Proxmox vGPU Gaming Tutorial - Share Your GPU With Multiple VMs!

[@CraftComputing]

Check out the EVGA Z15 Keyboard to start your custom keyboard obsession: amzn.to/3jRhlvP

[@Romnipotent]

When are you going to start printing moulds with 3 layer material keycaps with resins and harder enamels to make your own? You know... craft your own :D

[@flexxx222]

I really like the "perky" RGB lighting on this keyboard :) 0:34

[@lifegivesulemonsmakelemonade]

@craftcomputing Can please make video on red hat virtualization setup

[@FlaxTheSeedOne]

In the installation document on the VGPU unlock part in the seccond to last command you spelled "deboot" instead of "reboot" Not a big deal, just thought i'd let you know. This would also be great on a github page With some more text and Instruction for People to find this who don't search on youtube :)

[@GodmanBG]

1:47 yes I am looking for a linux desktop with access to full gpu acceleration windows VM , has this tutorial been released, I can't seem to find it?

[@talideon]

It's 2021. The world is a dumpster fire, but at least Crysis can run with full settings in a VM!

[@CraftComputing]

Two Crysii in VMs!

[@sekritskworl-sekrit_studios]

@@CraftComputing Well played!

[@jackmclane1826]

TWICE!

[@igordasunddas3377]

I thought the plural of Crysis is Crysis Remake? 😆

[@TheNone724]

2022: Allow me to introduce myself

[@vaidkun]

./ does not mean execute and run, dot represents current directory. so it means file in current directory. It is sometimes necessary to add full path or the current directory marker when executing files, so that may be source of confusion.

[@Appri]

hey man, how can you pass through multiple vGPUs to a single VM with id emulation?

[@SB-qm5wg]

Wow, that's crazy. vGPUs profiles

[@kimcosmos]

Parsec free trial is worrying and no ubuntu server yet. Recommend any FLOSS alternatives?

[@allrack1]

Any chance for an updated on PROXMOX 8 with latest nvidia drivers ?

[@GearSeekers]

Not all heroes wear capes.

[@PendakwahTeknologi]

True. spiderman, wonder woman, garfield. etc.

[@nopixel_media]

@@PendakwahTeknologi10/10

[@mikebroom1866]

The limitations of the vgpu memory really starts to make the 24GB of the 3090 seem more sane.

[@CraftComputing]

*stares lovingly at my 3090*

[@mikebroom1866]

Same! The fact that I couldn't get a 3080 seems less painful.

[@TaufanRezzafriMochamad]

unfortunately 30xx doesn't work with vgpu unlock, I hope someday nvidia enable sriov on consumer card, at least for 30xx

[@McTwistedTwisties]

I look strangely at ny 16gb 3080 laptop edition

[@shadowtheimpure]

I finally gave up and spent $2000 on a 3090. It hurts, but f*** it.

[@Jake9066]

"Depending on your internet, this will take 2 to 5 minutes to complete." You drastically underestimate how terrible some internet connections can be.

[@Joseph_Coutts]

10-20 mins with Australian internet

[@vonkruel]

Very clear instruction without a bunch of fluff. You've done a service to the community here.

[@WrightBrosVideos]

Windows line endings would be different than the Linux line endings when performing git clone. Other than that the files would appear identical. Maybe that has something to do with the scripts not working?

[@MilesProwerTailsFox]

no

[@twitchyarby]

That's what I was thinking

[@siberx4]

I agree this is a likely cause; depending on how he was comparing the files, this difference might have been "hidden" from his editor/tools. Git has a variety of options for how it handles line endings on clone that depend on the configuration and operating system, and it's entirely possible that in this case the clone was converting some line ending somewhere it shouldn't be. As a test, changing the git global settings on the proxmox server to leave line endings completely as-is would allow you to validate this hypothesis.

[@deefdragon]

@@siberx4 technically the easiest test would be to check file hashs.

[@mrsubrange]

Md5sum is the best test between git and zip file. After that, narrow down the reason using other tools. I'm not convinced that there's "no difference" of course there is a difference...

