Skip to content
Updated minutes ago
nvidia

B200 NVL (pair)

nvidia · blackwell · 360 GB HBM3e · 1800W TDP

VRAM

360 GB

BF16 TFLOPS

4500

Bandwidth

16000 GB/s

From

$10.50/hr

Calculate ROI with this GPU →

Spec Sheet

VRAM360 GB HBM3e
Memory Bandwidth16000 GB/s
BF16 TFLOPS4500
FP16 TFLOPS4500
FP8 TFLOPS9000
INT8 TOPS9000
TDP1800W
InterconnectNVLINK
NVLink Bandwidth1800 GB/s
Max per Node4
PCIe Gen6
CUDA Compute Capability10
Tensor CoresYes

Pricing by Provider

ProviderOn-DemandReservedSpotBadge
coreweave$14.00/hr$10.50/hr-Cheapest
lambda$11.49/hr--

Compatible Models (280)

Training Capabilities

Estimated GPU count for full fine-tuning (AdamW, BF16) and QLoRA

Model SizeFull Fine-TuneQLoRA
7B model1 GPU1 GPU
13B model1 GPU1 GPU
70B model4 GPUs1 GPU

Energy Efficiency

Estimated tokens/second per Watt for popular models

Mistral 7B
1.22 t/s/WFP8
Qwen 2.5 7B
1.17 t/s/WFP8
Llama 3.1 8B
1.11 t/s/WFP8
DeepSeek V3
0.24 t/s/WFP8
Llama 3.1 70B
0.13 t/s/WFP8
Qwen 2.5 72B
0.12 t/s/WFP8

Similar GPUs

GPUVRAMBF16 TFLOPSBW (GB/s)From
B300288 GB280012000$0.00/hr
B100 SXM192 GB17508000$4.50/hr
GB200 NVL72 (per GPU)192 GB22508000$6.50/hr
GB300 NVL72 (per GPU)192 GB25008000$7.50/hr
B200 SXM180 GB22508000$4.49/hr

Methodology Note

Performance estimates for the B200 NVL (pair)are based on InferenceBench's roofline performance model with CUDA kernel-level optimization including FlashAttention v2 and PagedAttention. Memory calculations account for model weights (360 GB HBM3e available), KV-cache allocation, and activation memory. Throughput predictions use the B200 NVL (pair)'s rated 16000 GB/s memory bandwidth and 4500 BF16 TFLOPS compute capacity as roofline ceilings, with empirical correction factors per GPU architecture (blackwell). See our full methodology.

Frequently Asked Questions

How many AI models can run on B200 NVL (pair)?

The B200 NVL (pair) can run 280 AI models from our database within a single node. Compatible models range across various parameter sizes depending on the quantization precision (BF16, FP8, INT4). Smaller models fit on a single GPU while larger models may require multi-GPU setups up to 4x B200 NVL (pair).

What is the B200 NVL (pair) inference throughput?

The B200 NVL (pair) delivers 4500 BF16 TFLOPS and 9000 FP8 TFLOPS with 16000 GB/s memory bandwidth. Actual inference throughput (tokens/sec) depends on the model size, precision, and batch size. Use our calculator for model-specific throughput estimates.

How much does B200 NVL (pair) cost per hour?

The B200 NVL (pair) is available starting from $10.50/hour via coreweave. Prices vary by provider and pricing tier (on-demand, reserved, spot). Compare pricing across all providers in the table above.