TRC free Cloud TPU quota for accepted researchers high confidence
Accepted researchers get no-cost Cloud TPU quota in their own Google Cloud project, usable within minutes of accepting an invitation. The About page describes access to more than 1,000 Cloud TPU devices (also described as thousands of Cloud TPU chips). Per-participant TPU type, device count, and access length are not stated; access is granted on a temporary basis. No dollar credits or cash amounts are offered. Supported frameworks: TensorFlow, PyTorch, Julia, JAX. Only TPU use is free; other Google Cloud services are billed (see gotchas).
- Eligibility
- Researchers accepted into the TRC program. Anyone may express interest, but invitations go only to approved applicants, and the approval criteria are not published. The official pages name no student-only, startup-only, or country restriction, and they do not name students or startups as a target group. Students are not excluded by the text, but no student track exists. Startups are not mentioned at all, so a startup has no stated path in.
- How to claim
- 1) Submit the TRC interest form, reached from the Express interest button on https://sites.research.google/trc/about/ (direct link https://forms.gle/uwWjUewyxZ5okL3W7). 2) Wait for an invitation. Invitations are sent on a rolling basis to approved applicants; no review timeline is published. 3) Accept the invitation and agree to Google's Terms and Conditions and Privacy Policy. 4) The Cloud TPU quota appears in your Google Cloud project. The quota is not compatible with CMLE, CAIP, or Vertex-based workflows; the FAQ recommends GCE tutorials instead.
- Expiry
- No end date or renewal terms are published. Access is only described as 'on a temporary basis'. The pages carry no dates.
- Gotchas
- (1) Approval is discretionary and undisclosed; expressing interest does not guarantee an invitation. (2) Required in return: share TRC-supported research publicly (peer-reviewed papers, open-source code, blog posts, or similar), give Google detailed feedback, and keep research consistent with Google AI Principles. A suggested acknowledgment line is 'Research supported with Cloud TPUs from Google's TPU Research Cloud (TRC)'. (3) Non-TPU GCP services are not free: the FAQ names small VMs (n1-standard-2) and Cloud Storage buckets for training data, described as 'generally minimal'. Networking and egress are not addressed. Billing questions go to trc-support@google.com. (4) Access is temporary with no stated length, so do not plan on it as permanent compute.
- Source
- sites.research.google