Keeping a Cool Head with #1 DLC Server Supplier

Embracing and cooling the latest high-performance GPU and CPUs for AI is hot topic in more ways than one.

Dell Technologies has shipped more DLC (direct liquid cooled) servers than any other server OEM¹ and is at the vanguard of the server technologies designed to support AI and other demanding server workloads. Dell has used that experience and expertise to design a portfolio of purpose-built direct liquid cooled and air-cooled servers to support the latest CPUs and GPUs. Incorporating Dell smart cooling—the DLC-enabled server portfolio—lets Dell customers take compute density per rack and cooling efficiency to the next level, while air-cooled platforms allow deployment in almost any data center.

Leveraging Dell thermal engineering expertise, Dell Smart cooling harnesses multi-vector cooling to manage thermals and fan behavior across the entire portfolio. During product development, Dell thermal engineers use simulation-driven design changes to improve speed and efficiency, like uncovering significant airflow gains allowing higher thermal design power (referred to as TDP) than the traditional legacy physical iterations. This method increases accuracy and reduces product development time.

DLC servers - direct liquid cooled - Dell - PowerEdge - Dell Technologies
Figure 1 Simulation Driven Product Development

PowerEdge XE: High Density and Silicon Diversity in Action

PowerEdge XE servers are acceleration-optimized, purpose-built for Artificial Intelligence (AI) and High Performance Computing (HPC). With superior acceleration and diverse GPU options, these powerful platforms are optimized to turn data into action faster. The portfolio offers a choice of GPU/brand and a variety cooling options and capabilities; Air, LAAC liquid assisted air cooling and DLC direct liquid cooling. This silicon diversity allows researchers and data scientists to fine-tune their systems for optimal performance, whether they’re processing natural language, analyzing vast datasets, or training complex neural networks.

DLC servers - direct liquid cooled - Dell - PowerEdge - Dell Technologies
Figure 2 Choice of Air or liquid cooled GPU accelerated in XE server portfolio.

XE8640 Air Cooled Four-way GPU Server

The PowerEdge XE8640 is a 4U air-cooled platform offering four NVIDIA H100 Tensor core 700W SXM5 GPUs fully interconnected with NVIDIA NVLink. Its air-cooled design supports the highest wattage CPUs and GPUs—up to 35 C ambient temperatures. Designed to let customers deploy this server in almost any data center, the XE8640 simplifies data center operations using Liquid Assisted Air Cooling (LAAC) cooling. This method uses liquid to collect the heat from the CPUs and GPUs, which is then transported to an air-cooled radiator housed at the rear of the server. The radiator is then cooled by leveraging the data center chilled air.

DLC servers - direct liquid cooled - Dell - PowerEdge - Dell Technologies
Figure 3 XE8640 LAAC assembly the air-cooled closed loop internal liquid Cooled Module for GPUs.

XE9640 Density and Cooling-optimized GPU Diversity Platform

The PowerEdge XE9640 is a 2U dual socket server using full DLC to cool both the CPU and GPU. The XE9640 builds on Dell’s proven track record with liquid cooling to support either four Intel Data Center Max GPU series 1550 fully interconnected with XeLink or four NVIDIA H100 SXM GPUs fully interconnected with NVLink. The XE9640 also supports the use of standard PCIe cards to offer a choice of connectivity options. Direct liquid cooling technology provides a more efficient and cost-effective solution than traditional air-cooled systems. It delivers double the rack density of the air-cooled sister product, the Dell XE8640 (2U vs. 4U) rack footprint while reducing the demands for chilled air in the data center. Based on Dell’s proven track record of liquid cooling, these servers are offered in 42U or 48U racks with liquid cooling components, including cooling distribution units (pumps), rack manifolds (like a PDU but for coolant) and services to build and install entire DLC solutions. Customers are already utilizing this platform today, like Cambridge University, which built the UK’s fastest AI supercomputer cluster based on the Dell XE9640.

DLC servers - direct liquid cooled - Dell - PowerEdge - Dell Technologies
Figure 4 XE9640 the internal liquid Cooled Module for NVIDIA GPUs.

Both the XE8640 and XE9640 equip the customer with incredible processing power, as seen in recent performance results using BERT language model inferencing workload (PowerEdge XEx640 performance results). Deploying the XE9640 would double the inferencing performance per rack compared to the XE8640. Dell recently published a comparison infographic of the benefits of each of the XE9640 and XE8640 configurations, including highlighting the typical return on investment of DLC in 1.3 years. StorageReview, a third-party analyst organization, also worked with Dell product management to produce a hands-on, lids-off PowerEdge XEx640 performance results, exploring the features of both server platforms.

XE9680 Extreme Acceleration Eight-way GPU Server

Designed for customers with the need to train the most demanding machine learning and deep learning large models, the Dell PowerEdge XE9680 is a dual-socket, 6U air-cooled rack GPU server offering customers more choice when it comes to AI performance. Designed to support eight NVIDIA H100 SXM5 or eight NVIDIA A100 SXM4 GPUs and shortly upcoming AMD Instinct™ MI300X accelerator, offering customers more choice when it comes to AI performance.

Innovation Driving Progress

Dell’s commitment to innovation and expertise in server technologies shines through its leadership in shipping DLC servers and its comprehensive portfolio designed to meet the demands of AI and other intensive workloads. By prioritizing performance, efficiency and versatility, Dell enables customers to harness the power of advanced computing while maximizing cooling efficiency. As the landscape of server technology continues to evolve, Dell remains at the forefront, driving progress and empowering organizations to achieve their computational goals with confidence. Dell Technologies makes finding the right server for your AI workload and your data center easier than ever. Learn more about Dell Smart Cooling at Data Center Power and Cooling Solutions.

1 Disclaimer: Estimate based on internal analysis of shipments between 2018 and 2022.

Travis Vigil

About the Author: Travis Vigil

Travis Vigil is Senior Vice President leading Portfolio and Product Management for Dell’s Infrastructure Solutions Group (ISG). He and his team are responsible for Dell’s Server, Storage, Data Protection, CI/HCI, Networking and Solutions businesses. He has over 20 years of Product Management, Marketing and Business Operations experience with technology companies including Intel and Dell. In previous roles at Dell, he served as Senior Vice President for Storage and Data Protection Product Management, and Senior Vice President for Business Operations focused on Dell’s Server, Storage and Networking Businesses. He has a B.S. from Stanford University and an M.B.A. from Northwestern University’s Kellogg School of Management.