Parts/Accelerators/Google TPU v4
Accelerators · TPU v4 · 2020
Google TPU v4
4,096-chip supercomputer configuration joined by optical circuit switches, with SparseCores for embedding workloads.
Why this part
Google says TPU v4 has been deployed since 2020 and was offered to customers in 2022. Its documentation gives 275 TFLOPS (bf16 or int8) and 32 GiB of HBM per chip; those figures are Google's own peak numbers. Like every TPU it is capacity Google rents rather than a part anyone can buy.
Systems using this part
- Systems
- 2
- Rmax underneath
- 0.0 GFlop/s
- Units recorded
- 36,864
2 of 2 edges disclose a quantity
| System | Role | Country | Units | System Rmax | Supplied by | Rank |
|---|---|---|---|---|---|---|
| Google TPU v4 ML Hub (Oklahoma) 2022-05 | Accelerator | USA | 32,768 | - | Alphabet | - |
| Google TPU v4 supercomputer 2020 | Accelerator | USA | 4,096 | - | 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.