Dell Technologies has unveiled a new series of advanced PowerEdge servers and complementary hardware designed to meet the demanding needs of large-scale AI deployments. These cutting-edge innovations, announced recently, are tailored to boost performance and efficiency in data centres involved in sophisticated computations.

At the forefront is the PowerEdge XE9712 server, noted for its high-performance dense acceleration capabilities suitable for large language model (LLM) training and real-time inferencing. This server is powered by Nvidia’s GB200 NVL72, featuring up to 36 NVIDIA Grace CPUs and an impressive array of 72 NVIDIA Blackwell GPUs. These GPUs are interconnected through an NVLink domain, enabling the setup to function as a single GPU unit. This configuration promises to deliver up to 30 times faster real-time inferencing for trillion-parameter LLMs, significantly enhancing computational efficiency and speed.

The portfolio also includes the Dell PowerEdge M7725, engineered for dense compute performance. Targeting sectors such as research, government, fintech, and higher education, the M7725 offers scalability between 24,000 to 27,000 cores per rack. This is facilitated by 64 or 72 two-socket nodes leveraging the power of 5th generation AMD Epyc processors. For optimal thermal management, the server utilises a combination of direct liquid cooling and traditional air cooling methodologies.

In addition to these computational enhancements, Dell is addressing the growing need for efficient data storage and management through its PowerScale storage devices. These devices are poised to boost AI application performance, providing simplified global data management solutions. The enhanced PowerScale models feature faster metadata processing and leverage the Dell Data Lakehouse discovery system. New 61TB drives allow for an increase in storage capacity while halving the data centre storage footprint. Furthermore, PowerScale now incorporates InfiniBand capabilities alongside support for 200GbE Ethernet adapters, achieving up to a 63% increase in throughput.

To house and operate this sophisticated hardware, Dell is introducing the Integrated Rack 7000 (IR7000). Built to OCP (Open Compute Project) standards, the IR7000 is designed to cater to the rigorous demands of accelerated computing. It offers greater density, innovative sustainable power management, and cutting-edge cooling technologies that can efficiently dissipate heat. The IR7000 is specifically engineered to support future deployments capable of up to 480KW cooling, capturing nearly all the heat generated.

This integrated, plug-and-play rack-scale system can accommodate not only Dell hardware but also off-the-shelf networking solutions, providing flexibility in deployment configurations. Such advancements underscore Dell's commitment to driving efficiency and performance in AI and advanced computational environments, catering to the evolving needs of modern data centres.

Source: Noah Wire Services