[@heropointgaming3864]

yoo I was literally searching for this and here comes craft computing with the yt notification. TNX dude

[@strawbemily3182]

cant wait for that linux with windows vm vid. thats what my setup is but i think i have it setup sub-optimally and want to see how jeff does it!

[@Marco_Onyxheart]

I'm hoping Jeff can get it working without having to log out of Linux. Right now, I have a single GPU setup, but it logs me out of Linux when I launch the VM. I want to still see my Linux system on my side displays. I've been trying to get it to work with the Intel IGP as a backup, but they just go black when I launch the VM, probably because the Intel GPU isn't actually rendering anything, but using PRIME offloading instead.

[@igordasunddas3377]

I kinda also hope that...

[@Dave-kq7gv]

"This is a pretty straightforward process..." -- 40 mins later after a shitload of explanation-less command line parameters "Done!"

[@CraftComputing]

"This is a fairly straightforward process..... so long as you enter all 300 lines EXACTLY as I say"

[@Standbackforscience]

Actually, he explained a lot of stuff along the way, which he could have skipped over if he wanted. The process of virtualizing a GPU is pretty complex, and covering all the theory behind it would take hours. I think this video is best seen in the context of someone who has tried and failed and several times, and understands the general pitfalls. It's also more proof that it's possible, getting an Nvidia to do something like this has been impossible at this point, so I think most of us were perfectly happy to skim through it all just to see that it actually worked. Which it did. Thanks so much @CraftComputing!

[@aaronb4957]

This video deserves a comment for the algorithm.

[@Th3ace223]

What an amazing setup. I wonder how easily you could swap GPU profiles. I love the idea of giving a VM 12GB vGPU memory when I know im the only user, then swapping back down to 6GB when another user is going to connect.

[@HowdyFolksGaming]

“The comments section of this video does not work nearly as well as a Google search bar.” Amen brother. Amen.

[@georgeashmore9420]

Git repositories can have recursive Git repositories that require a different pull command that zip may avoid

[@johnnyxp64]

I'm only like 10 minutes into the video but I still wish to thank you for the wonderful editing! this is how a good tutorial should be.

[@novellahub]

The keyboard game is getting pretty strong with this one.

[@erusman8863]

Cool been waiting for this. I have had some problems enabling pcie passthrough in Proxmox. But loving Proxmox, I can run multiple Debian machines, Linux Mint... etc. Thank you!

[@Mehdital89]

Windows, pfsense, openwrt... Everything!

[@RaidOwl]

45 min??? *sets chair to recliner mode*

[@thedeathwalkerlore]

Did you miss a few steps? Mostly about actually using the mdevctrl stuff... I see you put it in a txt file for later, then never use it...

[@blkspade23]

You know ./ does not mean run.. Yeah, I'll be "that" guy today.

[@CraftComputing]

psssst.... WINE is an emulator.

[@damonmassey4426]

@@CraftComputing s/is/is not/ =)

[@blkspade23]

@@CraftComputing Lol, I'll give you that one.

[@d0hanzibi]

Video we needed, hacks we deserved. This is pure awesomeness, i hope vGPU will be increasingly more used and available for all of us home tinkerers.

[@nicholastoo858]

What’s the advantage of Hyper V GPU P method over this? And vice versa.

[@CraftComputing]

I was I was going to get to it, but.... 46 minutes later....

[@franzpleurmann2585]

Now if this would be working with amd gpus as well because the big navi gpus offer significantly more RAM. Still this is awesome. Thank you so much for the tutorial ❤️

[@MrLordbeavis]

And I just got a amd gpu

[@craigw4644]

This is the wildest tutorial video ever watched; very impressive, understood the whole process. I'm debating building a Windows game server or using Proxmox and passing through a video card on a Windows VM for gaming. I will not need to share resources, so the whole card will be dedicated to games. My research on the subject is mixed. Thanks for posting, looks like it required a lot of homework.

[@PendakwahTeknologi]

Hey, we have the same mouse. Long live elecom.

[@adibbh]

Hello Pendakwah Teknologi. I am fans for both of you. From Kodiang.

[@noblej7897]

Me who can barely run a game with all the hardware.

