Introducing SCALE, a toolkit that allows CUDA code to be compiled for AMD GPUs without any modifications

British startup Spectral Compute has announced that it has developed a toolkit called SCALE that compiles CUDA programs to run on AMD GPUs without any modifications.

SCALE GPGPU Programming Language

Announcing the SCALE BETA
https://scale-lang.com/posts/2024-07-12-release-announcement

Spectral Compute CEO Michael Sondergaard explained the background to the development of SCALE, saying, 'You write code once and it should be possible to build and run it on any hardware platform.' 'This has been possible for many years with CPUs, so why can't we do it with GPUs?'

SCALE works similarly to NVIDIA's CUDA toolkit, but with the ability to generate binaries for non-NVIDIA GPUs when compiling CUDA code. SCALE can be used to close the compatibility gap between generic CUDA programs and other hardware vendors, allowing one code to be used across a variety of hardware platforms.

Some CUDA code embeds PTX, which is intermediate code during compilation, inline, or expects the Nvidia CUDA compiler to operate independently, but SCALE aims to achieve source compatibility with CUDA, including these points.

Although SCALE is not open source, it can be used for free by following certain conditions such as copyright notation . At the time of writing, only the beta version is available.

The documentation includes instructions on how to install and use SCALE, and there are also examples demonstrating SCALE's functions , so be sure to check it out if you're interested.