Ultra-Large Model Training & High-Density Computing
The NVIDIA B300 Tensor Core GPU features dual-module architecture designed for ultra-large model training and high-density computing workloads.
NVL16 Architecture Revolution
NVIDIA B300 redefines the limits of ultra-large scale AI computing
Key Highlights
Discover the revolutionary capabilities that make B300 the ultimate choice for ultra-large model training and high-density computing.
Blackwell NVL16 Architecture
B300 features dual-module Blackwell NVL16 architecture, delivering exceptional performance for ultra-large model training and high-density computing.
Ultra-Large Model Training
Designed for training the largest AI models, B300 provides the computational power needed for trillion-parameter models and beyond.
High-Density Computing
B300's dual-module design enables high-density computing configurations, maximizing performance per rack unit for data center efficiency.
Advanced Memory Architecture
B300 features advanced memory architecture optimized for large-scale AI workloads, providing the memory capacity and bandwidth needed for complex models.
Technical Specifications
Comprehensive technical specifications for the NVIDIA B300 Tensor Core GPU.
| Technical Detail | B300 SXM | B300 NVL |
|---|---|---|
| Technical Detail | B300 SXM | B300 NVL |
| FP64 | TBD | TBD |
| FP64 Tensor Core | TBD | TBD |
| FP32 | TBD | TBD |
| TF32 Tensor Core² | TBD | TBD |
| BFLOAT16 Tensor Core² | TBD | TBD |
| FP16 Tensor Core² | TBD | TBD |
| FP8 Tensor Core² | TBD | TBD |
| INT8 Tensor Core² | TBD | TBD |
| GPU Memory | Dual Module | Dual Module |
| GPU Memory | Bandwidth | TBD | TBD |
| Decoders | TBD | TBD |
| Confidential Computing | Supported | Supported |
| Max Thermal Design Power (TDP) | Up to 1600W (configurable) | Up to 1600W (configurable) |
| Multi-Instance GPUs | TBD | TBD |
| Form Factor | SXM | PCIe Dual-slot air-cooled |
| Interconnect | NVIDIA NVLink™: TBD PCIe Gen5: 128GB/s | 2- or 4-way NVIDIA NVLink bridge: TBD PCIe Gen5: 128GB/s |
| Server Options | NVIDIA HGX™ B300 partner and NVIDIA-Certified Systems™ with 4 or 8 GPUs | NVIDIA MGX™ B300 NVL partner and NVIDIA-Certified Systems with up to 8 GPUs |
| NVIDIA AI Enterprise | Included | Included |
1. Preliminary specifications. May be subject to change. 2. With sparsity.
Experience Next-Level Performance
Experience the power of ultra-large model training with NVIDIA B300. Contact our experts to learn how this revolutionary GPU can accelerate your AI initiatives.