Dell Technologies Inc. today launched five new PowerEdge servers using Advanced Micro Devices Inc.’s 5th Generation EPYC processors and targeted at artificial intelligence development and model deployment.
The announcements represent “a massive expansion of our partnership with the AMD,” said Varun Chhabra, senior vice president of product marketing at Dell.
The PowerEdge XE7745 is designed for enterprise-class workloads, with support for up to eight double-width or 16 single-width Peripheral Component Interconnect Express-based graphic processing units and AMD EPYC processors in a 4U air-cooled chassis. The internal GPU slots are paired with eight additional PCIe 5.0 slots for network connectivity.
PCIe 5.0 offers significantly higher data transfer rates than its predecessors, with double the bandwidth of PCIe 4.0 and up to 32 gigatransfers — or the number of signal transitions per second — per lane. The XE7745 will be available globally starting in January.
Open chassis
The new PowerEdge R6725 and R7725 servers are optimized for scalability with a Data Center Modular Hardware System chassis design that enables enhanced air cooling and dual 500-watt CPUs. Dell said the R7725 offers up to 66% increased performance and up to 33% increased efficiency at the top of the stack. DCMHS is a standardized, modular approach to designing and building data center infrastructure developed under the Open Compute Project.
“We built a brand new chassis with DCM’s designs that will enhance our air cooling capability and can do 192 cores per CPU and 500-megawatt CPUs for unmatched power and efficiency,” said Arunkumar Narayanan, Dell’s senior vice president of server and networking products. “We will be industry-leading in the ability to do air cooling in a 2U capability with 500 watts.”
All three platforms can support up to 50% more cores and up to 37% increased performance per core based on Dell’s analysis. They collectively consolidate up to seven five-year-old servers into one, lowering CPU power consumption by 65%. They can be remotely monitored, managed and updated using Dell’s Integrated Dell Remote Access Controller. which runs independently of the operating system and allows for remote management even when the server is powered off.
The PowerEdge R6715 and R7715 servers based on 5th Gen EPYC processors are faster than the previous generation and use also double memory capacity with support for 24 dual in-line memory modules in compact 1U and 2U chassis configurations. Availability is planned for November.
AI Factory on Hugging Face
Dell is also enhancing its AI Factory, which is rolled out in March. Described as an “end-to-end AI enterprise solution” for training, tuning and running AI models, it combines Nvidia Corp. GPUs with Dell’s compute, storage, client device and software products and professional services.
Dell said new generative AI capabilities using PowerEdge XE9680 servers with AMD Instinct MI300X accelerators simplify AI deployment, enhance security, allow for scalable and modular architectures and reduce time-to-value by up to 86%. The configuration is available as the Dell Enterprise Hub on Hugging Face Inc.’s AI platform, providing custom containers and scripts for the deployment of models such as Llama and Mixtral.
The XC9680 was “the most successful product launch in Dell’s history,” Narayanan said. “We have had more than $10 billion of revenue on that one product. Now we are launching our next generation.”
The containerized models are optimized to boost inferencing performance based on the model and server and leverage the Hugging Face Text Generation Inference, an inference framework designed to serve and scale large language models efficiently. Dell’s Professional Services for Generative AI are being expanded to support AMD environments.
“We’ve collaborated very closely with AMD and Hugging Face to make it easy for customers to deploy advanced models like Llama and Mixtral on the PowerEdge XC9680 infrastructure that takes advantage of AMD’s MI300 accelerators,” Chhabra said. “This partnership delivers custom containers, scripts and dedicated technical resources so that customers can have the peace of mind that they can deploy the most popular open-source gen AI models on the Hugging Face platform and know that it has been tested by Dell, AMD and Hugging Face.”