SCALE tool lets NVIDIA CUDA applications run on AMD Radeon GPUs without modifying code

As an Amazon Associate, we earn from qualifying purchases. TweakTown may also earn commissions from other affiliate partners at no extra cost to you.

At its most basic level, Compute Unified Architecture (CUDA) allows general-purpose processing and other tasks to run on NVIDIA GPUs with extensive language support. Since its inception, CUDA has been a game changer. It has been used to tap into the power offered by GPUs for advanced simulation, physics calculations, scientific modeling, crypto mining, and, yes, machine learning.

2

The latter, and the current AI boom, have been one of the main reasons NVIDIA GPUs have been the go-to for generative AI workloads - developers and researchers alike have been using the CUDA toolkit for years. Widely viewed as the industry standard, this has left GPUs from competitors like AMD looking elsewhere for its software.

That could change as a new startup called Spectral Compute has unveiled SCALE, a toolkit that allows CUDA applications to be natively compiled for AMD's RDNA GPUs. According to its creators, "SCALE does not require the CUDA program or its build system to be modified."

"We believe that it should be possible to write code once and build/run it on any hardware platform," Spectral Compute's CEO, Michael Sondergaard, said. "This has been a reality for CPU code for many years, so why not GPUs? We set out to directly solve this problem by bridging the compatibility gap between the popular CUDA programming language and other hardware vendors."

SCALE works similarly to the NVIDIA CUDA Toolkit. It effectively "impersonates" the Toolkit, so tools and scripts "just work." It has also been developing for seven years. It doesn't rely on NVIDIA's code for its CUDA compatibility, so developers can work from a single codebase to compile an AMD GPU-ready version of an application. Not using NVIDIA code could be why SCALE could avoid being shut down by NVIDIA.

SCALE creator Spectral Compute has successfully tested it with applications like Blender, Llama-cpp, XGboost, FAISS, GOMC, and more - all running on AMD GPUs.