- Registrado
- 28 de Ago, 2015
The stock answer is that there's code for at least the Intel NPU in some acceleration API that Linux has.
The subsequent questions I've seen about 'okay but what uses that API' were met with silence.
The Linux Thread - The Autist's OS of Choice
- Iniciador del tema tehpope
- Fecha de inicio
-
🔧 Site instability resolved. You can report double-posts and broken attachments. For bigger issues, use the Technical Grievances thread.
🇵🇦 Nuestro primer dominio localizado está en español en kiwifarms.pa. Our first localized domain is on Spanish on kiwifarms.pa. -
Want to keep track of this thread?
Accounts can bookmark posts, watch threads for updates, and jump back to where you stopped reading.
Create account
The subsequent questions I've seen about 'okay but what uses that API' were met with silence.
- Registrado
- 17 de Nov, 2023
Patches welcome, lol.
I am seriously considering selling my 4060Ti and buying RX 6800 instead. From what I've seen I may get 20% uplift in performance for cheaper with better driver support. So has AMD reached "it just works" territory on Linux? I am not bothered by ROCM much, and I don't give a shit about gaytracing one bit.
- Registrado
- 21 de Nov, 2020
For what?
I "upgraded" to a 7900 XTX and I have audio issues and AI stuff is still a pain in the ass, although improving, over my previous 3070.
- Registrado
- 28 de Ene, 2018
If you want to do anything AI, nvidia is pretty much the only option. If you don't, AMD is probably the better value, especially in Linux.
- Registrado
- 19 de Dic, 2022
Yup. My experience has been as good as with Intel, plug and play.
If you’re willing to tinker, AMD through rocm-PyTorch works “just fine”. The performance sometimes isn’t there but the setup has become relatively easy, as long as you don’t mind docker. Pull the correct container, pass through your GPU and whatever folder your AI stuff is in, do the setup in docker, take a snapshot, and then just run the snapshot to have it “just work”. It’s more steps than CUDA, but because it forces you into docker, it’s also less prone to breaking.
- Registrado
- 30 de Mar, 2023
I feel like a 6000 series AMD GPU will have more mature drivers compared to a brand new 7000 series
- Registrado
- 21 de Nov, 2020
I don't think 1.5 years counts as "Brand New". (December 2022)
- Registrado
- 30 de Mar, 2023
that doesn't dispute my point. if you keep nitpicking like that you'll never find a bride.
- Registrado
- 21 de Nov, 2020
I have AI working pretty well on the 7900, thanks.
- Registrado
- 12 de Dic, 2022
I've basically narrowed down my choices to Debian, Fedora or Arch lol
I feel like Fedora is nice and all but dnf feels slow as fuck, if anyone can help with that I might consider using Fedora full time lol
I was also thinking of using Endeavour but I feel like it'd be more satisfying to set up Arch.
I feel like Fedora is nice and all but dnf feels slow as fuck, if anyone can help with that I might consider using Fedora full time lol
I was also thinking of using Endeavour but I feel like it'd be more satisfying to set up Arch.
- Registrado
- 29 de Mayo, 2023
depends on your autism.
fedora if you're a austic redhat faggot.
debian if you're a slow and stable autist,
or arch if you're a loony troon.
pick your poison, it's autistic no matter what you choose.
- Registrado
- 17 de Dic, 2022
- Registrado
- 1 de Abr, 2024
I settled on fedora at least for my newer cheap laptop. It just works really well.
My thinkpad, i'm messing around with a bunch of random stuff right now. Actually endeavor os with i3 window manager was one of them. Ater stumbling learning to do the config its actually really nice to use. But i decided to go ahead and download a bunch of random (well not completely random) distros to pop in on the thinkpad. Since the old i5 processor seems to be really holding it back i might as well test stuff with it until i can get a better one.
I picked garuda linux with i3, and also another iso of it with sway. This one is supposed to be an arch based distro but dialed in for better performance in a few ways.
Also gecko rolling release with gnome. Which is supposed to be a more polished opensuse.
I was going to also download redcore linux which is basically gentoo. But you can skip all the compiling everything yourself part. But the download was taking forever so i canceled it.
