Dell PowerEdge AI Acceleration Announcements at NVIDIA GTC

Dell PowerEdge announcements at NVIDIA GTC include new eight-way GPU options, Ethernet AI fabrics and a peek at Dell's first ARM CPU.

The NVIDIA GPU Technology Conference (GTC) is NVIDIA’s premier event for AI innovators, developers and enthusiasts. Dell Technologies, a leader in AI infrastructure, is showcasing some of our latest technology collaborations with NVIDIA at the event.

In the booth and in online virtual sessions, we’re demonstrating Dell Generative AI Solutions with NVIDIA and professional services designed to bring AI to your data.

Dell - PowerEdge - Dell Technologies - servers - PE XE9680 - NVIDIA - NVIDIA GTCThe Dell PowerEdge XE9680 is our flagship eight-way GPU accelerated server for generative AI training, model customization and large-scale AI inferencing. The Dell PowerEdge XE product family gives you the freedom to choose either air-cooled or ultradense liquid-cooled configurations. PowerEdge XE servers give you even more choices for accelerated computing, like the powerful eight-way NVIDIA Tensor Core H100 found in the PowerEdge XE9680 or the four-way NVIDIA H100 HGX accelerated PowerEdge XE9640 and PowerEdge XE8640 servers. PowerEdge XE servers are NVIDIA-Certified Systems and include advanced NVIDIA Quantum-2 InfiniBand and Spectrum-X Ethernet networking solutions, as well as NVIDIA AI Enterprise software.

The highly flexible PowerEdge XE9680 celebrates three generations of NVIDIA GPUs. The XE9680 supported the NVIDIA A100 Tensor Core GPUs, based on the NVIDIA Ampere architecture, and NVIDIA H100 Tensor Core GPUs, based on the NVIDIA Hopper architecture, at launch. Today, Dell announces support for NVIDIA’s new Blackwell-architecture GPUs for AI acceleration and the following updates.

Dell PowerEdge XE9680 Support for the NVIDIA HGX H200

The PowerEdge XE9680 adds support for NVIDIA H200 Tensor Core GPUs and BlueField-3 SuperNICs. The NVIDIA HGX H200 refresh is based on the same NVIDIA Hopper eight-way GPU architecture of the PowerEdge XE9680 with NVIDIA HGX H100 with improved HBM3e memory. With an expanded 141 GB of memory per GPU, PowerEdge XE 9680 is expected to accommodate more AI model parameters for training and inferencing in the same air-cooled 6RU profile for a lower TCO. The NVIDIA BlueField-3 SuperNIC accelerates and offloads the cluster network traffic common in large AI models, enabling direct GPU-to-GPU communication and speeding up time to solutions.

Dell - PowerEdge - Dell Technologies - servers - PE XE9680 - NVIDIA - NVIDIA GTC

The Air-cooled PowerEdge XE9680 with NVIDIA HGX B100

The PowerEdge XE9680 with the NVIDIA Blackwell family HGX B100 will offer enterprises a next-gen eight-way GPU for generative AI with more processing power at the 700W power profile of the NVIDIA HGX H100. The air-cooled PowerEdge XE9680 with NVIDIA HGX B100 GPUs will feature a whopping total 1.54 TB of NVIDIA NVLink interconnected HBM3e GPU memory and enable your business to gain accurate GenAI insights with models containing trillions of parameters.

Dell’s First Liquid-cooled PowerEdge XE9680 Configuration: the NVIDIA HGX B200

We’re announcing Dell’s first liquid-cooled PowerEdge XE9680 server configuration. The DLC PowerEdge XE9680 with NVIDIA HGX B200 is expected to offer 25x more AI inferencing performance, with 20x lower TCO and 20x less energy utilization than the NVIDIA HGX H100.

Dell’s NVIDIA GB200 NVL72 Multi-node Scale-up Server Architecture Will Surpass Eight GPU Server Large AI Model Performance with Up to 72 NVLink Interconnected GPUs

In another first, Dell announces its first-ever ARM CPU with the introduction of the NVIDIA GB200 Superchip. Dell will collaborate with NVIDIA to offer a next-generation compute platform based on the NVIDIA hybrid Grace ARM CPU with on-chip Blackwell microarchitecture B200 GPUs.

This new architecture will become the new “king of the hill” for inferencing and training the largest AI models with many trillions of parameters. The GB200 NVL72 multi-node scale-up server architecture will offer an ultradense 72 NVIDIA NVLink interconnected GPUs in a single rack.

PowerEdge Support for NVIDIA BlueField-3 Ethernet Adapters

Dell PowerEdge will offer customers even more choice with support for NVIDIA BlueField-3 SuperNICs for NVIDIA Spectrum-X high-performance Ethernet AI fabrics. This high-performance fabric combines the NVIDIA Spectrum-4 switch and BlueField-3 SuperNICs as part of the NVIDIA OVX and NVIDIA HGX reference architectures for powering NVIDIA Omniverse and NVIDIA AI Enterprise workloads.

Want to learn more about these innovative and powerful solutions for AI? We invite you to go to our booth at the NVIDIA GTC conference or register online for the free virtual conference. Watch our in-person and virtual sessions to see how Dell Technologies and NVIDIA can help you accelerate your AI journey.

Please go to Dell’s NVIDIA GTC announcement blog to learn more about Dell AI Factory with NVIDIA and the Dell Technologies press release to learn more about new product updates and Dell Generative AI Solutions announcements.

Robert McNeal

About the Author: Robert McNeal

Robert McNeal is the Product Marketing lead for PowerEdge accelerated computing products at Dell Technologies. In this role, Robert oversees marketing activities for the Dell PowerEdge R760xa and XE family of GPU-accelerated servers. Robert is part of a team focused on solutions for Artificial Intelligence, Machine Learning, Deep Learning (AI/ML/DL) and High Performance Computing (HPC). Previously, Robert held technical marketing and solutions roles supporting Dell’s data storage products. Robert moved from his native Washington, D.C., to Seattle, Washington, in 2011 to work as a Solutions Architect for the unstructured data storage start-up that eventually became Dell PowerScale. Robert still lives in Seattle, where he spends his free time exploring the great outdoors with his family and faithful canine companion. Robert holds a bachelor’s degree in Liberal Arts and Sciences from Virginia Tech in Blacksburg, Virginia.