Chip comparison
Google TPU v4 vs Google TPU v6 (Trillium)
Side-by-side specs
Straight from each vendor's datasheet. Winner in bold: higher throughput, memory, or transistor count; lower power draw or price. Sparse (2:4) throughput roughly doubles the FP8/INT8 numbers shown.
| Field | Google TPU v4 | Google TPU v6 (Trillium) |
|---|---|---|
| Process node | TSMC N7 | — |
| TDP | 192 W | 200 W |
| Memory | 32 GB HBM2 | 32 GB HBM3 |
| Memory bandwidth | 1,200 GB/s | 1,600 GB/s |
| BF16 (dense) | 275 TFLOPS | 918 TFLOPS |
| FP8 (dense) | — | 1,836 TFLOPS |
| INT8 (dense) | 275 TOPS | 1,836 TOPS |
| Form factor | OAM (per-chip in v4 pod) | OAM (Trillium pod, up to 256 chips) |
| Announced | 2021-05-18 | 2024-05-14 |
| Released | 2022-05-01 | 2024-12-01 |
Robots running Google TPU v4
No robots publicly running it yet.
Robots running Google TPU v6 (Trillium)
No robots publicly running it yet.
Data centers with Google TPU v4
Data centers with Google TPU v6 (Trillium)
Common questions
Google TPU v4 vs Google TPU v6 (Trillium): which is faster for training?
Google TPU v6 (Trillium) has 6.68x the dense FP16/BF16 throughput of the other (Google TPU v4: 275 TFLOPS; Google TPU v6 (Trillium): 1,836 TFLOPS). Real-world training speed also depends on memory bandwidth, interconnect, and software stack: see the spec diff for those inputs.
Google TPU v4 vs Google TPU v6 (Trillium): which has more memory?
Google TPU v6 (Trillium) carries more HBM (Google TPU v4: 32 GB HBM2; Google TPU v6 (Trillium): 32 GB HBM3). Memory capacity sets the largest model that fits on one chip without partitioning; memory bandwidth sets inference throughput.
Google TPU v4 vs Google TPU v6 (Trillium): which is more power-efficient?
Dense FP16/BF16 TFLOPS per watt: Google TPU v4 1.43 (275 ÷ 192 W); Google TPU v6 (Trillium) 9.18 (1,836 ÷ 200 W). Google TPU v6 (Trillium) wins at the die level; system-level efficiency also depends on cooling and interconnect.
Google TPU v4 vs Google TPU v6 (Trillium): which is more widely used?
Google TPU v4: 0 robots and 2 data centers publicly running it. Google TPU v6 (Trillium): 0 robots and 1 data center. Empty here means no operator has said publicly, not zero adoption in the market.
See also: every chip comparison · Google TPU v4 full page · Google TPU v6 (Trillium) full page.