Chip comparison
Cerebras WSE-3 (Wafer-Scale Engine 3) vs Google TPU v4
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 | Cerebras WSE-3 (Wafer-Scale Engine 3) | Google TPU v4 |
|---|---|---|
| Process node | TSMC N5 | TSMC N7 |
| Transistors (B) | 4000 | — |
| Die size | 46225 mm² | — |
| TDP | 23000 W | 192 W |
| Memory | 44 GB on-die SRAM | 32 GB HBM2 |
| Memory bandwidth | 21,000,000 GB/s | 1,200 GB/s |
| FP16 (dense) | 62,500 TFLOPS | — |
| BF16 (dense) | — | 275 TFLOPS |
| INT8 (dense) | — | 275 TOPS |
| Form factor | wafer-scale (single-wafer system) | OAM (per-chip in v4 pod) |
| Announced | 2024-03-13 | 2021-05-18 |
| Released | 2024-04-01 | 2022-05-01 |
Robots running Cerebras WSE-3 (Wafer-Scale Engine 3)
No robots publicly running it yet.
Robots running Google TPU v4
No robots publicly running it yet.
Data centers with Cerebras WSE-3 (Wafer-Scale Engine 3)
No data centers publicly running it yet.
Data centers with Google TPU v4
Common questions
Cerebras WSE-3 (Wafer-Scale Engine 3) vs Google TPU v4: which is faster for training?
Cerebras WSE-3 (Wafer-Scale Engine 3) has 227.27x the dense FP16/BF16 throughput of the other (Cerebras WSE-3 (Wafer-Scale Engine 3): 62,500 TFLOPS; Google TPU v4: 275 TFLOPS). Real-world training speed also depends on memory bandwidth, interconnect, and software stack: see the spec diff for those inputs.
Cerebras WSE-3 (Wafer-Scale Engine 3) vs Google TPU v4: which has more memory?
Cerebras WSE-3 (Wafer-Scale Engine 3) carries more HBM (Cerebras WSE-3 (Wafer-Scale Engine 3): 44 GB on-die SRAM; Google TPU v4: 32 GB HBM2). Memory capacity sets the largest model that fits on one chip without partitioning; memory bandwidth sets inference throughput.
Cerebras WSE-3 (Wafer-Scale Engine 3) vs Google TPU v4: which is more power-efficient?
Dense FP16/BF16 TFLOPS per watt: Cerebras WSE-3 (Wafer-Scale Engine 3) 2.72 (62,500 ÷ 23000 W); Google TPU v4 1.43 (275 ÷ 192 W). Cerebras WSE-3 (Wafer-Scale Engine 3) wins at the die level; system-level efficiency also depends on cooling and interconnect.
Cerebras WSE-3 (Wafer-Scale Engine 3) vs Google TPU v4: which is more widely used?
Cerebras WSE-3 (Wafer-Scale Engine 3): 0 robots and 0 data centers publicly running it. Google TPU v4: 0 robots and 2 data centers. Empty here means no operator has said publicly, not zero adoption in the market.
See also: every chip comparison · Cerebras WSE-3 (Wafer-Scale Engine 3) full page · Google TPU v4 full page.