I am going to buy a new graphics card and can’t choose between Nvidia and AMD. I know that Nvidia has bad reputation in Linux community but how really it works? And I heard recently their drivers got better. What can you recommend?
P. S. I don’t want any proprietary drivers (so I am talking about Nouveau or any other FOSS Nvidia driver if it exists)
If those are your criteria, I would go with AMD right now, because only the proprietary driver will get decent performance out of most nVidia cards. Nouveau is reverse-engineered and can’t tap into a lot of features of newer cards especially, and while I seem to recall there is a new open-source driver in the works, there’s no way it’s mature enough to be an option for anyone but testers.
I have no beef in this argument, and I’m certainly not biased in relation to AMD/Nvidia. However, my 980Ti, my 2070S and now my 4070S have all run really well under Linux. I run KDE Neon and a quick ‘sudo apt install nvidia-driver-570’ installs the latest beta’s in under 5 mins, if I want to roll back the driver a quick ‘sudo apt install nvidia-driver-565’ has me back on the latest feature branch. Yeah, Wayland adoption under Nvidia was slow, and Nvidia’s earlier choices weren’t what anyone could call ‘ideal’ - But momentum is building, and as a result I’ve been using Wayland for about eight months now without issue. Before that, X11 was largely faultless running Nvidia hardware/drivers.
People say Nvidia struggle in relation to VKD3D performance. I’m not too sure what they’re doing, but VKD3D runs fine here.
It’s the one advantage we have over Mac users: We can run AMD, Intel and Nvidia. We also have ongoing OGL support, native Vulkan support, better game support under Steam, a larger user base under Steam, and the amazing Proton implementation.
Whether it be AMD or Nvidia, I personally think it’s Linux for the win. EDIT: I in no way see value for money in the new 5080/5090 cards and I eagerly await what AMD has to offer (although I won’t be switching from my 4070S for quite some time yet).
If you’re unwilling to use proprietary drivers AMD or Intel if yout friend. If you use proprietary drivers NVIDIA is mostly fine now.
The only reason I still go Nvidia is because I self host AI, which afaik takes advantage of CUDA and just runs overall better on Nvidia cards, or at the very least is easier to set up. Really, the top reason is that it’s the devil I know right now.
If I didn’t self host AI, I would 100% go AMD. Especially if you don’t want to use proprietary drivers. That being said, my old gaming laptop runs NixOS with Nouveau and there have definitely been improvements since I first tried it years ago, but I don’t do much gaming on it. It’s more a TV media station these days (so I can avoid the stupid smart TV bloat agenda, where your TV gets gradually slower and fits less increasingly-bloating apps over time).
If it’s just about self-hosting and not training, ROCm works perfectly fine for that. I self-host DeepSeek R1 32b and FLUX.1-dev on my 7900 XTX.
You even get more VRAM for cheaper.
This is very good to know. I read that ROCm can be a pain to get up and running, but I read that months ago and this space is moving fast. I may switch over when I can if this is the case. My 3080 is feeling it’s age already. Thank you!
That used to be the case, yes.
Alpaca pretty much allows running LLM out of the box on AMD after installing the ROCm addon in Discover/Software. LM Studio also works perfectly.
Image generation is a little bit more complicated. ComfyUI supports AMD when all ROCm dependencies are installed and the PyTorch version is swapped for the AMD version.
However, ComfyUI provides no builds for Linux or AMD right now and you have to build it yourself. I currently use a simple Docker container for ComfyUI which just takes the AMD ROCm image and installs ComfyUI ontop.
Definitely bookmarking this reply. I haven’t tried ComfyUI yet, but I’ve had it starred on Github from back when it was fairly new. I’m no stranger to building from source, but I have not dived into Docker yet, which is becoming more and more of a weakness by the day. Docker is sometimes required by some really cool projects and I’m missing out.
ROCm
I’m curious. Say you are getting a new computer, put Debian on, want to run e.g. DeepSeek via ollama via a container (e.g. Docker or podman) and also play, how easy or difficult is it?
I know that for NVIDIA you install the (closed official) drivers, setup the container insuring you get GPU passthrough, and thanks to CUDA from the driver, you’re pretty much good to go. Is it the same for AMD? Do you “just” need to install another package or is there more tinkering involved?
I’m curious. Say you are getting a new computer, put Debian on, want to run e.g. DeepSeek via ollama via a container (e.g. Docker or podman) and also play, how easy or difficult is it?
On the host system, you don’t need to do anything. AMDGPU and Mesa are included on most distros.
For LLMs you can go the easy route and just install the Alpaca flatpak and the AMD addon. It will work out of the box and uses ollama in the background.
If you need a Docker container for it: AMD provides the handy
rocm/dev-ubuntu-${UBUNTU_VERSION}:${ROCM_VERSION}-complete
images. They contain all the required ROCm dependencies and runtimes and you can just install your stuff ontop of it.As for GPU passthrough, all you need to do is add a device link for
/dev/kfd
and/dev/dri
and you are set. For example, in a docker-compose.yml you just add this:devices: - /dev/kfd:/dev/kfd - /dev/dri:/dev/dri
For example, this is the entire Dockerfile needed to build ComfyUI from scratch with ROCm. The user/group commands are only needed to get the container groups to align with my Fedora host system.
