@MrEric377 3 года назад

I got a bingo! with AMD 3400G and a Quadro p400. So now I have a bare bones with proxmox which is running Freenas and a linux VM running docker/rancher that has plex configured to run HW transcoding. Can it get any better than this?

@TechnoTim 3 года назад

B I N G O! Nice work!

@khalobert1588 3 года назад

Is it possible to assign the same GPU to one WIndows, and one Linux VM? Not to run them in parallel, but actually one at a time while the other one is turned off.

@TechnoTim 3 года назад

Yup!

@LukasDehne 3 года назад

Which software do you use for your terminal session on windows?

@TechnoTim 3 года назад

See this ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-kL8iGErULiw.html

@fabiongo123 3 года назад

This is awesome Tim. Do you have any guides for AMD cards?

@LampJustin 3 года назад

Wonder how good the Navi cards r

@LampJustin 3 года назад

The Intel HD integrated GPUs with Quick sync are also really amazing. On a cheap i3 with a UHD you get more then 15 1080p streams if I recall correctly. It's so much better then buying an Nvidia card.

@TechnoTim 3 года назад

Fabio Gomez Diaz not yet but AMD also supports this!

@ABUNDANCEandBEYONDATHLETE 3 года назад

@@TechnoTim I built a Zen2 3970x, looking to start my channel NetworkNickOfficial (no content yet). Would you want to collab on an AMD CPU/GPU and Encoding setup?

@momendo 3 года назад

To fix the error where nvidia-smi command fails and dmesg reports this error: [ 20.054078] NVRM: GPU 0000:01:00.0: RmInitAdapter failed! (0x23:0x56:624) [ 20.054173] NVRM: GPU 0000:01:00.0: rm_init_adapter failed, device minor number 0 [ 20.069879] NVRM: GPU 0000:01:00.0: RmInitAdapter failed! (0x23:0x56:624) [ 20.070010] NVRM: GPU 0000:01:00.0: rm_init_adapter failed, device minor number 0 [ 20.089595] NVRM: GPU 0000:03:00.0: RmInitAdapter failed! (0x23:0x56:624) [ 20.089780] NVRM: GPU 0000:03:00.0: rm_init_adapter failed, device minor number 1 [ 21.616934] NVRM: GPU 0000:03:00.0: RmInitAdapter failed! (0x23:0x56:624) [ 21.617023] NVRM: GPU 0000:03:00.0: rm_init_adapter failed, device minor number 1 In host server, edit /etc/pve/qemu-server/100.conf and add this line: cpu: host,hidden=1,flags=+pcid

@krazypeople4 2 года назад

Jellyfin is an opensource project and doesn't charge you to use your nvidia drivers.

@Jay.M 3 года назад

Why do you have so many VMs instead of using containers within Proxmox natively? It would reduce your overheads

@TechnoTim 3 года назад

I containerize everything I possibly can. I have almost 60 containers running that you don't see in Kubernetes. Some of those VMs are just for testing.

@Dexter101x 3 года назад

I want to know if I could use my server that has a gpu, as a supercomputer

@TechnoTim 3 года назад

Sure, give it a shot!

@firebrickfilmstv 3 года назад

Any ideas for depowering the GPU when the VM is switched off? :)

@TechnoTim 3 года назад

Hey, not sure if it's possible! If you run nvtop, you can see the GPU and RAM clocks down quite a bit when not in use.

@IIAmrasII 3 года назад

I'm new to all this and was wondering what the full tensorflow script is? Part of it is hidden behind the webcam and the script isn't on your github page for this video. Wanting to verify that the gpu is being passed through.

@TechnoTim 3 года назад

Thanks! Check out the docs in the description. It has links to everything!

@williamall6130 2 года назад

I know this is really old, but I had the same problem and it seems the script he ran that was blocked by the camera was for MNIST training. That Script is: python -m tensorflow.models.image.mnist.convolutional

@IIAmrasII 2 года назад

@@williamall6130 Thank you!

@user-cc4es1hf5p 3 года назад

is this available on Rhel or Arch linux? Btw thanks for good video!

@TechnoTim 3 года назад

It should.. Thank you!

