NVIDIA B200-SXM5 Blackwell SXM6 180GB AI GPU

• Brand: NVIDIA
• GPU Model: NVIDIA B200
• Architecture: NVIDIA Blackwell
• Product Class: Data-center GPU / AI accelerator
• SKU Provided: B200-SXM5
• Official Form Factor: SXM6
• GPU Memory: 180GB HBM3e
• GPU Memory Bandwidth: Up to 8TB/s
• NVLink Generation: Fifth generation
• NVLink GPU-to-GPU Bandwidth: Up to 1.8TB/s
• Maximum Configurable GPU Power: Up to 1,000W
• FP64: 40 TFLOPS
• FP32: 80 TFLOPS
• TF32 Tensor Core: 1.1 / 2.2 PFLOPS
• BF16 Tensor Core: 2.25 / 4.5 PFLOPS
• FP16 Tensor Core: 2.25 / 4.5 PFLOPS
• FP8 Tensor Core: 4.5 / 9 PFLOPS
• FP4 Tensor Core: 9 / 18 PFLOPS
• INT8 Tensor Core: 4.5 / 9 POPS
• Multi-Instance GPU: Up to 7 MIGs at 23GB each
• PCIe: Gen5, platform-dependent
• Target Workloads: AI training, AI inference, HPC, data analytics
• Primary Platform: NVIDIA HGX B200 / qualified partner systems
• Compatibility: Qualified SXM6 B200 platforms; verify server
• Performance notation: Values shown as dense / structured-sparsity performance where applicable

Why Choose This Product?

• Large 180GB HBM3e capacity: Provides substantially more local GPU memory than earlier mainstream data-center accelerators, helping organizations run larger AI workloads without immediately increasing GPU count.
• High memory throughput: Up to 8TB/s bandwidth supports memory-intensive AI and HPC processing where feeding compute engines efficiently is critical. 
• Designed for multi-GPU scaling: Fifth-generation NVLink enables high-bandwidth GPU-to-GPU communication for supported HGX architectures, reducing the connectivity bottleneck in distributed GPU workloads. 
• Blackwell platform investment: Provides an architecture designed specifically around modern AI workloads, including low-precision acceleration, Transformer Engine enhancements, and accelerated data processing.
• Enterprise AI deployment: B200 is available as the core accelerator in NVIDIA HGX B200 systems, where eight GPUs provide up to 1.44TB aggregate GPU memory and up to 64TB/s aggregate GPU memory bandwidth. 
• Workload consolidation potential: High GPU memory and compute density can allow organizations to process larger workloads per server, potentially reducing the number of nodes required for particular applications.
• Flexible workload partitioning: Supported MIG configurations can divide GPU resources into isolated instances, improving resource allocation for multi-tenant or mixed-workload environments. 
• Built for infrastructure-level deployment: The SXM6 design targets purpose-built AI servers and HGX platforms, making B200 appropriate for organizations deploying dedicated accelerated-computing infrastructure rather than standard workstation GPUs.