GeForce RTX 3060 Ti arrives December 2, hits RTX 2080 SUPER level performance

Quoting: x_wing

Quoting: PJHonestly - I don't even know what you're talking about. I know just one thing - with Nvidia regular desktop user has OpenGL / Vulkan / CUDA / OpenCL working well out of the box (and sadly no Wayland). With Amd I don't - I need to install AMDGPUPRO.

What I mentioned is the steps you have to do in order to get "good OpenCL" running in your system. Depending on your distro, you may or may not get the packages from the repo. But still, is quite simple to install using AMD packages.

Quoting: PJWhen I'm talking about mesa opengl is not good for creatives I mean it regularly fails in professional creative apps (Maya, Modo, Substance, Resolve etc). Often those apps don't work at all (with Modo I've been able to report mesa related issues and Modo was tweaked to work with it). Also when I've used amd mesa drivers it was the only time I've encountered hard lockups on Linux.

As I don't use those type applications, I cannot give an answer regarding of which could be the problem there. But, once again, for OpenCL the AMDGPU-PRO drivers work perfectly fine.

Quoting: PJI agree it shouldn't. But if you're creative it is. If you read my post you'll notice that I've specifically pointed out if you're a regular desktop user and a gamer mesa can be enough.
But if you're getting our of that comfort zone more than often it isn't.
I've managed a workstation with AMDGPU-PRO and I'm just reporting my findings. More than often after a kernel update I had to fix my box by reinstalling the driver. I get it that there may be a way to set it up better, so those won't break that easily / will get rebuilt. But that requires knowledge and setup that a creative / regular user shouldn't have to have. It's really not a way to expand linux user base.

It requires the same knowledge as with Nvidia. Both proprietary drivers ships their kernel modules using DKMS (with AMD not requiring a blob binary) so the risk to getting a broken system after a kernel update is exactly the same on both, with the difference that even if your DKMS build/installation process fails on a kernel update on your AMD system you will probably still be getting a working system if the builtin kernel drivers are new good enough for your GPU.

Either way, in both cases the best thing to do is to use a LTS kernel version as long as you can.

Quoting: PJI mean repos for distros like Ubuntu , OpenSUSE etc... And I don't care whether they're maintained by Nvidia or other organization. I'm just saying that from an average Joe perspective those are easier to handle. You enable the repo and you stop worrying about the driver - and again that it my experience, haven't had any major Nvidia driver related isssues for years.

AFAIK, those repos are maintained by the community, just like the community also have special repositories for the latest stable or bleeding edge versions of Mesa. If the average Joe knows how to install the latest Nvidia driver on a distro, he should also be able to keep to date the AMD counter part as it will require exactly the same steps.

I really fail to understand why so many Nvidia users says that AMD driver support is inferior when they are actually providing the type of solution as Nvidia and more.

I think what you fail to see is that Nvidia drivers are integrated, as latest version, in most distros. Most of the time with akmods or equivalent, which means you never have to deal with DKMS and reboot issues you talk about.
For AMD, when the kernel is okay (so, not when new card, new/latest Vulkan support etc), same for Mesa, which is locked to 3 month release cycle. As such, for nvidia, when a new release come out I just have to time "sudo dnf update"/ or "update"tickbox in sw and I am done. It's official repo, not third party. It requires literally 0 knowledge.
With AMD, you need to go the road of git kernel/ git Mesa with third party or self tinker. And upgrading kernel outside of distro cycle can/will cause issues at one point or another... As such, Nvidia way is usually more safe.
As you said, AMD provides also a dkms, but dkms is just bad, it fails at each new kernel release and is an absolute horror. Whereas, while it's true that Nvidia provides dkms on their website, it is thankfully not what is provided by distros - and surprise, never had issues with akmods, defo had with dkms. Also, for AMD is rarely just kernel/dkms, is also mesa git etc, which exposes you to even more issues.
For nvidia, at most, you just pull the rpm for Nvidia driver, and change the wget url if you need absolutely immediately, or you just wait tomorrow when it will be released in distro.

Now, once again, Nvidia is not perfect either, esp on compute side. But on gaming, it is click and play. New card support is day one/two, without tinker, without forceful updates etc.
As for the creatives/ML guys, Nvidia is defo the simplest road, by far and large. Support is once again basically out of the box, and you get CUDA which is often a hard requirements + the tensor cores of course. With AMD, you once again go the tinker road, and forget CUDA.

Quoting: x_wing[...]
I really fail to understand why so many Nvidia users says that AMD driver support is inferior when they are actually providing the type of solution as Nvidia and more.

Based purely on anecdotal observation, I suspect it's because there's been many HOWTO guides for NVIDIA drivers written by "mainstream" news sites and its often mentioned in Youtube videos, whereas I have yet to see anyone mention how to set things up with AMD.

Quoting: 3zekielI think what you fail to see is that Nvidia drivers are integrated, as latest version, in most distros. Most of the time with akmods or equivalent, which means you never have to deal with DKMS and reboot issues you talk about.