@jasonperry6046 3 года назад

So.... I have been thinking about building a system for just this reason for awhile. Here is my thing, I want power efficiency, I want scalability, and I want it automated. I am not home all the time and don't want to leave a powerful machine running in my basement. Have you done a video covering this? Is it possible to leave a efficient rig running and have others hibernating waiting for a work load?

@TechnoTim 3 года назад

Jason Perry soon!

@jasonperry6046 3 года назад

@@TechnoTim well.... now I have the waiting for Christmas feeling. cant wait, thanks.

@techlordgyan9961 3 года назад

Is windows worth it for coding cause I have installed Ubuntu in my virtual box

@TechnoTim 3 года назад

I do 50% of my coding on Mac and the other 50% on Windows. I think Linux is great too. If you use WSL on Windows, it gives you a Linux like experience in Windows.. Check this out for how to set it up. ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-kL8iGErULiw.html

@camerontgore 3 года назад

So how many of these hoops do you have to jump through if you're using an amd gpu?

@constantinamundsen2487 3 года назад

amd drivers are more open, it shoul dbe roughly the same, but maybe easier to get the drivers.

@camerontgore 3 года назад

@@constantinamundsen2487 Awesome! Thanks

@constantinamundsen2487 3 года назад

@@camerontgore but i want to warn you that many gpu server applications rely on cuda cores. Id say theres more applications for nvidia than for amd cards in a server environment as nvidia is also leading the datacenter gpu market.

@TechnoTim 3 года назад

Cameron Gore agreed! About the same!

@Dainis_M 3 года назад

Hi, Could you please make a video about Wireguard installation in Rancher

@TechnoTim 3 года назад

Possibly!

@dec0077sa 3 года назад

Instead of passing through an entire GPU to a VM, can you do the same with a vGPU and still use it for docker/kubernetes?

@TechnoTim 3 года назад

If you have a vGPU, I don't see why this wouldn't work!

@dec0077sa 3 года назад

@@TechnoTim Well, the drivers for vGPUs are not the same series as consumer Nvidia cards or even those for "normal" Quadros, so I don't know if the drivers you installed would support the virtual card. Besides, these video cards would also require licenses (and a dedicated license server), which means an additional service running on the VM, but how would this work from inside a docker image or a kubernetes pod?

@nagarutu 2 года назад

Good video, but Im stuck in testing the tensor flow, the code when you launch the tensor flow is hidden by your picture and I cant find it in the linked site,

@TechnoTim 2 года назад

Thank you! Sorry about that! The code is here techno-tim.github.io/posts/gpu-passthrough-linux/

@williamall6130 2 года назад

@Ouldsmobile 3 года назад

What is the script you ran from the tensorflow docker shell? Video doesn't show the command, just the result. :-)

@TechnoTim 3 года назад

The command should be in the docs, which is in the description!

@Ouldsmobile 3 года назад

@@TechnoTim No worries, I did check the docs first and it wasn't listed. I managed to sort it out though, so it's all good.

@shoebucket 3 года назад

@@Ouldsmobile Could you share with the rest of the class, please? :)

@Ouldsmobile 3 года назад

@@shoebucket What I meant was, I got nvidia passthrough to docker working on my setup. Never did end up getting the tensorflow test to work. :-/ I just relied on making sure the nvidia-smi worked in docker, then setup my plex container to use nvidia and watched nvtop, once I saw it was using the gpu for transcoding I knew it was working.

@jordanmfarmer 2 года назад

@@TechnoTim You keep saying this is in the documentation but it isn't bro.

