Unrivalled
Performance in…
AI Reasoning and Inference
37 TB of fast memory and 130 TB/s NVLink bandwidth power real-time inference at a scale previously impossible in a single rack.
Trillion-Parameter LLM Training
1440 FP4 PFLOPS with sparsity accelerates large language model training to new levels of speed and cost efficiency.
High-Density HPC Workloads
2,592 Arm Neoverse V2 CPU cores alongside 20 TB of GPU memory ensure CPU-bound and GPU-bound tasks run in lockstep.
Large-Scale Data Processing
Up to 576 TB/s GPU memory bandwidth eliminates bottlenecks across the largest analytic and simulation pipelines
Key Features of
NVIDIA GB300 NVL72
Faster AI
Inference
The GB300 NVL72 delivers 2x the attention-layer acceleration and 1.5x more AI compute FLOPS than standard Blackwell GPUs, helping you run larger reasoning and inference workloads with higher throughput and lower latency.
Expanded GPU Memory
With 1.5x more HBM3e memory capacity, you can support longer context windows, larger batch sizes, and more demanding AI models while maintaining maximum throughput.
Next-Gen Accelerated Computing
Blackwell Ultra brings major advances in accelerated computing, delivering the performance, efficiency, and scalability needed for next-generation AI and HPC workloads.
High-Speed AI Networking
Each GPU connects through dual NVIDIA ConnectX-8 SuperNICs, providing up to 800 Gb/s of high-speed networking for ultra-efficient AI cluster communication and RDMA performance.
High-Bandwidth GPU Interconnect
Fifth-generation NVIDIA NVLink enables high-bandwidth communication between every GPU in the rack, accelerating large-scale AI reasoning and distributed workloads.
