B200 vs H100
Spec-sheet and pricing comparison of NVIDIA B200 and NVIDIA H100 SXM with links to continuously measured LLM inference benchmarks on identical workloads.
Spec-sheet comparison
| B200 | H100 | B200 / H100 | |
|---|---|---|---|
| Memory per chip | 180 GB HBM3e | 80 GB HBM3 | 2.25x |
| Memory bandwidth | 8 TB/s | 3.35 TB/s | 2.39x |
| Dense FP8 compute | 4,500 TFLOP/s | 1,979 TFLOP/s | 2.27x |
| Dense FP4 compute | 9,000 TFLOP/s | Not supported | n/a |
| TDP | 1,000 W | 700 W | 1.43x |
| Hourly rate (neocloud tier) | $2.07/hr | $1.55/hr | 1.34x |
| Scale-up world size | 8 chips | 8 chips | 1.0x |
Ratios are spec-sheet values; see the live compare pages for measured deltas.
Frequently asked questions
- Which has more memory, B200 or H100?
- B200 offers 180 GB HBM3e per chip versus 80 GB HBM3 on H100 (2.25x the capacity).
- How do B200 and H100 prices compare?
- At the neocloud tier the SemiAnalysis TCO model rates B200 at $2.07/hr versus $1.55/hr for H100. Hourly price alone is misleading; the per-dollar compare pages divide measured throughput by these rates.
- Is B200 faster than H100 for LLM inference?
- On paper B200 has 2.27x the dense FP8 compute of H100, but delivered tokens per second depend on the model, framework, precision and interactivity target. InferenceX measures both chips daily on identical workloads; see the live compare pages for current results.
See live benchmark results
Every number above is static hardware data. Delivered tokens per second, cost per million tokens and energy per token are measured continuously on the dashboard:
Go deeper with the SemiAnalysis models
InferenceX measures delivered inference performance. The SemiAnalysis institutional models cover the market behind these chips: who ships them, who buys them, and what they cost to own.
SemiAnalysis Accelerator & HBM Model
SKU-level AI accelerator shipments, pricing and specifications, from foundry wafer starts and HBM supply through customer-level installed base, quarterly with multi-year forecasts.
SemiAnalysis AI Cloud TCO Model
The source of the hourly rates on this page: all-in GPU cost of ownership built up from server capex, power, colocation and cost of capital, with rental price scenarios and a full cluster finance suite.