@cedricmarcoux5795 3 года назад

Hello, for my part, to get all this to work, i have add to modify the daemon.json like this: { "default-runtime": "nvidia", "runtimes": { "nvidia": { "path": "/usr/bin/nvidia-container-runtime", "runtimeArgs": [] } }, }

@jvlogs6140 3 года назад

Hay Tim great video can you recommend any hardware for me as I am looking at building my first server for this kind of setup I have a spare nvidia 1070 I will using and am looking at going down the amd route 6 to 8 core unsure on motherboard specs I should be looking for any help would be appreciated 👍😉

@TechnoTim 3 года назад

Check out some of my kits.kit.co/TechnoTim

@jvlogs6140 3 года назад

@@TechnoTim thanks Tim will do

@ProGamer1115 7 месяцев назад

How would I do this on Intel iGPU or AMD

@EricZarnosky 3 месяца назад

Look at Jim's Garage Channel. He has a good video on doing it with Intel GPUs

@jorgegomez374 3 года назад

New at servers. Right now I am running my stuff on raspberry pi have a old mac whit the Intel processors I edited videos there. If install linux and install docker think will work? Thank guys great video

@xxcr4ckzzxx840 3 года назад

Should work just fine.

@LampJustin 3 года назад

That will work fine, but how old is "old"? If it's Intel 4k + (so Haswell and beyond) using hardware transcoding on these is pretty dang awesome and a much better experience then with Nvidia. Make it yourself easy and install libva and plex-media-server on Ubuntu LTS without docker. Especially when you're new to this.

@TechnoTim 3 года назад

Jorge Gomez what modzilla said! Thank you!

@stuartwilson2277 3 года назад

:( NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver. Make sure that the latest NVIDIA driver is installed and running.

@dustinkrejci6142 3 года назад

I was looking for a way for me to use Docker to make a computer lab for Remote Desktop though the internet in a Kubernetes cluster. Does anyone know if there is a way to do this? I want to have a Remote Desktop computation for my future company.

@TechnoTim 3 года назад

Check out my Proxmox GPU passthrough video.

@dustinkrejci6142 3 года назад

@@TechnoTim oh I have seen it already! :D thanks for the referral.

@dustinkrejci6142 3 года назад

@@TechnoTim I have a project for you, what if I had a company like shadow cloud for Remote Desktop. Yet on the backend my company uses Docker/Kubernetes with GPU PASSTHROUGH so I can give my virtual assistants a standardized Remote Desktop in the cloud. There’s a customer database too for people to sign up for my service. “ but this is like Docker VDI/ Kubernetes management of Remote Desktop for The Whole Internet. I want a Docker to cache downloads as part of my stack. I second world countries that I selling this service to needs to be networked optimized, because of the slow internet speeds in Urban and Rural places to the remote office workers this is meant for. I know most of what I want but I have no coding experience to get there. I’m on the scent to the next big tech evolution. I can talk more but I don’t want to drown in words. Is there such (solution) a way that’s already been built? If not can you build such the core of the tech I’m trying to put into words? It’s what I am talking about even possible?

@axelolaussonholtenas5087 3 года назад

How do you come up with your vm names?

@TechnoTim 3 года назад

Whatever I can think of that is astronomy or science related :)

@axelolaussonholtenas5087 3 года назад

@@TechnoTim COol!

@BlazinNova 3 года назад

What ssh software is that looks nice?

@TechnoTim 3 года назад

Thanks ! It’s Windows Terminal with WSL and oh my zsh. You can set up your environment like mine here ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-kL8iGErULiw.html

@BlazinNova 3 года назад

Thanks for the response I really am enjoying your videos hopefully I can start my own home lab in the future. Also are you able to do a video on a MergerFS/Snapraid setup in the future using proxmox I’ve been looking into doing this. Since I don’t want to format my exist hard drives that have data and I can’t really move them anywhere. If it’s possible of course but great videos keep up the great work

@bulcub 3 года назад

This is a great video, I'm about to setup a bare metal linux server, and wanted to know, would I still need to all that is mentioned to run gpu passthrough? I will be using a 1050 nvidia card? thanks

@TechnoTim 3 года назад

Thank you. You will still need to do most of this, just not any of the items that apply directly to virtualization.

@kingundfaker 3 года назад

So will you make a video about proxmox web api? :D

@TechnoTim 3 года назад

Does anyone use it :)

@kingundfaker 3 года назад

@@TechnoTim I don't think so :D My Homelab (Nextcloud especially) is set to Wake on Lan because of costs. Therefore it is nice to have because I don't have to log into the Proxmox Dashboard to start my Containers and VM's but generally I don't think anyone would use it. :D

@bikhard0 5 месяцев назад

it really needs update for this guide. not working on latest drivers n ubuntu

@maxdmytrenko Год назад

You should make an update to this video, some things have changed.

