NVIDIA H100 SXM vs Google TPU v4 (历代旗舰): Spec Comparison & Buyer's Guide
In AI infrastructure selection, NVIDIA H100 SXM and Google TPU v4 (历代旗舰) are two accelerators frequently compared. This article contrasts them item by item — architecture, compute, memory, power, and release cadence — to help you quickly judge which fits training or inference workloads.
Spec Comparison Table
| Vendor | NVIDIA H100 SXM | Google TPU v4 (历代旗舰) |
|---|---|---|
| Vendor | NVIDIA | |
| Architecture | Hopper GH100 | TPU v4 |
| Process | TSMC 4N | TSMC 7nm |
| Release Date | 2022 3 GTC | 2020-07-15 |
| FP8 Compute | 3,958 TFLOPS | — |
| FP16 Compute | — | — |
| FP32 Compute | 67 TFLOPS | — |
| INT8 Compute | — | 275 TOPS |
| Memory Type | — | — |
| Memory Capacity | 80 GB HBM3 | 32 GB HBM2 |
| Memory Bandwidth | 3.35 TB/s | — |
| TDP Power | 700 W | 170 W |
Key Differences
- Power: Google TPU v4 (历代旗舰) has a TDP of 170 W, lower than NVIDIA H100 SXM's 700 W, friendlier to datacenter PUE and cooling.
- Memory capacity: NVIDIA H100 SXM packs 80 GB HBM3, more than Google TPU v4 (历代旗舰)'s 32 GB HBM2, more comfortable for single-card hosting of very large models.
Selection Advice
- When chasing extreme single-card compute and a mature toolchain, prioritize NVIDIA H100 SXM; if budget, power wall, or local support are hard constraints, Google TPU v4 (历代旗舰) often fits better. Use this site's AI Compute Card Comparison Tool to validate multiple chips side-by-side before deciding.
FAQ
What are the main differences between NVIDIA H100 SXM and Google TPU v4 (历代旗舰)?
The core difference is architecture and compute density: NVIDIA H100 SXM uses Hopper GH100, FP8 ~3,958 TFLOPS, memory 80 GB HBM3; Google TPU v4 (历代旗舰) uses TPU v4, FP8 ~No public FP8 data, memory 32 GB HBM2. See the comparison table above.
What is the TDP (power) of NVIDIA H100 SXM?
NVIDIA H100 SXM has a TDP of 700 W; actual whole-system power also includes board, fans, and PUE.
Which is better for large-model training / inference?
Training values memory capacity, bandwidth, and multi-card interconnect; inference values single-card throughput and power efficiency. Combine the "Key Differences" and "Selection Advice" above with your batch size, model size, and SLA.
How much do NVIDIA H100 SXM and Google TPU v4 (历代旗舰) differ in memory capacity?
NVIDIA H100 SXM is 80 GB HBM3, Google TPU v4 (历代旗舰) is 32 GB HBM2; the gap directly affects loadable model size and context length.
Related Pages
- NVIDIA H100 SXM
- Google TPU v4 (历代旗舰)
- AI 算力卡对比工具 — Compare 2–4 chips side-by-side online