By using learntpu.dev you agree to these terms. If you do not agree, please do not use the site.
Educational content, provided as is
The course is provided for educational purposes, free of charge, as is and without warranty of any kind, express or implied, including any warranty of accuracy, completeness, fitness for a particular purpose or non-infringement.
The content was checked against the cited sources, but:
- hardware specifications, software APIs and prices change, and some figures were not yet published or confirmed when written;
- figures tagged "provisional", "derived" or "unverified" are estimates and labelled as such;
- performance numbers are idealised lower bounds or model predictions, not measurements of your workload.
Do not rely on the course as the sole basis for purchasing, capacity-planning, engineering or business decisions. Confirm figures with the vendor's current documentation. To the extent permitted by law, the maintainer is not liable for any loss or damage arising from use of the site or its content.
Independence and trademarks
learntpu.dev is an independent project. It is not affiliated with, sponsored by or endorsed by Google LLC, NVIDIA Corporation, Amazon Web Services, Meta Platforms or any other company mentioned.
Google, Google Cloud, Cloud TPU, TPU, Ironwood, Trillium, JAX, XLA and related names are trademarks of Google LLC. NVIDIA, H100, B200, GB200, Blackwell, Rubin, NVLink, CUDA and NCCL are trademarks of NVIDIA Corporation. AWS, Trainium and Neuron are trademarks of Amazon.com, Inc. or its affiliates. PyTorch is a trademark of the Linux Foundation. Other names are the property of their owners. They are used here only to identify and describe the products discussed.
Third-party material and links
The course quotes and summarises published work, chiefly How to Scale Your Model (Austin et al., Google DeepMind) and vendor documentation, with attribution, for commentary and teaching. Links to external sites are provided for reference; their content and terms are their owners' responsibility.
Reuse of this course
The course content (text, quizzes, flashcards and slides) is licensed under Creative Commons Attribution 4.0 International (CC BY 4.0): you may share and adapt it, including commercially, if you give credit to "TPU Deep Dive (learntpu.dev)", link to the license and say whether you made changes. The site's code, including the interactives and scripts, is under the MIT License. Quoted or adapted third-party material remains under its original authors' terms. See the LICENSE file for details.
AI assistance and errors
The course was written with the help of AI (Anthropic's Claude) and reviewed by the maintainer. It may contain errors. See About for how it was checked and how to report a mistake.
Acceptable use
Do not attempt to disrupt the site or its hosting, and do not use automated access in ways that place unreasonable load on it. Ordinary search-engine crawling is welcome.
Changes and contact
These terms may be updated; the date at the top of the page reflects the latest version. Questions can be raised by opening an issue on GitHub.