B200 Specifications
Blackwell architecture introduces FP4 precision and fifth-generation NVLink, delivering up to 4x the effective AI throughput of H100 while supporting trillion-parameter model training.
180 GB
HBM3e Memory
2.25x the capacity of H100 for very large models.
1.8 TB/s
NVLink Bandwidth
Fifth-gen NVLink, 2x the bandwidth of H100.
4,500
TFLOPS FP8
9,000 TFLOPS at FP4 with Blackwell engine.
1.6X
Faster Inference
vs H100 for GPT-3 175B.
NVIDIA B200 on Denvr AI Cloud
1T Parameter Training
Scale across 8-GPU NVLink nodes for 1,440 GB of total VRAM and 1.8 TB/s per-GPU interconnect.
Next-Gen Inference
Serve the largest models with full context size using 180 GB per GPU. FP4 quantization enables higher throughput and lower cost-per-token than Hopper.
Step Up from H100
B200 delivers up to 4x the AI throughput of H100 with 2.25x the memory. For teams needing faster iteration on large models.
Managed Storage
High-performance Weka filesystem and local NVMe available. No external storage to provision for datasets, checkpoints, or model artifacts.
Configurations
Per-minute billing with on-demand and reserved options. All configurations available as bare metal, VM, or model endpoints.
Platform
GPUs
On-Demand
VRAM
vCPUs
Memory
Local Storage
Interconnect
NVIDIA B200 SXM
8
180 GB
224
2048 GB
6x 3.8TB NVMe
RoCE or IB 3200G
Reserved only