B200 SXM vs H200 SXM
Side-by-side comparison of the NVIDIA B200 SXM and the NVIDIA H200 SXM for AI inference workloads.
Specifications
| Spec | B200 SXM | H200 SXM |
|---|---|---|
| Generation | blackwell | hopper |
| Memory Type | HBM3e | HBM3e |
| VRAM | 180 GB | 141 GB |
| Memory Bandwidth | 8,000 GB/s | 4,800 GB/s |
| BF16 TFLOPS | 2,250 | 990 |
| FP16 TFLOPS | 2,250 | 990 |
| FP8 TFLOPS | 4,500 | 1,979 |
| INT8 TOPS | 4,500 | 1,979 |
| TDP | 1,000 W | 700 W |
| Interconnect | nvlink | nvlink |
| NVLink Bandwidth | 1,800 GB/s | 900 GB/s |
| Max GPUs per Node | 8 | 8 |
| PCIe Gen | Gen 6 | Gen 5 |
| CUDA Compute Capability | 10 | 9 |
Pricing
B200 SXM
| Provider | On-Demand | Reserved | Spot |
|---|---|---|---|
| coreweave | $7.50/hr | $5.50/hr | - |
| lambda | $5.99/hr | $4.49/hr | - |
| runpod | $7.20/hr | - | - |
H200 SXM
| Provider | On-Demand | Reserved | Spot |
|---|---|---|---|
| lambda | $3.49/hr | $2.69/hr | - |
| coreweave | $4.25/hr | $3.19/hr | - |
| runpod | $4.69/hr | - | - |
| tensordock | $3.80/hr | - | $2.90/hr |
Cheapest available rate: B200 SXM at $4.49/hr vs H200 SXM at $2.69/hr — H200 SXM is +67% cheaper
Efficiency Metrics
TFLOPS / Watt
2.3
B200 SXM
1.4
H200 SXM
BF16
VRAM / Dollar
40.1
B200 SXM
52.4
H200 SXM
GB/$/hr
Bandwidth / Watt
8.0
B200 SXM
6.9
H200 SXM
GB/s/W
Models (FP16, 1 GPU)
220.0
B200 SXM
193.0
H200 SXM
Model Compatibility (FP16, Single GPU)
Only on B200 SXM (27)
- Code Llama 70B
- Dolphin 2.9 72B
- DeepSeek R1 Distill 70B
- Llama 3 70B 1M Context
- Llama 2 70B
- Llama 3 70B
- Llama 3.1 70B
- Llama 3.3 70B
- WizardMath 70B
- Hermes 3 70B
- HelpSteer2 Llama 3.1 70B
- Llama 3.1 Nemotron 70B Instruct
- Llama 3.1 Nemotron 70B Reward
- Nemotron 70B
- Qwen 2.5 72B
- Qwen 2.5 Math 72B
- Qwen 2.5 VL 72B
- Llama 3.1 70B Turbo
- Claude Sonnet 4
- o1-mini
- +7 more
Both (193)
- Yi 1.5 34B
- Yi 1.5 9B
- Yi Coder 9B
- Jamba 1.5 Mini
- GTE Qwen2 7B
- Marco O1
- Qwen 1.5 MoE A2.7B
- Qwen 2 Audio 7B
- Qwen 2.5 14B
- Qwen 2.5 32B
- Qwen 2.5 3B
- Qwen 2.5 Coder 32B
- OLMo 2 13B
- OLMo 2 7B
- Amazon Nova Lite
- Amazon Nova Pro
- OpenELM 3B
- BGE Large EN v1.5
- BGE M3
- Baichuan 2 13B
- +173 more
Only on H200 SXM (0)
None
Summary
The B200 SXM (blackwell generation) offers 180GB of HBM3e with 2,250 BF16 TFLOPS and 8,000 GB/s memory bandwidth at 1000W TDP.
The H200 SXM (hopper generation) offers 141GB of HBM3e with 990 BF16 TFLOPS and 4,800 GB/s memory bandwidth at 700W TDP.
The B200 SXM has +28% more VRAM, allowing it to run larger models without multi-GPU setups.
From a cost perspective, the H200 SXM is more affordable at $2.69/hr vs $4.49/hr for the B200 SXM.