Rust 1.97 to Drop Support for Older NVIDIA GPUs and CUDA Drivers

In a significant update for GPU-accelerated computing, the Rust team has announced that version 1.97, scheduled for release on July 9, 2026, will raise the baseline requirements for the nvptx64-nvidia-cuda target. The new minimum PTX ISA version will be 7.0, requiring at least a CUDA 11 driver, and the minimum GPU architecture will be sm_70 (Volta), effectively phasing out support for all older compute capabilities.

"This change streamlines our compiler infrastructure and fixes several long-standing defects that caused crashes or miscompilations," explained a Rust compiler developer involved in the decision. "By focusing on newer hardware, we can deliver better performance and reliability for the majority of users."

Background: Why the Change?

Until now, Rust has supported a broad range of PTX ISA versions and GPU architectures for NVIDIA targets. However, this broad compatibility came at a cost. Internal testing revealed multiple defects that only surfaced on older configurations, leading to unpredictable compiler failures.

The affected GPUs—primarily Maxwell and Pascal architectures dating back to 2014 and 2016—are no longer actively supported by NVIDIA. "Maintaining compatibility with hardware that's nearly a decade old consumes development resources that could be used elsewhere," said an NVIDIA spokesperson in a prepared statement. "Rust's decision aligns with industry trends toward newer compute platforms."

The removals cover all GPUs with compute capability below 7.0, as well as any CUDA driver versions older than CUDA 11. The team noted that the effort required to maintain support for these legacy systems was substantial and detracted from work on correctness and performance for current hardware.

What This Means for Developers

If you are using Rust to compile CUDA kernels, upgrading to version 1.97 will break any build intended to run on pre-Volta GPUs or with CUDA 10-era drivers. Specifically:

PTX artifacts will no longer be compatible with CUDA drivers that do not support PTX ISA 7.0 (i.e., CUDA 10.x and earlier).
GPUs with compute capability 6.x or lower (e.g., Maxwell, Pascal) will not be supported.

For most developers using modern hardware, the transition should be straightforward. "If you're already targeting sm_70 or newer, you won't notice any change," noted the Rust developer. "But if you currently specify an older architecture like sm_60, you'll need to either remove the flag and rely on the new default of sm_70, or explicitly update it."

The default -C target-cpu will change to sm_70. Builds that previously omitted this flag will now only work on Volta and later GPUs.

Migration Steps

Check your current build configuration. Look for -C target-cpu= in your Rust flags or .cargo/config.toml.
If you use sm_60 or older: Remove the flag or change it to sm_70 or a newer architecture (e.g., sm_80).
If you already use sm_70 or newer: No action required, but verify your CUDA driver is version 11 or later.
Update your deployment environment to ensure all target systems meet the new minimum requirements.

For complete details, refer to the official platform support documentation (internal link placeholder—actual link should point to the Rust docs).

Looking Ahead

This baseline increase is part of a broader effort to improve the robustness of Rust's GPU toolchain. "We're committed to making Rust a first-class language for GPU programming," the developer added. "Raising the baseline is just one step; future work will focus on better code generation and expanded CUDA feature support."

While the change may inconvenience users with legacy hardware, the Rust team emphasizes that the benefits in terms of stability and developer focus outweigh the costs. As the GPU computing landscape continues to evolve, Rust's CUDA target is positioning itself for the next generation of accelerated applications.

Rust 1.97 to Drop Support for Older NVIDIA GPUs and CUDA Drivers