As far as installing arch. At least now days. Just run the install arch script. It basically showes you what you need to do. Having that and the wiki for any extra advice makes it not too bad to figure out.
The more difficult things for me so far have been just trying to solve random errors, and bugs that happen. That aren't actually easy to find documentation on.
Also setting up a window manager. It wasnt super hard, i just wasnt used to it. I think i learned more doing that though. (Also. I might be converted to running those, i want to get g-tile on fedora with gnome so its like pop!_os, because at that point it might be perfect, but otherwise I really like the wms)
Última edición:
- Registrado
- 30 de Mar, 2023
That's not really narrowing down your choices, as those are just the upstream distros for completely different design philosophies.
Debian is the slow+steady distro, designed to not really push out new software until its thoroughly tested. Cinnamon is the best match for it as it's a highly stable DE that rarely needs updates to fix things. LMDE (Linux Mint Debian Edition) adds minor improvements that make it easier to use and work with, and Linux Mint provides newer packages and a better application manager and way to handle Nvidia drivers.
Arch Linux is the build-it-yourself distro that gives you full control over your OS whether you want it or not. I'd say KDE best matches it's design philosophy with a "notifications" system that lets you program any action to occur on a trigger (login, device disconnected, etc). The derivatives mostly aim to make installation and use easier, but you can just use Arch with the archfi installation script
Fedora is focused on having the latest and greatest packages, at the cost of those packages being more buggy. GNOME is the default DE for Fedora but it's a garbage DE for garbage people.
- Registrado
- 16 de Sep, 2019
I've been using ROCM on bare metal and it's been functioning pretty well, but yeah, Docker is a much easier install and it's easier to keep ROCM up to date.
My 7800xt has had zero issues on Mint. I only had to change the kernel to a newer version to get it working, but that's what you get for new hardware on Linux. A 6800 shouldn't have any issues.
- Registrado
- 12 de Dic, 2022
KDE Arch/EOS is probably what I'm gonna use tbh if Fedora's packages just suck ass
- Registrado
- 13 de Dic, 2022
One kind of similar example is OpenSSL will take advantage of SHA cpu acceleration extensions. So it comes down to software.
Honestly I don’t really get how the NPU stuff is different enough from GPUs to matter. My limited understanding is it almost seems like a cut down GPU that only does fast matrix math (but honestly I haven’t read up on this at all and this is only from headlines), so just like iGPUs allowed for a windows vista aero fad, the inserting LLMs everywhere fad is in full swing.
But if my random assessment is correct, then if you have a discrete GPU then you have already been able to do whatever the NPU does.
There was that open source voice assistant.
Also just the lazy tavern, LM Studio, stable diffusion.
Canonical may try to push a “Copilot+” into some Ubuntu non-LTS.
Usually fedora, to their credit, implements pretty bleeding edge features (still not going to daily drive it though), so I would look at what they do.
Josie and the Sussy Cats
kiwifarms.net
- Registrado
- 5 de Jun, 2024
Good afternoon, I am a retard.
When I was a boy, my dad loved compiling his own Gentoo installations on various generic PCs scattered throughout the house.
He would spend days and days on Gentoo Linux. I've never seen anything like it.
The above anecdote has no relation to my question:
Will Linux offer me advantages in terms of running machine learning on my PC, or should I just stick to Windows and LM Studio because I'm a tard?
I want to create a system that I can feed public-domain passages to and it generates reading comprehension questions. My system has a 3080 in it so I've got 10 GB VRAM.
When I was a boy, my dad loved compiling his own Gentoo installations on various generic PCs scattered throughout the house.
He would spend days and days on Gentoo Linux. I've never seen anything like it.
The above anecdote has no relation to my question:
Will Linux offer me advantages in terms of running machine learning on my PC, or should I just stick to Windows and LM Studio because I'm a tard?
I want to create a system that I can feed public-domain passages to and it generates reading comprehension questions. My system has a 3080 in it so I've got 10 GB VRAM.
- Registrado
- 29 de Mayo, 2023
honesty, they kinda are the same but they are designed to accelerate different workloads.
gpus are more suited for graphical workloads while npu's are more suited for computational workloads.