Nvidia reigns supreme in professional GPU applications, largely due to its CUDA platform, the backbone for countless tools from Blender to AI software. However, a new tool called Scale, developed by Spectral Compute, seeks to dismantle this dominance. While competitors to CUDA have emerged, such as AMD’s ROCm, Scale stands out as a ‘drop-in replacement.’ It functions as a compiler, allowing CUDA applications to be compiled directly for AMD GPUs, without the need for porting or language changes. Spectral Compute emphasizes that Scale seamlessly integrates with existing build tools and scripts. Scale’s strength lies in its compiler nature, setting it apart from translation tools like ZLUDA, which also aim to run CUDA applications on ROCm. While ZLUDA, funded by AMD, has been a valuable resource, it has its limitations and has reportedly lost AMD’s backing. Scale, on the other hand, is not open source and not funded by AMD. It’s offered under a free license, comprising a Nvidia CUDA Compiler (NVCC)-compatible compiler, a CUDA runtime implementation, and an open-source wrapper for CUDA-X calls. The free version allows commercial and private use but explicitly absolves Spectral Compute of any liability. This disclaimer might prove crucial, as Nvidia’s license agreement explicitly prohibits reverse-engineering CUDA for non-Nvidia platforms. Currently, Scale supports AMD’s RDNA 2 and RDNA 3 GPUs without any issues. It has also undergone basic testing on RDNA 1 GPUs, with support for older architectures in development. While the focus is on AMD, Intel GPUs could potentially gain support in the future. Although CUDA applications are heavily associated with Nvidia GPUs, breaking down these barriers paves the way for a more open software ecosystem. However, Scale’s long-term viability remains uncertain. While Spectral Compute claims that Scale doesn’t depend on CUDA, Nvidia could potentially take steps to remove or hinder its functionality. This potential clash underscores the ongoing battle for dominance in the GPU market, where innovation and strategic moves are crucial for maintaining a competitive edge.
New Tool Aims to Break Nvidia’s CUDA Dominance with AMD GPU Support
