The nvptx64-nvidia-cuda target is a compilation target for NVIDIA GPUs. When using this target, the final output is PTX. Two version choices shape that output:
- a GPU architecture (for example,
sm_70, sm_80, …), which determines which GPUs can run the PTX, and
- a PTX ISA version, which determines which CUDA driver versions can load (and JIT-compile) the PTX.
In Rust 1.97 (scheduled for release on July 9, 2026), the baseline PTX ISA version and GPU architecture for nvptx64-nvidia-cuda will be increased. These changes affect both the Rust compiler (rustc) and related host tooling, and they make it impossible to generate PTX artifacts compatible with older GPUs and older CUDA drivers.
The new minimum supported versions will be:
- PTX ISA 7.0 (requires a CUDA 11 driver or newer)
- SM 7.0 (GPUs with compute capability below 7.0 are no longer supported)
Why are the requirements being changed?
Until now, Rust has supported emitting PTX for a wide range of GPU architectures and PTX ISA versions. In practice, several defects existed that could cause valid Rust code to trigger compiler crashes or miscompilations. Raising the baseline addresses these issues and enables more complete support for the remaining supported hardware.
Removing support affects users of the architectures being removed. In this case, the most recent affected GPU architectures date back to 2017 and are no longer actively supported by NVIDIA. We therefore expect the overall impact of this change to be limited.
Maintaining support for these architectures would require substantial effort. These removals let us focus development efforts on improving correctness and performance for currently supported hardware.
What happens when I update to Rust 1.97?
If you need to target a CUDA driver that does not support PTX ISA 7.0 (CUDA 10-era drivers and older), Rust 1.97 will no longer be able to generate PTX compatible with that environment. Similarly, if you need to run on GPUs with compute capability below 7.0 (for example, Maxwell or Pascal), Rust 1.97 will no longer be able to generate compatible PTX for those GPUs.
Assuming you are targeting a CUDA driver compatible with CUDA 11 or newer and using GPUs with compute capability 7.0 or newer:
- If you do not specify
-C target-cpu, the new default will be sm_70, and your build should continue to work (but will no longer be compatible with pre-Volta GPUs).
- If you currently specify an older
-C target-cpu (for example, sm_60), you will need to either:
- remove that flag and let it default to
sm_70, or
- update it to
sm_70 or a newer architecture.
- If you already specify
-C target-cpu=sm_70 (or newer), there should be no behavioral changes from this update.
For more details on building and configuring nvptx64-nvidia-cuda, see the platform support documentation.
Instagram Engineering
Shopify Engineering