NVIDIA GB300 NVL72
Reserve
NVIDIA GB300 NVL72
Deploy larger AI models, process longer context windows and scale high-performance inference with the NVIDIA GB300 NVL72, delivering 72 Blackwell Ultra GPUs, 37 TB of fast memory and 1440 FP4 PFLOPS in a single rack.
Reserve your allocation today with no deposit required
Why Reserve
NVIDIA GB300 NVL72
Built for Advanced AI Reasoning
Powered by NVIDIA Blackwell Ultra, the GB300 NVL72 delivers massive gains in AI reasoning performance with 2x faster attention processing compared to standard Blackwell GPUs, helping teams run larger and more complex models efficiently.
288 GB HBM3e Memory Per GPU
Each GPU includes 288 GB of HBM3e memory, giving you more capacity for larger batch sizes, longer context windows, and higher inference throughput across demanding AI workloads.
37 TB of High-Speed Memory
With 37 TB of combined GPU and CPU memory across the system, the GB300 NVL72 is designed to support trillion-parameter models and memory-intensive AI workloads without performance bottlenecks.
130 TB/s NVLink Connectivity
Fifth-generation NVLink delivers 130 TB/s of total bandwidth between all 72 GPUs, enabling ultra-fast communication and efficient scaling for distributed AI and HPC environments.
Integrated NVIDIA Grace CPUs
2,592 Arm Neoverse V2 CPU cores work alongside the GPUs with high-bandwidth shared memory, accelerating preprocessing, orchestration and data-intensive tasks without slowing GPU performance.
Power the New Era of Generative AI
Build and run real-time inference on trillion-parametre large language models. Enable faster insights, more accurate models, and more efficient operations across a variety of fields.
Transform generative AI and accelerated computing in data processing, electronic design automation, computer-aided engineering and quantum computing.