[@MCHEROBRINEBROS]

I've never clicked faster

[@maxmalmberg4755]

Super exsided for the hyper V, hyper V is my choice of virtuellzon program

[@bluegizmo1983]

Now if we could just get someone to get this working on UnRAID and put out a tutorial, or even better a simple plugin we can install!

[@criostasis]

Just bought my first rack, finally going to get around to mounting up my servers and try and get this going, then no support for 30 series cards 😢.

[@legofan2284]

Since 3000 series cards don't actually exist (at least I've seen no evidence apart from review samples), that's totally fine

[@mitcHELLOworld]

@@legofan2284 I know you’re joking but crazy enough according to steam hardware survey the 3000 series has already outsold the 2000 series. Gpu demand has just hit record highs with the pandemic and such.

[@PlaceholderforBjorn]

@@mitcHELLOworld That I didn't know! Plus the demand from miners during the value increase in bitcoins. I hope for a lot of cheap second hand GPU's in 2022. Then I maybe can build two decent computers for my sons.

[@mitcHELLOworld]

@@PlaceholderforBjorn agreed!

[@WhiteSkyMage]

A couple of questions for you: Can you do a project like this for AMD cards? Will this vGPU method work on a RTX 3090? Using this method, is possible to create a "local game cafe" streaming server or rather "regional cloud gaming server" to be used by gamers for competitive/casual gaming? Is there any latency overhead if used locally? Can I connect a VM directly to a monitor instead of encoding and streaming the game to a thin client?

[@imnitin97]

Got any answers to this questions

[@magraopb]

My GRID K2 only support driver 367 version, anyone find this drivers?

[@zaneoblaneo7624]

Can I convince you to post a SHA-256 checksum of the "NVIDIA-GRID-Linux-KVM-450.80-450.80.02-452.39.zip" file? I'm curious to see if the version I found on the open internet is the same as the one that came bundled from Nvidia.

[@CraftComputing]

Mine comes up fe9fa15475f02846cfde9ea944f2b23d1dae5359 Checked on Windows via certutil -hashfile

[@zaneoblaneo7624]

​@@CraftComputing Can confirm that if you know how to google dork a file, it's really easy to find the correct driver online, without waiting for Nvidia to approve your grid license. Just PLEASE check the checksums before using it. Also, nice idea to use a windows builtin tool to check the CRC, though I usually use 7-zip's context menu CRC tool, because you don't have to open a command line. The checksums I have are as follows: SHA256: CD8284509773FEEB57D1BFD9EA5F9302E24E67153E87339C7E5BC58A061FC383 SHA1: FE9FA15475F02846CFDE9EA944F2B23D1DAE5359 BLAKE2sp: DEB51C322A8565985D07375B1733F14379CD71400A827113FF345BE31EBA7C72

[@stefanl5183]

Thanks!

[@LuffyOniiSan]

@@zaneoblaneo7624 Could you give some pointers on how to google dork it? Can't seem to find it myself. Thanks!

[@Klessc]

@@zaneoblaneo7624 at first i wanted to try this out.. now waiting for the approval but i doubt ill get that, now im digging in on learing how to dork the file... lol i didnt even know this was a thing, been attempting to get the file sense day 1 of the upload.

[@TechteamGB]

*Takes notes* ...

[@SomeNameGoesHere]

At the 15:00 to 17:00 mark, is there a particular reason the /usr/lib/systemd/system version of each file doesn't get /usr in the path of the modified line when it is there in the original ExecStart? It looks like the /usr/lib/nvidia/systemd versions kept /usr in the modified line.

[@Scitch87]

Craft Computing: "Share your GPU with multiple VMs!" My half broken GTX660: * heavy sweating *

[@mbmumford]

My fully working 660M: * laughs hysterically * …No

[@camerontgore]

Knees weak, arms heavy...

[@erwintwr1193]

very well explained sir! Pity about the vRam having to be equally split, but hope that improves in future. Waiting for the Hyper-v solution as well. ( Might be a good idea, if it is on the same hardware, to compare what performance you are getting vice-versa?)

