Parts/Accelerators/Google TPU v5p
Accelerators · TPU v5 · 2023
Google TPU v5p
Systolic-array training accelerator deployed in 8,960-chip pods over a custom optical circuit-switched fabric. Never submitted to a public HPL list.
Why this part
Never sold. A TPU exists only as rented capacity inside Google Cloud, which means the world's third or fourth largest pool of AI compute is invisible to every analysis that counts units shipped. The architecture is a systolic array rather than a general-purpose parallel processor: extremely efficient at dense matrix multiplication and unsuited to anything else, which is a defensible trade when you own both the hardware and the models that run on it.
Systems using this part
- Systems
- 5
- Rmax underneath
- 0.0 GFlop/s
- Units recorded
- 181,560
| System | Role | Country | Units | System Rmax | Supplied by | Rank |
|---|---|---|---|---|---|---|
| Google Council Bluffs East TPU Campus 2024-11 | Accelerator | USA | 54,300 | - | Alphabet | - |
| Google Pryor North TPU Campus 2024-07 | Accelerator | USA | 23,500 | - | Alphabet | - |
| Google Columbus TPU Campus 2024-04 | Accelerator | USA | 28,200 | - | Alphabet | - |
| Google New Albany TPU Campus 2024-04 | Accelerator | USA | 66,600 | - | Alphabet | - |
| Google TPU v5p pod | Accelerator | USA | 8,960 | - | Alphabet | - |
“Units recorded” sums only the edges where a public source states a quantity. It is a floor, not a total, and should never be read as installed base.