Spectral Compute has raised $6 million in seed funding. The company aims to make AI software compatible with a wider range of hardware.
Most modern AI applications are built on Nvidia's proprietary CUDA programming model. This locks developers into using Nvidia's expensive and supply-constrained GPUs. This dependency creates what some call the "Nvidia tax." Spectral Compute is building an AI CUDA compiler named SCALE. It allows programs written for Nvidia hardware to run on other chips without performance loss. SCALE directly compiles CUDA code for alternative GPUs in seconds. This process avoids the need for developers to rewrite their software from scratch. The toolkit currently supports AMD's GPUs. The company plans to expand compatibility to all AI accelerators in the future. Its vision is to let developers write CUDA code once and run it anywhere.