[@FUNK_DAS_ANTIGAS]

can you make a tutorial how to build a server local to stream ? windows server with OBS. thanks

[@112Haribo]

It could be interesting if you showed us the Hyper-V way of doing this as well, since you said it was easier.

[@lasselindberg]

I would also like to see it on VMware

[@deltawhiplash1614]

@@lasselindberg me too😭😭😭

[@thematrixian8710]

He did post a video for this now :)

[@112Haribo]

@@thematrixian8710 yup I noticed. Thx for the heads up 👍🏻

[@thematrixian8710]

@@112Haribo my pleasure I was guessing you'd already be aware, just in case others stumble upon this :)

[@JackDespero]

Thank you very much for the tutorial. But it seems to me that it is completely crazy that for such a common operation, one needs to do so many things instead of, you know, just typing a single command, running a single script that someone put together, or just flipping a single switch. And the fact that you have to contact Nvidia and wait for the drivers?? Why??

[@AxelHue-yq7nu]

I work with CAD+3D Slicing Softwares.... Now i bought a $1000 Nvidia Card .... and how to email them in order to get a 90-day-trail Licence ??? Clear case : I shall pay monthly at the end ? Are they gone mad ? If i overthink it. its clear , Nvidia thinks : If we sell one card and someone uses it for 10 PCs then we sold 9 cards less. But i bought a card in order to user it alone BUT onsite and if i travel. I wont spent 1000 bucks for a card in order to get a life time fee. And even if i read the "unlock" thing.....what happens if the smuggle code through updates and revers it ? Does anyone has a solution ? Does sometone knows how it is about other manufactures ? thx

[@tubefulable]

My hypothesis: I think the fact that you didn get a "vgpu patch unlock applied" with your git version of the project, is because debug isn't enabled (today it is at row 570 in vgpu_unlock_hooks.c). Maybe you changed it by hand and forgot to mention it, or you downloaded a version that had debug enabled by default. Since debug has been enabled by default for a couple of weeks (you can see it in the git history). Bottom line: your git clone was newer than your zip file. (I was disappointed that, in my case, following your tutorial didn't give me the "patch applied" message, but then my girlfriend pointed out LOG wasn't enabled in the source file).

[@thesixwings2077]

What a good girlfriend!

[@watch4fun13]

That saved me from going insane. Thank you and your girlfriend.

[@andreasw.3354]

One question: will this affect the passthrough of my GPU to several LXCs which I use right now?

[@Pavlogal]

Boosting the algorithm and also want to say I'm excited for the linux+windows vm and GPU-P videos!

[@bertjan43]

Bit late to the party, but I love the tutorial. Untill the point where I have to get an Nvidia Enterprise account. It seems the only way to do that now is to buy a Nvidia Licensed Product. Is there any other way to get the drivers?

[@ytg7257]

I'm also trying to figure it out.

[@rahulchandra152]

Quick tip linux comes with a uuid generator. Just type in uuidgen into command line

[@aliwalil4160]

Dude, you are awesome. I don't need doing gpu passthrough right now but your content is invaluable.

[@jeditec7479]

I tried with a 1070, it showed as a P40 in proxmox, I selected 2gb instances, But when I install the drivers the card shows ok in the system but if I try to use parsec it shows the 43 message. Tried hiding the CPU to show host cpu, still no luck. 43 error or BSOD, if someone has success with a 1070 please share the details.

[@DavidAshwell]

Hardware manager on your Windows VM shows your GPU working properly? No caution symbol? In my pass through attempts, once I get to a fully functional GPU in the VM (with the correct driver working), then parsec just works.

