Review GPU / TPU

GPUs and TPUs are types of processors designed to perform many mathematical operations in parallel, making them useful for training and running many AI models.

A graphics processing unit, or GPU, originated in graphics workloads and became widely used for machine learning. A tensor processing unit, or TPU, is a specialized accelerator designed for tensor-heavy machine-learning computation.

Accelerator availability, memory, software support, energy use, cost, and supply chains influence which models can be built and where they can run.

A team trains a neural network across several GPUs, then chooses a smaller accelerator configuration for production inference.

A GPU or TPU is not an AI model and chip labels alone do not establish complete system performance.

Accelerators perform the parallel math; models and systems decide how it is used.

* [[compute|Compute]] * [[deep-learning|Deep Learning]] * [[inference|Inference]] * [[layered-sovereignty|Layered Sovereignty]] * [[concentration-risk|Concentration Risk]]

* [NVIDIA, *CUDA C++ Programming Guide*](https://docs.nvidia.com/cuda/cuda-c-programming-guide/) — vendor-primary documentation for GPU parallel-computing architecture and programming. * [Google Cloud, *Introduction to Cloud TPU*](https://docs.cloud.google.com/tpu/docs/intro-to-tpu) — vendor-primary documentation for tensor processing units and machine-learning workloads.

* Place accelerators within the broader control analysis in [[layered-sovereignty|Layered Sovereignty]].