Nvidia dominates the professional GPU market not only because of their superior graphics cards, but also because of their CUDA platform. This platform serves as the foundation for popular applications like Blender and AI software. A new tool called Scale, developed by Spectral Compute, aims to break the barriers of this closed ecosystem.
Unlike other alternatives like AMD ROCm, Scale is designed as a seamless replacement for CUDA. It acts as a compiler that allows CUDA applications to be directly compiled for AMD GPUs without the need for code translation. According to Spectral Compute, existing CUDA programs can be used as is with Scale, maintaining compatibility with current build tools and scripts.
Scale stands out because it functions as a compiler, not just a translation tool. While other tools like ZLUDA exist to run CUDA applications on AMD GPUs, they are not as effective. Scale, though not open source and not funded by AMD, consists of a CUDA-compatible compiler, a CUDA runtime implementation, and a wrapper for CUDA-X calls. The free version is available for both commercial and private use, but Spectral Compute absolves itself of any liability.
Spectral Compute confirms that Scale works well with AMD’s RDNA 2 and RDNA 3 GPUs, with plans to support older architectures in the future. While currently focused on AMD, support for Intel GPUs may be possible down the line. Opening up support beyond Nvidia GPUs signifies a move towards a more inclusive software environment, though Nvidia’s response to Scale remains uncertain.
It’s worth noting that while most CUDA users likely have Nvidia GPUs, Scale’s development signifies a step towards breaking down restrictive software barriers. The future longevity of Scale is uncertain, as Nvidia may seek to counter its compatibility with non-Nvidia platforms.
Get your weekly breakdown of PC gaming tech