[@jeditec7479]

​@@DavidAshwell I was able to make it work, I saw an error in journalctl Jul 07 03:37:02 n2 nvidia-vgpu-mgr[28788]: notice: vmiop_log: ######## Guest NVIDIA Driver Information: ######## Jul 07 03:37:02 n2 nvidia-vgpu-mgr[28788]: notice: vmiop_log: Driver Version: 471.11 Jul 07 03:37:02 n2 nvidia-vgpu-mgr[28788]: error: vmiop_log: (0x0): Incompatible Guest/Host drivers: Guest VGX version is newer than the maximum version supported by the Host. Disabling vGPU. Jul 07 03:37:02 n2 nvidia-vgpu-mgr[28788]: error: vmiop_log: (0x0): VGPU message 1 failed, result code: 0x6a Jul 07 03:37:02 n2 nvidia-vgpu-mgr[28788]: error: vmiop_log: (0x0): 0x1c, 0x8, 0x100, 0x100, Jul 07 03:37:02 n2 nvidia-vgpu-mgr[28788]: error: vmiop_log: (0x0): 0x100, 0x0, '471.11', 'r470_96-9', 'DVSReal r470_96 471.11 DVS-Applications' Jul 07 03:37:02 n2 nvidia-vgpu-mgr[28788]: error: vmiop_log: (0x0): VGPU message 47 failed, guest VGX version not initialized... Jul 07 03:37:02 n2 nvidia-vgpu-mgr[28788]: error: vmiop_log: (0x0): VGPU message 47 failed, result code: 0x56 So it was basically removing the newest driver and downloading an old one, 452.39 is the supported driver according to release notes docs.nvidia.com/grid/11.0/grid-vgpu-release-notes-generic-linux-kvm/index.html Now everything works ok.

[@jeditec7479]

​@CrunchteamHD It works, make sure to enable debugging in vgpu_unlock_hooks.c look for this message. /* Debug logs can be enabled here. To enable it, change 0 to 1. */ Remember to uninstall and recompile the driver, after that you might have issues with the drivers version. Make sure to use the following drivers NVIDIA vGPU Software Version NVIDIA Virtual GPU Manager Version NVIDIA Windows Driver Version NVIDIA Linux Driver Version 11.1 450.80 452.39 450.80.02 Send me a message if you have any issue, check my other comment where I share the details.

[@jeditec7479]

@CrunchteamHD thats great news, if you are able to make CUDA work inside windows let me know, I am getting acceleration and can play video games, but I intended to have another VM for machine learning but it seems cuda is not working. BTW which versions of drivers did you used?

[@Zoltag00]

The obvious difference is the existence of the git repo (.git directory). It would be interesting to know if the existence of this is a problem Another difference (admittedly on my Windows machine) is the file types - When I clone the git repo locally, the files are bigger and I think that is due to them being cloned with line endings (as opposed to line endings in the zipped copy). I know that git can be configured to alter the line endings depending on what you need on your system, so thats another angle I'd be looking at

[@zerberoshades8946]

Actually, I just watched the vid and my first thought regarding this was

[@chandrakantkumar7699]

Really great tutorial, i am planing to work with quadro p2000 gpu. Waiting for hyper-v guide. 😄

[@TheBauwssss]

Run a diff on those two supposedly identical files, and I am willing to bet €100 that those two files are in fact not quite as identical as you claim 🤔😉

[@CraftComputing]

i.imgur.com/SByACjo.png I accept PayPal, Bitcoin, or just commenter tears.

[@tahasever]

Hello, I am just wondering if the 3000 and 4000 series rtx cards are now supported or not

[@Tofflus]

me too!

[@markcentral]

Thanks for this tutorial! The only place where I got a bit lost was when you made changes to the Nvidia Source files, removed the module with dkms, and then re-inserted it. I am not an expert at these things, but I was expecting the nvidia module would have to be recompiled *before* re-inserting the new version to incorporate changes made to the .c and .Kbuild files

[@mrsubrange]

I agree, this sounded weird in my ears too ... Any explanation anyone?

[@ronnyhaldorsen6213]

Thanks for great tutorial. I see that Proxmox 7.0 has been released today. I hope you can update your tutorial to this version.

[@antargaming1619]

I already did hyper v after I saw a form post on this but I’ve been waiting for your vid on this. I’ll be using a titan xp 12 gb and a 7940x

[@NickyNiclas]

Damn, I really hope someone knows what causes that issue because I've had it too while doing other things. Super strange. The files MD5 the same and everything.

