Chip comparison
Google TPU v4 vs NVIDIA B200 (Blackwell)
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 | NVIDIA B200 (Blackwell) |
|---|---|---|
| Process node | TSMC N7 | TSMC 4NP |
| Transistors (B) | — | 208 |
| TDP | 192 W | 1000 W |
| Memory | 32 GB HBM2 | 192 GB HBM3e |
| Memory bandwidth | 1,200 GB/s | 8,000 GB/s |
| FP16 (dense) | — | 2,250 TFLOPS |
| BF16 (dense) | 275 TFLOPS | 2,250 TFLOPS |
| FP8 (dense) | — | 4,500 TFLOPS |
| INT8 (dense) | 275 TOPS | 4,500 TOPS |
| Form factor | OAM (per-chip in v4 pod) | SXM6 |
| Announced | 2021-05-18 | 2024-03-18 |
| Released | 2022-05-01 | 2024-11-01 |
Robots running Google TPU v4
No robots publicly running it yet.
Robots running NVIDIA B200 (Blackwell)
No robots publicly running it yet.
Data centers with Google TPU v4
Data centers with NVIDIA B200 (Blackwell)
Common questions
Google TPU v4 vs NVIDIA B200 (Blackwell): which is faster for training?
NVIDIA B200 (Blackwell) has 16.36x the dense FP16/BF16 throughput of the other (Google TPU v4: 275 TFLOPS; NVIDIA B200 (Blackwell): 4,500 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 NVIDIA B200 (Blackwell): which has more memory?
NVIDIA B200 (Blackwell) carries more HBM (Google TPU v4: 32 GB HBM2; NVIDIA B200 (Blackwell): 192 GB HBM3e). Memory capacity sets the largest model that fits on one chip without partitioning; memory bandwidth sets inference throughput.
Google TPU v4 vs NVIDIA B200 (Blackwell): which is more power-efficient?
Dense FP16/BF16 TFLOPS per watt: Google TPU v4 1.43 (275 ÷ 192 W); NVIDIA B200 (Blackwell) 4.50 (4,500 ÷ 1000 W). NVIDIA B200 (Blackwell) wins at the die level; system-level efficiency also depends on cooling and interconnect.
Google TPU v4 vs NVIDIA B200 (Blackwell): which is more widely used?
Google TPU v4: 0 robots and 2 data centers publicly running it. NVIDIA B200 (Blackwell): 0 robots and 3 data centers. Empty here means no operator has said publicly, not zero adoption in the market.
See also: every chip comparison · Google TPU v4 full page · NVIDIA B200 (Blackwell) full page.