Cool AI in the data center: Dell presents heat exchangers for AI hardware

At its in-house exhibition in Las Vegas, Dell presented a heat exchanger for computing hardware and a data platform for AI applications. Under the umbrella of its so-called Dell AI Factory, the US manufacturer intends to provide a product suite designed to support companies in setting up and operating AI workloads in their data centers. In particular, Dell is targeting key challenges faced by its customers when using artificial intelligence, such as energy efficiency, data quality, security and operating expenses.

Heat exchanger allows higher rack density

The PowerCool Enclosed Rear Door Heat Exchanger (eRDHx) is a heat exchanger with a closed airflow system. It is designed for the heat development of AI hardware and, according to the company, can capture up to 100 percent of the waste heat. The system works with increased water inlet temperatures of 32 to 36 degrees Celsius. According to Dell, this should reduce cooling costs by up to 60 percent. At the same time, the heat exchanger enables up to 16 percent higher rack density and an air cooling capacity of up to 80 kilowatts per rack.

Dell has also updated its AI Data Platform. With Project Lightning, it now includes a parallel file system which, according to Dell, can achieve twice the data throughput of competing systems. In this way, Dell aims to reduce the training time of language models for complex AI workflows. The US manufacturer has also expanded its Data Lakehouse, which now includes vector and full-text search, as well as SQL functions for language models. Possible applications for the new features include recommendation systems or semantic searches and recognition of customer intentions.

Nvidia integration accelerates S3 data throughput

Dell has also released several new features for its ObjectScale platform. This includes greater integration with Nvidia's BlueField-3 data processing unit and Spectrum 4 networking platform. In addition, ObjectScale now supports S3 object storage via Remote Direct Memory Access (RDMA). With these updates, Dell aims to increase object memory throughput by up to 230 percent while reducing CPU utilization by up to 98 percent compared to traditional S3 data transfer.

The US manufacturer also presented Linear Pluggable Optics (LPO) in the area of network technology. AI and HPC applications can be connected without a digital signal processor using the company's own Ethernet switches from the PowerSwitch series. In this way, the company aims to reduce energy consumption and costs. Dell is also adding the SN2201 and SN5600 switches from the Spectrum-X series and the Quantum-X800 InfiniBand switches from Nvidia to its network portfolio. They achieve transfer rates of up to 800 gigabits per second.

The US manufacturer also presented several new servers from the PowerEdge series for the use of AI workloads at Dell Technologies World 2025. Dell also added a laptop with a dedicated NPU for local AI conferences in the enterprise sector to its portfolio.

(dahe)