[@ravenpsi4088]

Finally!!! I was waiting for this, uploaded 15 minutes ago and still think i came in late xD

[@natpeterson7988]

I'm wondering if the issue with the git vs scp was a file permissions issue, had issues with git persisting them before. Pretty weird nonetheless

[@Dqixol]

Thank you so much! I have been waiting for vgpu on consumer cards for so long! Hope we can have it supported officially on next gen gpus.

[@OARomanov]

Oh my god this is absolutely perfect tutorial! Jeff, thank you very much for it! Please go on and greetings from Siberia!

[@xxriverbeetxx1.065]

if i have a tesla gpu idont need this 14:47 or?

[@FSchack]

Still no solution for Ampere cards?

[@animalmthr]

I wonder if the strange issue with the unlock script could have to do with end of line (EOL). I am not sure if Git change the line break depending on what system you clone to - but that's the only thing I can come up with.

[@bb2ridder757]

nice play on the 2 gamers 1 cpuXD

[@insanesk8123]

Love the tutorial! Glad you finally got it working. I am wondering if the issue with the script has something to do with Windows vs Unix line endings.

[@JPEaglesandKatz]

Your tutorials and vids are something special... Clear and easy to follow. And I love the fact that you enable us to get the files for more easy copy/pasting!!! Keep on rocking!! :)

[@wujekcientariposta]

crazy thing is this would be even more effective on an amd gpu with it's ton of vram on a budget.

[@_rene]

Hey thanks for the great video and excellent instructions. I currently have an amd gpu do you also have a video or link to setup with this?

[@richardleaneagh4274]

foot pedal page turner might be an idea to replace auto scroll seen a lot of musicians use it with tablets

[@BobBrittonBespoke]

For anyone that finds this video at a later date... the instructions of dmesg showing 'vGPU unlock' is NOT accurate. I got this system working 100% and it NEVER showed that message. Just a heads up, that one thing cost me many hours of head scratching.

[@bobs-u9i]

Thanks mate - This has probably just saved me many hours too - I couldn't get it to display... I'll continue on with the video now and keep my finders crossed :D

[@adamgleason1338]

What did you have to do different? I have spend two days on it and can't get the dmesg thing to show up either but I can't get it to work. My VMs won't boot with the args: -device... line in the VM file.

[@stephenreaves3205]

How did you download the vgpu unlock zip file? Git clone will pull from the main branch but if you downloaded a release from github then that's essentially a git tag and should be stable. main is hardly ever garaunteed to be stable

[@bindkeys]

What's stopping the 3000 series from being supported, the patch or NVIDIA support? I'm looking at a GPU and a 3060 would be great because of the high vmem.

[@it3963]

did u find anything?

[@KeeperOfTheGate]

And the legend of Craft Computing continues.

[@lucavignati2958]

very very good tutorial! i will try it asap! do you think there is a way to create a "cluster" of gpus ( 4x 1080 for example) and then create a virtual gpu from this sort of cluster??

[@krispcode]

Can you do an updated version of this? Been trying to do stuff but it just doesnt like working

[@HarryWho102]

I can not thank you enough. You supply this information and your experience for no cost. Plus you have a smile whilst doing it. Good-on-ya.

[@Dave-kq7gv]

I have a 24-core threadripper with 2x 2070 supers....could this work with that? I don't know of an equivalent quadro card [edit] either way, cool vid man! I'm both thrilled & jealous you got it working 👍👍

[@CraftComputing]

Absolutely! The 2070 Super would work very well as an RTX 6000, just like I showed in this video. That would make a pretty killer build.

[@ytg7257]

Watched this almost in May 2024. Now I want to do this in my setup but not for gaming (just because I wouldn't mind having more decent GPU acceleration in my VMs). But I'll try to do this not as a dedicated Proxmox server but directly on Qemu/KVM with two Nvidia cards (one for host, other for vGPU). Any chances in accomplishing that?

[@darrenfalconer3267]

I use aster to achieve similar setup. Allows the windows machine to use multiseat access. I plug a monitor in to each video output, then plug in a keyboard/mouse set for each screen and set the users up. Some software is an issue if it is loaded in one log in but I'm trying to learn sanboxie to get around that. I think the overhead is easier than virtual machines and doesn't require configuring hardware and needing virtual disk's.