That's not true. Even such packages can mess up your system big time. I had such experience with Nvidia in the past. Every time you install a new kernel you have to rebuild dkms for Nvidia, which the package itself can support. But there is no guarantee it it would work correctly any more than for AMD.

Distros don't need to focus on such solution for AMD with packages, because AMD provide the driver in upstream kernel. If you want to go out of your way and deal with dkms mess - you always can. But do it on your own. With Nvidia that's the only solution, unless you want to use nouveau.

So I agree with x_wing. Nvidia doesn't offer any more convenience than AMD installation wise. It actually offers less.


Last edited by Shmerl on 4 December 2020 at 3:37 pm UTC

Quoting: 3zekielI think what you fail to see is that Nvidia drivers are integrated, as latest version, in most distros. Most of the time with akmods or equivalent, which means you never have to deal with DKMS and reboot issues you talk about.
For AMD, when the kernel is okay (so, not when new card, new/latest Vulkan support etc), same for Mesa, which is locked to 3 month release cycle. As such, for nvidia, when a new release come out I just have to time "sudo dnf update"/ or "update"tickbox in sw and I am done. It's official repo, not third party. It requires literally 0 knowledge.

Just like with AMD if you use your distro official repo, you get official support for certain versions of their packages (which version? well, that depends of your distro). Lets be honest here: it is not Nvidia how give this easiness, is your distro maintainers that do such thing. So, Nvidia release a driver that is installed using DKMS or akmods... which is probably the same that AMDGPU-PRO provides (And I say "probably" because I never used rpm packages of AMDGPU-PRO, but I would guess that they use akmods).

Quoting: 3zekielWith AMD, you need to go the road of git kernel/ git Mesa with third party or self tinker. And upgrading kernel outside of distro cycle can/will cause issues at one point or another... As such, Nvidia way is usually more safe.

You only have to do that if you need the latest bits of AMDGPU and the same thing can happen in your distro if you use Nvidia. In order to get the latest version of your driver you will probably have to switch to an unstable/third party repository, completely removing the "safety" you mention as it is possible that your distro upgrades to a kernel that doesn't work with your Nvidia driver.

Quoting: 3zekielAs you said, AMD provides also a dkms, but dkms is just bad, it fails at each new kernel release and is an absolute horror. Whereas, while it's true that Nvidia provides dkms on their website, it is thankfully not what is provided by distros - and surprise, never had issues with akmods, defo had with dkms. Also, for AMD is rarely just kernel/dkms, is also mesa git etc, which exposes you to even more issues.

Not sure what you mean here. What the distros provides is what Nvidia releases with changes in order to make it work with the distro kernel (if necessary). In the other hand, using dkms/akmods is completely optional on AMD but you can do exactly the same if you want (any distro can adapt the released AMD packages to their distro and use it if they want). The difference is that is far easier to simply update your kernel instead of dealing with a dynamic module. As you say, it's a absolute horror that will probably fail with each new kernel release... which is a common problem on Nvidia.

As I said before: you should only use the dynamic module for AMD GPUs if you can't upgrade your kernel.

BTW, if in your opinion dmks is just bad, then Nvidia is also just bad as this is the way their driver is installed on Debian based distros.

Quoting: 3zekielThey version that you will want to use completely depends
For nvidia, at most, you just pull the rpm for Nvidia driver, and change the wget url if you need absolutely immediately, or you just wait tomorrow when it will be released in distro.

And the same can be said for each release of the kernels. How easier you can upgrade to the latest bits of your GPU driver is completely dependent of your distro maintainers, so giving the credit of this to Nvidia is just lame. My point here is that AMD provides the exact same options as Nvidia plus the flexibility of avoiding the DKMS modules all together and use the builtin driver of the latest kernel if the user wants (which is in fact the best solution).

I find hard to believe that getting the latest kernel or Mesa version for your distro is way more difficult than getting the latest Nvidia drivers.

Quoting: 3zekielAs for the creatives/ML guys, Nvidia is defo the simplest road, by far and large. Support is once again basically out of the box, and you get CUDA which is often a hard requirements + the tensor cores of course. With AMD, you once again go the tinker road, and forget CUDA.

Once again, this "simplest road" is related to the work that the maintainers do on the distro. If you compare what Nvidia provides and what AMD provides with their proprietary drivers you will find that both gives the exact same stuff.


Last edited by x_wing on 4 December 2020 at 3:57 pm UTC

Quoting: CybolicBased purely on anecdotal observation, I suspect it's because there's been many HOWTO guides for NVIDIA drivers written by "mainstream" news sites and its often mentioned in Youtube videos, whereas I have yet to see anyone mention how to set things up with AMD.

It's probable, but that doesn't justify the common idea that Nvidia gives better support for driver installation. IMO, there is a big bias going on regarding AMD drivers.


Last edited by x_wing on 4 December 2020 at 4:01 pm UTC

Ha, I love the argument that it is simpler to swap out a kernel instead of using dkms to automatically compile the driver when there is an automatic kernel update...

