NVIDIA DGX Rubin NVL8 Intel Xeon 6776P 2.3TB HBM4 AI Supercomputer System with 3 Year Std. Support
NVIDIA DGX Rubin NVL8 AI Supercomputer, 8x NVIDIA Rubin GPUs, 2.3TB HBM4, 2x Intel Xeon 6776P, CX-9, 3yr Std Support
Email me when the availability or price changes
Finance this product
- Spend over £390.00 and Buy Now, Pay Later or spread the cost of your finance over 24-48 months with a 10% deposit.
- 48HR REPLACEMENT If you need to return this item, your replacement will be dispatched within 2 working days of your product arriving back at Scan.
Suggested Alternatives
PNY NVIDIA DGX B300 Deep Learning AI System, 8x B300 GPUs, 2100GB HBM3e, 72 petaFLOPS FP8 Performance, 3yr Std SupportPNY NVIDIA DGX 8x B300 2,100GB AI Server System
LN165851
No customer reviewKey Specifications
- PNY NVIDIA DGX B300 Deep Learning AI System
- 8x B300 GPUs
- 2100GB HBM3e
- 72 petaFLOPS FP8 Performance
- 3yr Std Support
£569,999.00Pre order
NVIDIA DGX Rubin NVL8 AI Supercomputer
The NVIDIA DGX Rubin NVL8 is a turnkey AI infrastructure platform built on the NVIDIA Rubin architecture, designed to accelerate demanding agentic AI, physical AI, large-scale Mixture-of-Experts models and massive-context inference workloads. Powered by eight NVIDIA Rubin GPUs with 2.3TB of HBM4 memory and up to 400 PFLOPS of NVFP4 inference performance, the DGX Rubin NVL8 provides the high-performance compute, memory bandwidth and GPU interconnect required for AI workloads at enterprise scale.
Two Intel Xeon 6776P processors provide host processing, while fourth-generation HBM4 memory across the eight GPUs delivers 176TB/s of memory bandwidth. Sixth-generation NVIDIA NVLink provides 28.8TB/s of total GPU-to-GPU bandwidth, with eight NVIDIA ConnectX-9 VPI networking interfaces supporting up to 800Gb/s InfiniBand and Ethernet. The system also incorporates NVIDIA BlueField-4 DPUs and NVIDIA Mission Control for integrated infrastructure and workload management.
Eight Rubin GPUs for High-Performance AI
Eight NVIDIA Rubin GPUs form the compute foundation of the DGX Rubin NVL8, delivering 400 PFLOPS of NVFP4 inference performance, 280 PFLOPS of NVFP4 training performance and 140 PFLOPS of FP8/FP6 training performance. The platform is designed to accelerate the complete AI lifecycle, from model training and post-training through to inference and complex reasoning workloads.
Massive HBM4 Capacity and Memory Bandwidth
The eight Rubin GPUs provide 2.3TB of HBM4 memory with up to 176TB/s of aggregate GPU memory bandwidth, allowing large datasets and AI models to be processed with high-speed access to GPU memory. This substantial memory subsystem is designed for demanding workloads including large-scale Mixture-of-Experts models and massive-context inference.
Sixth-Generation NVIDIA NVLink for GPU Communication
Four sixth-generation NVIDIA NVLink Switch systems provide 28.8TB/s of total NVLink bandwidth between the eight GPUs. High-speed GPU-to-GPU communication enables the DGX Rubin NVL8 to efficiently distribute large models and workloads across its GPU resources while supporting the massive model parallelism required by advanced AI applications.
Dual Intel Xeon Processing for Enterprise AI
Two Intel Xeon 6776P processors provide the host CPU foundation for the system, coordinating workloads, data movement and supporting infrastructure around the eight Rubin GPUs. The combination of dual Xeon processors and eight Rubin GPUs creates a balanced platform for large-scale AI training, inference and reasoning.
High-Speed ConnectX-9 Networking for AI Infrastructure
Eight OSFP ports serve eight single-port NVIDIA ConnectX-9 VPI interfaces, supporting up to 800Gb/s NVIDIA InfiniBand and Ethernet connectivity. Two 400G QSP112 interfaces serve a NVIDIA BlueField-4 DPU, providing additional high-speed networking capabilities for connecting DGX Rubin NVL8 systems and integrating them into larger AI infrastructure.
Integrated Software for AI Development and Management
The DGX Rubin NVL8 supports NVIDIA DGX OS, Ubuntu, Red Hat Enterprise Linux and Rocky, providing flexibility for enterprise AI deployments. NVIDIA Mission Control provides integrated infrastructure management spanning configuration, facility integration, cluster and workload management, as well as cooling and power events, while the NVIDIA AI software stack provides the software foundation for developing and deploying AI applications.
Built for Agentic and Next-Generation AI Workloads
The Rubin architecture incorporates a specialised multi-agent engine for reasoning workflows and a dedicated reinforcement learning engine designed to optimise memory movement in hardware. This enables the DGX Rubin NVL8 to target AI workloads requiring multi-step reasoning, long-context understanding, multimodal processing and real-time decision-making, alongside large-scale MoE models and next-generation AI factory infrastructure.
Three-Year Standard NVIDIA Support
The NVIDIA DGX Rubin NVL8 comes with three years of standard NVIDIA support, providing customers with manufacturer-backed assistance for the system throughout the standard support period.
• Eight NVIDIA Rubin GPUs.
• 2.3TB total HBM4 GPU memory.
• Up to 176TB/s GPU memory bandwidth.
• Up to 400 PFLOPS NVFP4 inference.
• Up to 280 PFLOPS NVFP4 training.
• 28.8TB/s sixth-generation NVLink bandwidth.
• Two Intel Xeon 6776P processors.
• Eight NVIDIA ConnectX-9 VPI interfaces.
• Up to 800Gb/s InfiniBand and Ethernet.
• Approx. 24kW system power usage.
• Three-year standard NVIDIA support.
