NVIDIA B100 SXM
Air-cooled HGX variant of the Blackwell architecture with 192 GB HBM3e memory. Lower power envelope than B200 (700 W vs 1000 W) at roughly 78% of the compute, suited to existing HGX H100 chassis upgrades.
VRAM
192 GB
Memory
HBM3e
Bandwidth
8000 GB/s
TDP
700W
Suitable Workloads
Large Language Models
Training and inference for models like GPT-4, Llama 70B+
Deep Learning Training
High-performance training for neural networks
Distributed Training
Multi-node training with fast interconnects
High-Throughput Inference
Optimized for batched inference workloads
Key Highlights
- Built on Blackwell architecture
- SXM form factor
Memory
Compute Performance
Architecture
Power & Physical
Interconnect
Pricing
Notes
Compute throughput shown with 2:4 structured sparsity. Drop-in replacement for HGX H100 8-GPU baseboards.
Models That May Fit on NVIDIA B100 SXM
Estimates based on INT8 quantization. Actual fit depends on framework and batch size.
Explore More
Added Apr 30, 2026
Last updated: Apr 30, 2026
Start building with the right model.
From model selection to production, one platform, no fragmentation.