If you are currently up to date on Debian Sid, then the AMD should work out of the box, but of course won't have any features that are in the AMDGPU PRO driver. Whereas with Nvidia cards, you will have to enable non-free repo and install 'nvidia-driver' package. But once you do that, it just works and they keep it well up to date. They don't always get the latest Mesa and sometimes (especially after a stable release) don't have the latest kernel.

There are plenty of advantages and disadvantages to each method. Neither is entirely better than the other.

In a perfect world, they would just work. Like maybe some sort of auto detection that could hit kernel.org or other once network pops up and looks to see if there is a minimal driver combination to be needed to get Xorg / Wayland running... but I guess we have things like fbdev for that.

Quoting: slaapliedjeWhereas with Nvidia cards, you will have to enable non-free repo and install 'nvidia-driver' package. But once you do that, it just works and they keep it well up to date.

It works, until it doesn't because it's not compatible with the latest kernel which you might need for completely unrelated reason. Leaving you to scratch your head, how to make your other hardware work and have a working Nvidia driver.

So it's clearly not "just works".


Last edited by Shmerl on 4 December 2020 at 4:07 pm UTC

Quoting: slaapliedjeHa, I love the argument that it is simpler to swap out a kernel instead of using dkms to automatically compile the driver when there is an automatic kernel update...

What is the difficulty of swapping kernels?

Quoting: slaapliedjeIf you are currently up to date on Debian Sid, then the AMD should work out of the box, but of course won't have any features that are in the AMDGPU PRO driver. Whereas with Nvidia cards, you will have to enable non-free repo and install 'nvidia-driver' package. But once you do that, it just works and they keep it well up to date. They don't always get the latest Mesa and sometimes (especially after a stable release) don't have the latest kernel.

That your distro doesn't provides the latest stable kernel or Mesa packages is problem with your distro. We are talking about of what Nvidia and AMD provides.

Quoting: slaapliedjeThere are plenty of advantages and disadvantages to each method. Neither is entirely better than the other.

In a perfect world, they would just work. Like maybe some sort of auto detection that could hit kernel.org or other once network pops up and looks to see if there is a minimal driver combination to be needed to get Xorg / Wayland running... but I guess we have things like fbdev for that.

Neither is better than the other, but somehow many users says that Nvidia still provides a better driver support.

Quoting: Shmerl

Quoting: slaapliedjeWhereas with Nvidia cards, you will have to enable non-free repo and install 'nvidia-driver' package. But once you do that, it just works and they keep it well up to date.

It works, until it doesn't because it's not compatible with the latest kernel which you might need for completely unrelated reason. Leaving you to scratch your head, how to make your other hardware work and have a working Nvidia driver.

So it's clearly not "just works".

The funny thing is, that used to be AMD, where their driver didn't work with the latest kernels and didn't work with the latest Xorg, and it'd take them months to fix it. There has only been the recent one with nvidia where that was ever the case, and that was the fault of the Linux kernel as they broke it intentionally.

I have been using nvidia for years, and it has pretty much been 'just works'.

Quoting: x_wing

Quoting: slaapliedjeHa, I love the argument that it is simpler to swap out a kernel instead of using dkms to automatically compile the driver when there is an automatic kernel update...

What is the difficulty of swapping kernels?

Really? Swapping kernels out for something you compiled yourself or fetched from some third party seems great to you??

Quoting: slaapliedjeIf you are currently up to date on Debian Sid, then the AMD should work out of the box, but of course won't have any features that are in the AMDGPU PRO driver. Whereas with Nvidia cards, you will have to enable non-free repo and install 'nvidia-driver' package. But once you do that, it just works and they keep it well up to date. They don't always get the latest Mesa and sometimes (especially after a stable release) don't have the latest kernel.

Quoting: x_wingThat your distro doesn't provides the latest stable kernel or Mesa packages is problem with your distro. We are talking about of what Nvidia and AMD provides.

There really isn't such a thing as 'latest' stable kernel, when you're running a stable distribution such as RH/CentOS, Debian Stable, Ubuntu LTS, etc. Sometimes you just don't want to be running unsupported configurations, which running the 'latest stable' kernel wouldn't be.

Quoting: slaapliedjeThere are plenty of advantages and disadvantages to each method. Neither is entirely better than the other.

In a perfect world, they would just work. Like maybe some sort of auto detection that could hit kernel.org or other once network pops up and looks to see if there is a minimal driver combination to be needed to get Xorg / Wayland running... but I guess we have things like fbdev for that.

Quoting: x_wingNeither is better than the other, but somehow many users says that Nvidia still provides a better driver support.

Yeah, they're just different, and as you said, sometimes it's just a matter of which distribution you use, on which is supported better. Unless you end up with some sort of AMD Linux distribution where they keep it up to date for future cards and prepares them for their users, then there is no 'perfect' solution at this point. Now maybe if Intel does something amazing with their dedicated GPUs, and makes the drivers compatible, then there'll be a better experience. But even then you'd still be tied to newer Mesa drivers, which is usually the sore point to the AMD ones.