Graphics Processing Unit - specialized hardware designed for parallel processing that accelerates AI training and graphics rendering. Like having thousands of workers doing calculations simultaneously instead of one at a time.
Training large AI models on GPUs can be 100x faster than using regular CPUs because GPUs can process many calculations at once.
All major clouds provide GPUs primarily through virtual machines: you choose a GPU-capable VM/shape (or attach a GPU) and run CUDA/ROCm-enabled workloads such as AI training, inference, rendering, and HPC. Naming differs (instance types/VM series/shapes), but the concept is the same: pay for GPU time plus supporting CPU, memory, and storage.
Explore real-world architectures from our community that use GPU: