NVIDIA's v580 Driver Branch Ends Support for Maxwell, Pascal, and Volta GPUs

[Tomorrow]

So it depends on what you call "support". No features? Sure. Some fixes, security updates and support in new systems? That may last until 2027~2029.

AMD has "support" for Vega the same way since 2022 but people beat AMD over the head without acknowledging this.
In their eyes AMD "abandoned" Vega. They also ignore 5 years of prior support.

I don't really give a crap about Windows usage

You might not, but ~70% of PC users that are on Windows, might.

 

[Dr. Dro]

AMD has "support" for Vega the same way since 2022 but people beat AMD over the head without acknowledging this.
In their eyes AMD "abandoned" Vega. They also ignore 5 years of prior support.

You might not, but ~70% of PC users that are on Windows, might.

Let me put this in the simplest of terms: Vega's code base has been maintained for security updates and very minor bug fixes. It is done, feature-complete and no further major changes are to be expected from here on out. This is the exact same "support" scheme that Nvidia has been doing for the Kepler architecture until recently, I believe the last one is the 475.14 driver from July 2024, one year ago.

The code base itself has not been updated since around late 2023. It's still the old 23.19 driver branch, even in newer releases such as 25.5.1. Those two fixes, btw? Everything that's been done to this driver since late 2024.

So when people call Vega "unsupported", they're calling it for what it really is. Its driver code has nothing in common with the 25.5.1 for RDNA+ (version 25.10.01.09), which is on a much much newer driver branch. AMD's versioning number here is misleading, even if it's not intentionally so (and I genuinely believe that it's not intentional) - they just do year.month.release to make it simple to figure it out from roughly when the driver is from.

 

[igormp]

AMD has "support" for Vega the same way since 2022 but people beat AMD over the head without acknowledging this.
In their eyes AMD "abandoned" Vega. They also ignore 5 years of prior support.

So what? Have you seen me talking about this? If not, I believe this reply of yours to me makes no sense whatsoever.
I don't care about AMD in general, and the only development I follow they doing is linux-related. Afaik vega is still pretty well supported by AMDGPU, just not by ROCm.

You might not, but ~70% of PC users that are on Windows, might.

And those should look for news related to Windows, not try to assume things related to Linux, which have different things going on on that side.

 

[dgianstefani]

And those should look for news related to Windows, not try to assume things related to Linux, which have different things going on on that side.

Although this announcement originates from NVIDIA's UNIX documentation, the unified driver codebase means Windows users will face the same cutoff.

From the article.

 

[igormp]

From the article.

I'm sorry, but @AleksandarK is not a proper source. They wrote it out of their own will, and this is not written in the original forum post from Nvidia.
As I had mentioned before, the codebase is unified but up to a point. It's likely that this will be the case, but spreading this as a de facto confirmation without an official announcement is just spreading rumors and misinformation.
Heck, even the rumor mill that VideoCardz is, and which you used as one of your sources worded it in a better way:

As a result, NVIDIA will likely stop releasing Game Ready drivers for one of the last remaining GeForce GTX product lines.

 

[AleksandarK]

I'm sorry, but @AleksandarK is not a proper source. They wrote it out of their own will, and this is not written in the original forum post from Nvidia.
As I had mentioned before, the codebase is unified but up to a point. It's likely that this will be the case, but spreading this as a de facto confirmation without an official announcement is just spreading rumors and misinformation.
Heck, even the rumor mill that VideoCardz is, and which you used as one of your sources worded it in a better way:

Unified codebases are a thing, just different compilation targets, with some OS specific quirks in the code. When the driver upade stops, it stops for all platforms usually. Nothing new there.

 

[igormp]

Unified codebases are a thing, just different compilation targets, with some OS specific quirks in the code. When the driver upade stops, it stops for all platforms usually. Nothing new there.

