What is CUDA?
CUDA is NVIDIA's programming platform for running general-purpose computations on NVIDIA GPUs. Much of the AI software ecosystem, including many training, fine-tuning and inference tools, was built CUDA-first. Some of these tools run only on NVIDIA hardware, or work best there. That matters when you choose a local LLM machine. The NVIDIA DGX Spark and Acer Veriton GN100 both use the NVIDIA GB10 chip, with 128 GB of unified memory and 273 GB/s of memory bandwidth, so they can run CUDA software. Apple machines such as the Mac Studio (M4 Max) and AMD-based systems such as the GMKtec EVO-X2 cannot run CUDA. They rely on other back ends, such as Apple's Metal, instead. For plain chat-style inference, this is often fine. If you plan to fine-tune models or follow tutorials that assume CUDA, an NVIDIA-based machine avoids compatibility problems.