spoiler
ARG UBUNTU_VERSION=24.04 ARG ROCM_VERSION=6.3 ARG BASE_ROCM_DEV_CONTAINER=rocm/dev-ubuntu-${UBUNTU_VERSION}:${ROCM_VERSION}-complete # For 6000 series #ARG ROCM_DOCKER_ARCH=gfx1030 # For 7000 series ARG ROCM_DOCKER_ARCH=gfx1100 FROM ${BASE_ROCM_DEV_CONTAINER} RUN apt-get update && apt-get install -y git python-is-python3 && rm -rf /var/lib/apt/lists/* RUN pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/nightly/rocm6.3 --break-system-packages # Change group IDs to match Fedora RUN groupmod -g 1337 irc && groupmod -g 105 render && groupmod -g 39 video # Rename user on newer 24.04 release and add to video/render group RUN usermod -l ai ubuntu && \ usermod -d /home/ai -m ai && \ usermod -a -G video ai && \ usermod -a -G render ai USER ai WORKDIR /app ENV PATH="/home/ai/.local/bin:${PATH}" RUN git clone https://github.com/comfyanonymous/ComfyUI . RUN pip install -r requirements.txt --break-system-packages COPY start.sh /start.sh CMD /start.sh
Very cool, thanks for the in depth explanation.
Honestly even on Windows I preferred AMD’s software suite compared to Nvidia control panel and GeForce Experience. Currently using a 7900XTX and pretty happy with it. Also I missed Radeon Chill when I was on Nvidia, didn’t expect to care about that at all, but I love it.
I don’t want any proprietary drivers (so I am talking about Nouveau or any other FOSS Nvidia driver if it exists)
In that case AMD, no doubt about it.
If you were considering proprietary drivers it would still be AMD but there would be some discussion about it.
The nouveau drivers are just barely enough to have a desktop, anything actually needing a GPU will perform very poorly (in my anecdotal experience with 4K). Or, to put it another way, choosing an NVIDIA card is choosing their proprietary drivers.
So you’re left with AMD (and Intel). The open amdgpu driver is pretty good and is suitable for gaming. Which I do.
I have no experience with Intel, but I believe their open drivers are pretty good.
So I recommend AMD.
Both work, just in different ways. I think AMD’s value proposition is better on Linux but if you were choosing between a 6700XT and a 4080 (for sake of example) of course the latter is still gonna be faster despite the drivers being a bit weirder to manage
My two cents.
I have quite a few Nvidia GPUs I still use (2080,3080ti,3090) but recently purchased two AMD cards. I have a 5700xt and 7800xt.
I recently started using Universal Blue Linux as my daily driver on most of my systems. Bluefin for my desktop with Nvidia, Bazzite for my gaming PC with AMD.
They do both work however I have still had more issues with NVIDIA than AMD. For example, running games tends to be buggier but that is specifically an Nvidia driver issue. I’m guessing most hot fixes come out for the windows driver first. For instance, FF7 Rebirth does not render world geometry on Nvidia on Linux. I do not have this problem under AMD
I started purchasing the AMD cards because I was growing tired of waiting for Nvidia stability on Linux.
Is it much better than it was before , yes Do you use Nvidia CUDA apps or AI? Check, that works! Is it still as smooth and seamless as AMD, nope, you’re still going to end up with regressions.
I think it’s only a matter time before Nvidia finally figured this out as they heavily rely on Linux as a platform in their own work. But right now your best user experience overall is going to be on AMD hardware.
From what i’ve heard if your not willing to use the nvidia proprietary drivers then DON’T go for nvidia you will get terrible performance and amd will always be significantly better.
If you consider the proprietary drivers then I think it depends on your use case. For example AMD is better value if your gaming without ray tracing if you want to play with ray tracing or do any kind of productivity Nvidia is generaly the better option. For machine learning Nvidia has much better compatibility with everything so you will have a better time and better performance, Although if you only care about running the largest models you can with the available vram then AMD gpu’s will have more vram for the price.
Intel arc is also always an option if you are aiming for a lower tire/mid range card. They have really price competitive cards and unlike amd they have very decent ray tracing and productivity capability’s. They also have lots more vram for the price compared to Nvidia.
Also I highly recommend buying a used graphics card, you help the environment, save a lot of money and if you don’t like the card you chose you can sell it for the same price your bought it and buy a different one.
Maybe if you could specify your use case and what cards you are currently looking at I could help you out more.
I have 2 PCs, both on Linux. One with an AMD XTX 7900 XT, the other one has an Nvidia 3080 TI.
The Nvidia one is running the latest proprietary drivers, and they suck HARD. They just are far inferior to AMD’s. The only reason to go Nvidia is to do local AI or video (editing / transcoding).
If your primary use is gaming and go Nvidia, you will be sabotaging yourself.
If you want Nvidia Reflex,DLSS and RTX and GSYNC,etc and your fine with installing out of tree proprietary drivers and fine with some minor issues(Like rarely breaking randomly) Nvidia If you don’t care about Nvidias features AMD.
In my experience older nvidia cards (~5 years old +) work fine, newer ones are very hit-or-miss
Amd cards of any age work pretty much perfectly as far as I can tellThough if the drivers not being proprietary is a hard line for you then amd is your only option really
Could be game specific, but there is no ground rendering in final fantasy. https://youtu.be/DxE-4ZxYxDA
A workaround’s been developed for the issue regarding FFVII.
https://github.com/ValveSoftware/Proton/issues/8408#issuecomment-2657340142
AMD cards work great with the open source driver. As i understand it, the nouveau driver is getting better but might not be there yet? So if the non-proprietary driver is a must you might be better off with AMD.