Universal Blue•2y ago

Dear god man. That's a lot of machines.

KamquatOP•6/23/24, 1:10 AM

Oh wait those are just a lot of displays

KKamquat Dear god man. That's a lot of machines.

HikariKnight•6/23/24, 1:13 AM

nvidia test VM (with gpu passthrough), main machine, HTPC (stuck on F39 due to polaris atm but moving to F40 soon once some patches land)
on the go hobby work laptop, legion go, deck (running steamos)

2 penguin plushies my friends sent me as secret santa gifts as jokes (the sunglasses on the small one was a dlc that got sent to me last year, didnt know my secret santa gift came with a season pass)

bunch of controllers i use for testing (plus another 40 in a drawer behind the desk)

Tsubajashi (ツバジャシ)•6/23/24, 1:13 AM

moved my dots and iso to my ventoy usb drive, wish me luck

TTsubajashi (ツバジャシ)moved my dots and iso to my ventoy usb drive, wish me luck 😄

HikariKnight•6/23/24, 1:13 AM

good luck!

HHikariKnight nvidia test VM (with gpu passthrough), main machine, HTPC (stuck on F39 due to p...

KamquatOP•6/23/24, 1:13 AM

...you have like 50 controllers?

antheas•6/23/24, 1:13 AM

gnome merged accent colors?

KamquatOP•6/23/24, 1:13 AM

PaimonShock

antheas•6/23/24, 1:13 AM

am i dreaming?

HHikariKnight nvidia test VM (with gpu passthrough), main machine, HTPC (stuck on F39 due to p...

Tsubajashi (ツバジャシ)•6/23/24, 1:14 AM

do you have experience in gpu passthrough when only nvidia gpus are in a pc?

Tsubajashi (ツバジャシ)•6/23/24, 1:14 AM

since im planning to make use of a second nvidia gpu in a windows vm, so it would be unpractical for me to unload the nvidia module fully

Aantheas gnome merged accent colors?

Kyle Gospo•6/23/24, 1:14 AM

We've been blessed

TTsubajashi (ツバジャシ)do you have experience in gpu passthrough when only nvidia gpus are in a pc?

HikariKnight•6/23/24, 1:15 AM

they need to be detected as different GPUs, if you have lets say 2x 3090s you need to add custom scripts tailored for your card into initramfs, this is not something we do

Tsubajashi (ツバジャシ)•6/23/24, 1:15 AM

i see, i would have 2 4090s, exactly the same manufacturer

Tsubajashi (ツバジャシ)•6/23/24, 1:15 AM

so i guess thatll be a bit harder?

HikariKnight•6/23/24, 1:17 AM

yeah since you need to make a custom script to manually bind the correct card by pci address to vfio-pci at boot before nvidia driver grabs the card as vfio-pci only understands vendorid:deviceid
which would be identical for the 2 cards

Tsubajashi (ツバジャシ)•6/23/24, 1:17 AM

hmm... i see

Tsubajashi (ツバジャシ)•6/23/24, 1:17 AM

guess theres a bunch of preparation i will do before the pc gets to me next week

HikariKnight•6/23/24, 1:18 AM

not something we support sadly, if you know how to set it up feel free to PR but i have no way to test it since i just do different cards as it is 1000x easier

Tsubajashi (ツバジャシ)•6/23/24, 1:18 AM

will definitely do a PR if i figure it out

HikariKnight•6/23/24, 1:18 AM

i use amd on host and nvidia on VM

Tsubajashi (ツバジャシ)•6/23/24, 1:18 AM

its more of an idea than something i instantly prep

HikariKnight•6/23/24, 1:19 AM

wendell has a guide on level1tech for debian based systems, no idea how to do the same in fedora tbh

Tsubajashi (ツバジャシ)•6/23/24, 1:19 AM

like, i just dont have any real usecase where i would have a good use for the second 4090 (my boss sends me the PC as a gift), so i figure something out where i can put it to good use

Tsubajashi (ツバジャシ)•6/23/24, 1:19 AM

and given there are some new windows games that dont yet work (as far as i know) i wanna try out, it may be worth a shot

HikariKnight•6/23/24, 1:19 AM

sell one of the 4090s and get an amd card?

HHikariKnight nvidia test VM (with gpu passthrough), main machine, HTPC (stuck on F39 due to p...

M2•6/23/24, 1:20 AM

About to join you with the multi-gpu passthrough.

MM2 About to join you with the multi-gpu passthrough.

HikariKnight•6/23/24, 1:20 AM

oh you are?

HHikariKnight sell one of the 4090s and get an amd card? 🤣

Tsubajashi (ツバジャシ)•6/23/24, 1:20 AM

i mean yea i could do that, but in case i would have a usecase for it, it most probably is for AI. ROCm is.... questionable to use tbh

Tsubajashi (ツバジャシ)•6/23/24, 1:20 AM

it has gotten better, but its still shit compared to what nvidia offers

HikariKnight•6/23/24, 1:20 AM

amd on host, nvidia in VM is how i do it

HikariKnight•6/23/24, 1:21 AM

however linux vm would need a physical monitor or rely on sunshine (without mesa breaking it)

Tsubajashi (ツバジャシ)•6/23/24, 1:21 AM

would probably be the other way around for me, i intent to use atleast one nvidia gpu on containers

Tsubajashi (ツバジャシ)•6/23/24, 1:22 AM

and WSL still has a weird design flaw where it doesnt release ram it used previously

HHikariKnight however linux vm would need a physical monitor or rely on sunshine (without mesa...

M2•6/23/24, 1:23 AM

Dummy plug

M2•6/23/24, 1:24 AM

Best thing ever

MM2 Dummy plug

HikariKnight•6/23/24, 1:24 AM

linux vm will need dummy plug + sunshine or physical monitor
no looking-glass host for linux (no maintainer for that portion and its currently broken)

M2•6/23/24, 1:24 AM

Oh. Yeah for Linux I always used a monitor

HikariKnight•6/23/24, 1:27 AM

im glad we made passthrough pretty simple to setup here since we have set clear guidelines
2 different cards (must have different vendor:devicevendor:device ids and have good iommu groups) and rest should in most cases just work

HikariKnight•6/23/24, 1:27 AM

if you want anything else, youre on your own, but we enabled what is required for you

M2•6/23/24, 1:28 AM

So I'm going to do it with zero layering.... Hopefully.....

M2•6/23/24, 1:28 AM

The only gotcha I have is trying to do the vfio stub for the second gpu

M2•6/23/24, 1:29 AM

But the goal is only kargs and containers

MM2 So I'm going to do it with zero layering.... Hopefully.....

HikariKnight•6/23/24, 1:29 AM

you will have to have dracut regeneration enabled unless you do all that required stuff in build time like you do the initramfs stuff in bluefin (which we now use in bazzite)