Google TPU v5p
by Google · paid · Last verified 2026-03-17
Google's most powerful TPU for large-scale AI training. Features 95GB HBM2e memory per chip and is designed to train the largest frontier models via massive pod-scale configurations connected by Google's proprietary ICI interconnect.
https://cloud.google.com/tpu/docs/v5p ↗C+
C+—Average
Adoption: C+Quality: A+Freshness: A+Citations: B+Engagement: F
Specifications
- License
- Proprietary
- Pricing
- paid
- Capabilities
- ai-training, inference, bfloat16-compute, int8-compute, pod-scale
- Integrations
- jax, tensorflow, pytorch-xla, gcp
- Use Cases
- frontier-model-training, large-scale-pretraining, research
- API Available
- Yes
- Tags
- tpu, data-center, training, inference, google, cloud
- Added
- 2026-03-17
- Completeness
- 100%
Index Score
58.7Adoption
55
Quality
96
Freshness
90
Citations
70
Engagement
0
Put AI to work for your business
Deploy this hardware alongside autonomous AaaS agents that handle tasks end-to-end — no babysitting required.