Nvidia prepares to move Maxwell, Pascal, and Volta GPUs to legacy driver status

Skye Jacobs

Posts: 584   +13
Staff
In context: Nvidia is preparing to transition its Maxwell, Pascal, and Volta GPU architectures to a legacy driver branch, signaling the end of an era for these iconic products. This transition reflects the company's focus on supporting more recent hardware capabilities, particularly in areas such as AI and ray tracing.

Nvidia's CUDA 12.8 release notes indicate that support for the older architectures is now considered "feature-complete" and will be frozen in an upcoming release.

This move marks a significant shift for Nvidia as it begins to phase out support for the remaining GTX-era architectures. While CUDA support will continue for Maxwell, Pascal, and Volta GPUs, they will no longer receive new features in future updates. It's important to note that this change does not immediately affect GeForce gaming driver support, as Maxwell and Pascal GPUs are still included in the support list for the GeForce RTX series driver.

Nvidia has not provided a specific date for the end of full support for these three GPU architectures, but the transition is expected to occur soon. Once this change takes effect, the GTX 16-series, based on the Turing architecture of the RTX 20-series, will be the only remaining GTX-series GPUs with full support.

The Maxwell architecture, introduced 11 years ago, represents the oldest of the outgoing GPU architectures still supported by Nvidia on the consumer side. It debuted with the GeForce GTX 750 series and was followed by the GTX 900 series. Maxwell brought significant performance-per-watt improvements over its predecessor, Kepler, and was particularly notable for its efficiency in mobile GPUs.

Pascal, introduced in 2016 with the GeForce GTX 1000 series, marked one of Nvidia's most significant architectural advancements in the 2010s. It utilized TSMC's 16nm finFET plus technology, doubling the density of Maxwell's 28nm node and delivering substantial performance gains. The GTX 1080, for instance, offered 60-65 percent higher performance than its predecessor, the GTX 980.

Volta, released in 2017, was primarily focused on AI applications and enterprise use. It introduced Tensor cores, specialized units designed for AI workloads, which provided nine times the performance of Pascal in AI-specific tasks. Volta was largely confined to the enterprise sector, with the Titan V being the only desktop GPU to feature this architecture.

For Linux users, most distributions will continue to support legacy versions of the Nvidia driver, ensuring that affected cards will remain functional for the foreseeable future. However, users should be aware that they will not receive new features or optimizations moving forward.

Permalink to story:

 
Why not make the drivers open source so the community can keep them updated? Not a perfect solution, but better than legacy support
 
Because that doesn't make Nvidia more moneys. Buying a new card does

Τhe US Military calls that QUOTE Planned Obsolescence UNQUOTE

Sound strategy for NVIDIA and one sure to increase profits. Good for us stockholders.
 
I mean, these cards are ancient and aren't getting any actual code level improvements nowadays anyhow.

Yeah, you can download a new driver for them, but all that is really doing is updating the driver version number, there aren't any real changes under the hood.

Only card here that is being cast under bus really is Poor Volta, there is no reason it can't get DLSS updates etc given it has Tensor cores etc.
 
I mean, these cards are ancient and aren't getting any actual code level improvements nowadays anyhow.

Yeah, you can download a new driver for them, but all that is really doing is updating the driver version number, there aren't any real changes under the hood.

Only card here that is being cast under bus really is Poor Volta, there is no reason it can't get DLSS updates etc given it has Tensor cores etc.

Wrong, those drivers still patch security holes.

Seems to me you're doin' it wrong, man.
 
Why not make the drivers open source so the community can keep them updated? Not a perfect solution, but better than legacy support
Positives for NVIDIA:
* Generate a little bit of goodwill with old customers and the opensource community

Negatives for NVIDIA:
* Encourages people to use an old product instead of buying a new one
* They'll have to make sure nothing that can reflect negatively on them is found in the code*
* They'll have to make sure they don't publish anything that's copyright protected (video acceleration is an issue iirc when it comes to that)
* If they did something really smart somewhere AMD/Intel could copy their homework

Financially it's basically all negatives.

* A dumb example of this could be something like the word `master` to signify that part of code overwrites the behaviour of other parts or sets the standard behavior. In 2025 sensitive snowflakes want to shun the word entirely because somehow it helps us get away from slavery or something? (I have trouble following the logic)
GIT(hub) stopped with the 'master' branch terminology. It's now 'main' branch.
 
Why not make the drivers open source so the community can keep them updated? Not a perfect solution, but better than legacy support
The R515+ drivers now have an dual-license MPL/GPLv2 'open source' driver avaiable, but it only works with Turing and later (Turing, Ampere, Ada). This is because the hardware moved in some proprietary (e.g., 3rd party, possibly where nVidia does not own the IP) capability into hardware.

nVidia is unable to support its 'open' driver with earlier architectures for this reason. It cannot open source some of the IP.
 
Back