I agree with what you said for the most part, but the driver in question it's unified up to a point, and it's not just a matter of "compilation targets" or "OS specific quirks".
One clear example of that is the example I've told more than once, the nvidia proprietary drivers vs the nvidia-open ones. Same target system, same compilation target, different kernel modules.
Adding to the above, and as far as I'm aware, most of the "unified" part is related to the userspace drivers. The kernel ones from Windows are really different from the Unix ones.

The whole difference when it comes to this v580 being the latest one for linux is that Nvidia will simply drop support for the non-open driver, with the nvidia-open only support GSP-enabled GPUs.

 

[Visible Noise]

What other driver branches?

Look it up. I don’t spoon feed people.

 

[Tomorrow]

Look it up. I don’t spoon feed people.

If you're basing your argument on that, then it's up to you to at least provide a rudimentary overview of what you mean.

But fine. Since you're so lazy i decided to be equally lazy and ask AI:

Branch Type	Description
Mainline (e.g., R570)	Covers recent GPUs (900 series and up), includes GRD and SD builds.
Legacy (e.g., R470)	Maintained separately for older GPUs (600/700 series), security updates only.
Studio/Game Ready	Same base driver, different packaging and testing emphasis.
Quadro/Enterprise	Separate release cadence and certification for pro apps.
Special branches	Vulkan beta, CUDA dev, OEM-modified releases.

 

[JustBenching]

Let me put this in the simplest of terms: Vega's code base has been maintained for security updates and very minor bug fixes. It is done, feature-complete and no further major changes are to be expected from here on out. This is the exact same "support" scheme that Nvidia has been doing for the Kepler architecture until recently, I believe the last one is the 475.14 driver from July 2024, one year ago.

The code base itself has not been updated since around late 2023. It's still the old 23.19 driver branch, even in newer releases such as 25.5.1. Those two fixes, btw? Everything that's been done to this driver since late 2024.

View attachment 406313

So when people call Vega "unsupported", they're calling it for what it really is. Its driver code has nothing in common with the 25.5.1 for RDNA+ (version 25.10.01.09), which is on a much much newer driver branch. AMD's versioning number here is misleading, even if it's not intentionally so (and I genuinely believe that it's not intentional) - they just do year.month.release to make it simple to figure it out from roughly when the driver is from.

Wish it was that simple and evidence convinced people. Nope, not when it comes to amd it won't.

 

[Visible Noise]

If you're basing your argument on that, then it's up to you to at least provide a rudimentary overview of what you mean.

But fine. Since you're so lazy i decided to be equally lazy and ask AI:

Branch Type	Description
Mainline (e.g., R570)	Covers recent GPUs (900 series and up), includes GRD and SD builds.
Legacy (e.g., R470)	Maintained separately for older GPUs (600/700 series), security updates only.
Studio/Game Ready	Same base driver, different packaging and testing emphasis.
Quadro/Enterprise	Separate release cadence and certification for pro apps.
Special branches	Vulkan beta, CUDA dev, OEM-modified releases.

This ought to rock your world:
570 and 575 driver branch updated on the same day.
535 and 550 branches getting updates after 570 branch.
Etc.

Linux AMD64 Display Driver Archive | NVIDIA

Linux AMD64 Display Driver Archive

www.nvidia.com

 

[Easo]

My 1070 Ti has served me VERY well, time to retire it with honours.

 

[agent_x007]

Wasn't Linux Pascal support... problematic ?
(due to it needing GRD driver but never getting it, and public one being more Maxwell 2.0 focused instead of Pascal [due to internal "black box" inside Pascal architecture])

 

[igormp]

Wasn't Linux Pascal support... problematic ?
(due to it needing GRD driver but never getting it, and public one being more Maxwell 2.0 focused instead of Pascal [due to internal "black box" inside Pascal architecture])

No? Are you confusing Nvidia's driver with nouveau?
Nouveau lacks the ability to do recloking due to the lack of firmware signing on post kepler.
I never had any issues with pascal when I used it on Linux with Nvidia's proprietary driver.