[@NdxtremePro]

Its 2024 and I just now realized Zelda BOTW stole the beginning cliff view from Crysis.

[@ivanbeach1665]

Thought for the day, normies actually work in factories making the absolute garbage that they eat and drink.

[@timurzaynutdinov3445]

​@UCp3yVOm6A55nx65STpm3tXQ , please when you use "grep" use it in case insensitive mode "grep -i". Thanks for the tutorial.

[@georgeashmore9420]

Can you pass through USB sticks from the thin client? I'd love a computer that follows me around the house

[@itsTyrion]

Meanwhile I can't even install the damn GRID driver. What a shame, I came here hoping I did something wrong so the problem can be fixed - apparently not ._. Below CONFIG options are missing on the kernel for installing NVIDIA vGPU driver on KVM host CONFIG_VFIO_MDEV_DEVICE Please install the kernel with above CONFIG options set, then try installing again

[@MariaInesHiggs]

@craft computing, the error you had on the script probably was converted to windows iso format. Try to use dos2unix command on it and see if the md5sum changes. I had the same error on another ocasion, and was only able to see the difference with strings $filename - it added the ^M to the end of each line, and diff was not able to pick it up.

[@VinnyG919]

p100 on proxmox only gets recognized by the grid driver when i set machine's iommu type from intel to virtioGPU any tips?

[@Thriving_in_Exile]

28:40 "Goes completely against _morality_...." Yes. Because nonsensical contracts and product segmentation from multi-billion dollar corporations dictates "morality." Sorry, *_what?_*

[@BansheeBunny]

2:23 Almost died laughing. 20:20 Could you split a 2080 Ti if you treated it as a 10 Gig card? 25:02 8 Gig of RAM seems light for gaming. If you have the resources, 12 Gig would be my target. 41:52 I usually let my stouts sit a while before drinking, as a rule of thumb. Give Death by Flapjacks a try from Oskar Blues Brewery. Well done video, will be looking forward to the next one.

[@CraftComputing]

2:23: THANKS! 20:20: A 2080 Ti will still have 11GB of ram availalbe. Use 'mdevctl types' to see available profiles. Similar to my 2080 in this video, you'll be able to do 1x8GB and 2x4GB, but you'll have 5x2GB available.

[@mdkrush]

What about using multiple GPUs? I have a gpu miner that is now useless... So I could have 12 GPU Proxmox??

[@rikka0_059]

Does it work on RTX graphic cards? I though Nvidia vGPU is available to GRID GPUs only.

[@ariyasuryaman]

hello sir, its IMPORTANT question, please answer honest, when i use NVIDIA GRID DRIVERS it just share about gpu memory and the vgpu ? or i can have full performance of physical gpu but only gpu memory splited ? i have success try about your video "2 gamers 1 pc paravirtualiazation with hyper-V" the result it just sharing gpu performance in to VMs dynamically, I want to use vgpu technology with all the physical performance of the gpu and only share the memory for each VMs, its that possible ?

[@OGXDev13]

Looks like the nvidia vgpu unlock script didn’t work for me. I’ve ran it twice and both do not output “vgpu patch applied” followed the video and copy and pasted line for line on the docs... can’t for the life of me get it to work. Any suggestions my setup is ryzen 7 1700X MSi x370 Pro Carbon with 2 gpus one amd rx 580 for console output and gtx 1080 ti

[@phooopy7941]

So the vgpu daemon and manager fails to start at boot and I get ModuleNotFoundError: No module named 'frida' when using systemctl status This is strange considering the fact that I did copy/paste and use pip3 install frida. Any idea what the problem might be?

[@ChínhDz90]

can you make a video about GPU for vmw workstations?? like Hyper-V and Promox

[@skawashkar]

No output for mdevctl types. But able to detect the Nvidia graphics card via nvidia-smi and nvidia-smi vgpu.

[@degrengolada2360]

11:01 i only got [ 0.036279] DMAR: IOMMU enabled does it work or not?