NVIDIA CUDA Can Now Directly Run On AMD GPUs Using The SCALE Toolkit
British startup Spectral Compute has unveiled SCALE, a GPGPU toolchain that enables NVIDIA’s CUDA to seamlessly function on AMD’s GPUs. The industry has been striving to break NVIDIA’s software dominance, seeking ways to eliminate the exclusivity status through various approaches. A recent development was the emergence of ZLUDA, an open-source project allowing CUDA libraries to interface with AMD’s ROCm, supporting Team Red’s GPUs. Now, a new contender has entered the scene with SCALE, enabling AMD users to harness NVIDIA’s CUDA capabilities, creating a powerful hybrid model.
Spectral Compute CEO Michael Sondergaard advocates for an open-source environment for GPUs akin to modern CPUs, promoting interoperability across platforms. He views SCALE as a solution bridging the compatibility gap between CUDA and other hardware vendors, unlocking market limitations. SCALE, developed over seven years, does not rely on NVIDIA’s code but constructs its CUDA-compatible toolchain, making it highly versatile across platforms like AMD’s RDNA GPUs. It streamlines developers’ work by maintaining a single version of the codebase, eliminating the need for translation layers or porting.
While the implementation of SCALE suggests a shift from NVIDIA’s CUDA exclusivity to broader accessibility, the toolkit itself is not open-source, accessible through a free software license. Spectral Compute has extensively tested SCALE on various applications using AMD’s RDNA 3 and RDNA 2 architectures, including Blender, Llama-cpp, XGboost, FAISS, and NVIDIA Thrust, among others. NVIDIA has previously expressed concerns about resources allowing CUDA on external components, as evidenced by warnings in their EULA. CUDA’s pivotal role in NVIDIA’s AI market dominance makes retaining its exclusivity a priority for the company.
