Support for several architectures has been marked as feature-complete with the latest release of CUDA runtime, this month. The chipmaker stated in its CUDA 12.8, release notesthat
“Architecture support for Maxwell, Pascal, and Volta is considered feature-complete and will be frozen in an upcoming release,” .
This decision will likely impact datacenter operators, and scientific institutions that still rely on older models. The youngest architecture is almost eight years, while the oldest will celebrate its eleventh birthday this coming year. They grow up so quickly.
For those still using these older cards, including Nvidia’s 700, 900 and 1000-series desktop processors, as well as the M, P and V-series of datacenter parts, there’s good news: they’ll continue to work for a while longer.
Nvidia says that features deprecated by the latest release are still functional, at least temporarily. Documentation could be removed and they could become “officially unsupported,” for future releases.
We understand that users will eventually be stuck with older unsupported software, and may have compatibility issues when upgrading to future operating systems.
These changes will primarily affect those who use CUDA for computing-intensive workloads on GPUs. Graphics drivers for Nvidia Maxwell cards are still available and active support is provided. Maxwell is running out of time. Nvidia stopped supporting its predecessor, Kepler in 2021. Maxwell may not be far behind its older sibling.
- DeepSeek suspended new registrations due to cyberattack
- Google DeepMind’s CEO said 2025 is the year we’ll start taking pills AI invented
- What does Microsoft’s NPU-obsession leave Nvidia AI PC ambitions in?
- Nvidia invests $500M in Israeli boffinry supercomputer.
Ofcourse, just because hardware may not be supported moving forward does not mean users will give up the hardware before the magic smoke escapes. The Texas Advanced Computing Center’s Stallion tiled-display system still runs on a bunch 13-year-old Quadro K5000s.
There’s still a lot of people running Maxwell, Pascal and Volta hardware. Sierra, the supercomputer at Livermore National Laboratory, is still running. It’s powered by IBM Power9 CPUs with Nvidia V100 accelerators. Summit, a nearly identical supercomputer, was decommissioned only last fall.
Nvidia V100-series components were also used by OpenAI for training GPT 3.5 which powered ChatGPT, and kicked off AI boom.
CUDA 12.8 deprecates hardware and also removes support for older operating systems, including Windows 10 21H2, Debian 11 and SUSE Linux enterprise server (SLES) 15.4 SP4 and OpenSUSE 154. The Register contacted Nvidia
for comment. We’ll let you all know if there is any response. (r)
