James Ding
Dec 15, 2025 19:01
NVIDIA’s MGX architecture is set to transform data centers with flexible, efficient designs that cater to the growing demands of AI-driven workloads.
The rapid advancements in artificial intelligence (AI) are reshaping the computing landscape, creating unprecedented demands on data centers. NVIDIA’s latest offering, the MGX modular reference architecture, is designed to meet these challenges with flexible and efficient solutions, according to NVIDIA.
Innovative Modular Design
NVIDIA MGX introduces a 6U (800 mm) chassis configuration, tailored for the next generation of accelerated compute and networking platforms. This architecture supports a range of computing needs, from today’s powerful platforms to future innovations, minimizing the need for disruptive redesigns. The MGX 6U platform allows for integration with multiple host-processor modules, including the next-generation NVIDIA Vera CPU, offering a unified server design compatible with various CPU architectures.
Enhanced Serviceability and Sustainability
The MGX platform emphasizes serviceability by providing accessible pathways for maintenance and easy replacement of key components such as network cards and power supplies. This design reduces operational overhead and enhances the efficiency of managing rack-scale infrastructure. Additionally, the MGX 6U supports the new liquid-cooled NVIDIA RTX PRO 6000 Blackwell Server Edition GPU, which boosts performance and thermal efficiency for demanding AI workloads.
Advanced AI Networking Capabilities
Featuring NVIDIA ConnectX-8 SuperNICs with integrated PCIe Gen 6 switches, the MGX 6U platform offers high-performance AI networking. This configuration supports up to 400 Gb/s of network bandwidth per GPU, effectively doubling per-GPU network capacity and eliminating I/O bottlenecks. This setup ensures scalable, high-performance AI workloads across multiple GPUs and nodes.
Security and Infrastructure Acceleration
To address the growing complexity of modern data centers, NVIDIA’s BlueField data processing units (DPUs) are integrated into the MGX architecture. These DPUs enhance security by offloading functions like encryption and real-time threat detection, enabling zero-trust security while preserving computing resources for AI tasks. This approach ensures that AI pipelines are protected from threats, maintaining consistent performance across AI factory deployments.
Future-Proofing AI Factories
With NVIDIA’s MGX architecture, data centers are poised to evolve alongside advancements in GPU technology. The architecture reduces engineering costs and accelerates time to market for new platforms, ensuring optimal performance for AI workloads. Systems featuring the liquid-cooled NVIDIA RTX PRO 6000 Blackwell Server Edition GPU are expected to be available from global builders in the first half of 2026, marking a significant step forward in data center evolution.
Image source: Shutterstock
