Nvidia deprecates CUDA support for aging architectures

Maxwell, Pascal and Volta, oh my! But fear not, driver support is still safe

Updated The end of the road is nearing for a range of aging Nvidia graphics cards, as support for several architectures was marked as feature-complete in the latest release of its CUDA runtime this month.

"Architecture support for Maxwell, Pascal, and Volta is considered feature-complete and will be frozen in an upcoming release," the chipmaker said in its CUDA 12.8 release notes.

The decision is likely to impact datacenter operators and scientific institutions still relying on the older models. The youngest of these architectures is almost eight years old, while the eldest will celebrate its 11th birthday this year. They grow up so fast.

The good news for anyone still rocking one of these aging cards, which include Nvidia's later 700, 900, and 1000-series desktop chips as well as its M, P, and V-series datacenter parts, is that they won't stop working for a good while longer.

According to Nvidia, features deprecated in the latest release will still work, at least for the moment. However, documentation could be removed, and they may become "officially unsupported," in future releases.

As we understand it, eventually users will find themselves stuck running older unsupported software and may run into compatibility issues with future operating system releases.

The changes will primarily impact those using CUDA to run compute-heavy workloads on GPUs. Graphics drivers for Nvidia's Maxwell generation of cards still appear to be available and actively supported. Having said that, the clock is ticking for Maxwell. Nvidia ended support for its predecessor, Kepler, back in 2021. So, it may not be much longer before Maxwell joins its sibling in the legacy bin.

Of course, just because the hardware may not be supported going forward doesn't mean users will give them up before the magic smoke escapes. Last we heard, the Texas Advanced Computing Center's Stallion tiled display system is still running on a bunch of 13-year-old Quadro K5000s.

There are still plenty of folks running Maxwell, Pascal, and Volta hardware. Livermore National Laboratory's Sierra supercomputer, powered by IBM's Power9 CPUs and Nvidia's V100 accelerators, is still chugging along. Meanwhile, its nearly identical sibling, Summit was only decommissioned last fall.

Nvidia's V100-series parts were also famously used by OpenAI to train GPT 3.5, which powered its ChatGPT and kicked off the AI boom.

In addition to deprecating hardware, CUDA 12.8 also drops support for a number of older operating systems including Windows 10 21H2, Debian 11, and SUSE Linux Enterprise Server (SLES) 15 SP4 and OpenSUSE 15.4.

The Register reached out to Nvidia for comment; we'll let you know if we hear anything back. ®

Updated to add

Nvidia has confirmed to us that users or institutions running applications on Maxwell, Pascal, and Volta cards will be able to continue using them with its long term support driver for GPU-accelerated applications.

However, as we understand it, code bases based on future releases of CUDA may not run on these architectures.

More about

TIP US OFF

Send us news


Other stories you might like