We haven’t yet released support for tcgen05, but we’ll deal with it the same way we deal with other inline PTX: parsing it and converting it to target-appropriate instructions together with the rest of the program.
This is something we’ve done already for the hopper-class tensorcore instructions, and the blackwell ones will map similarly, though likely with a kernel launch involved.
In some benchmarks, SCALE beats nvcc, and we have compiler optimizations in the pipeline that will improve those numbers over time.
> If all you want is to be able to easily use non-NVidia hardware then high level tools like PyTorch already let you do that
Somewhat true, but, CUDA is significantly larger than PyTorch and there's more to Accelerated Computing than just those types of applications supported there.
> OTOH if you want to be programming close to the metal to achieve top performance then you are probably not using CUDA in the first place, and using some CUDA translation layer on non-NVidia hardware would be an even worse idea.
SOTA mlperf submissions use CUDA to achieve their high levels of performance.
It's not a "translation layer", it's a native, ahead-of-time compiler that makes full use of the native hardware features. Here's an example of a feature (Shuffles) being compiled to take advantage of native hardware instructions, resulting in speedups: https://scale-lang.com/posts/2026-01-19-optimizing-cuda-shuf...
A guess would be some time next year — since our public launch our focus has generally been on API coverage and increasingly recently, on performance.
While performance improvements will always remain a target, we're soon at full coverage of the core CUDA APIs and will be shifting an increasing amount of effort towards developer tooling.
Would love to connect and hear more about what you like about SCALE and where you'd like it to go.
AMD is a part of our strategy, but it's not the end-game - we envision SCALE to be vendor neutral and have plans to support all of the competitive GPUs that come out in the future, including AMD, NVIDIA, Intel and any newcomers.
We do believe in open source software and we do want to move the GPGPU market away from fully closed languages. The future is open for discussion but regardless, the status-quo at the moment is a proprietary and dominant implementation which only supports a single vendor.
> I don't see a way for a new language to catch on nowadays that is not open source.
I do note that CUDA is itself closed source -- while there's an open source implementation in the LLVM project, it is not as bleeding edge as NVIDIA's own.
We're going to be publishing more details on later blog posts and documentation about how this works and how we've built it.
Yes, we're not open source, however our license is very permissive. It's both in the software distribution and viewable online at https://docs.scale-lang.com/licensing/
Ping me at [email protected]