@LampJustin 3 года назад

Thanks for the video, it's pretty great! But honestly why r people buying Nvidia cards just for Plex encoding. Why not use a much more power efficient Intel processor with QuickSync Video. The later revisions (starting at Haswell and later) are actually pretty amazing and don't have a lock and a VRAM limitation. Someone even got more then 15 1080p stream to de- and encode on an i3 I think.

@TechnoTim 3 года назад

Modzilla fair! I have one for more than plex and one leftover from another PC. Also, It’s hard to passthrough the onboard gpu to a vm, at least in my experience. Thank you!

@shoebucket 3 года назад

It's also a lot easier -- and far less expensive -- to purchase a legacy NVIDIA gpu and plug it in a PCIe slot, than to get a whole new motherboard/cpu/ram setup.

@LampJustin 3 года назад

@@shoebucket I'm sorry but I'd like to disagree there. A legacy Nvidia card won't be able to de-/encode HEVC (10Bit) and that (at least for me) is a total dealbreaker as I try to encode all my movies in that format. It's more space efficient and almost all my devices are now able to direct play it. The support for HEVC 10Bit HDR is only getting more important. Having a normal Nvidia GPU that could do 2 HEVC reencodes wouldn't be a good option as that I can do with my Ryzen 1600 without a hitch and that is an 80 bucks cpu. In my country power draw is another big consideration to be made and with another gpu the server will draw at least 15-20 Watts more when in idle and that's too much in my book. Intel really optimized the idle draw. It would be much lower. So in the end you at least won't pay more compared to a discrete GPU. I have a mITX chassis so having one PCIe slot for a 10Gbit/s nic is very welcome. So in conclusion you get 10Bit HEVC HDR support, have a lower power draw, don't have to upgrade for a pretty long time and you preserve another x16 slot. But maybe I'm missing sth, I dunno. (And yes I know about the driver hack that enables the full use of the nvenc, but that's not a proper solution)

@shoebucket 3 года назад

@@LampJustin Perhaps what I meant by "legacy" wasn't what you were thinking. All I meant was, it's a whole lot cheaper and easier to add a PCIe card to an existing system than to rip all of the guts out of a system and replace them with new architecture. Plus, there may be proprietary hardware to consider (surely I'm not the only person in the world who is using retired enterprise gear -- a Dell PowerEdge 420, for example -- to run their homelab stuff) Will the performance boost be as much as if you could completely replace the rest of the underlying hardware? Probably not, but it's still a benefit to folks like me.

@LampJustin 3 года назад

@@shoebucket fair, different perspective ;)

@CruzMonrreal 3 года назад

"...We'll want to add a couple of flags to our config. " "...First we'll set our distribution.

@TechnoTim 3 года назад

Cruz Monrreal hey! Thank you for the feedback! This is very helpful for future videos! I never know how much context to give and at the same time that makes the video longer. Also, the docs are in the description, and I did mention that. Thanks again for the feedback! This will make future videos better!

@Stephen-wh7vl 3 года назад

Yeet

@TechnoTim 3 года назад

Stephen ya ya yeet

@brian_3236 3 года назад

I can not even get the drivers to download and have even tried it without the 450. ( root@omv-nas:/# sudo apt-get install --no-install-recommends nvidia-cuda-toolkit nvidia-headless-450 nvidia-utils-450 libnvidia-encode-450 Reading package lists... Done Building dependency tree Reading state information... Done E: Unable to locate package nvidia-headless-450 E: Unable to locate package nvidia-utils-450 E: Unable to locate package libnvidia-encode-450 )

@TechnoTim 3 года назад

Which version of Ubuntu are you running? This works fine for me on 20.04

@brian_3236 3 года назад

@@TechnoTim My server is OMV (OpenMediaVault) Using Debian 10 (Buster). I did have the full Nvidia driver installed at one time for my GTX 1050 but removed it cause i could not figure how to get it into docker. I have JellyFin in a docker and really want to get hardware decoding working to stop the hesitating video on my phone when I am not home .