Dell Technologies Inc. is making its case for keeping customers’ artificial intelligence workloads on-premises with a long list of new and enhanced products being announced today at Dell Technologies World in Las Vegas under the banner of its recently introduced Dell AI Factory.

They include new flash storage systems tuned to AI workloads, a parallel file system integrated into the PowerScale storage servers, and a set of networking appliances that deliver faster throughput and lower latency.

Dell is also expanding its partnership with graphics processing unit maker Nvidia Corp., announced last year, to include new server, edge and workstation products and services for AI development and deployment. It’s also setting up a hub on Hugging Face Inc.’s AI development platform to speed generative AI onboarding for its customers.

The announcements continue Dell’s campaign to position itself as the go-to vendor for on-premises AI development. “We’ve seen on-premises processing delivering 70% greater total cost of ownership compared to the public cloud” when processing AI training and inferencing workloads, said Sam Grocott, Dell’s senior vice president of product marketing.

Flash storage for AI workloads

Dell said the new PowerScale F910 all-flash file storage system provides a performance boost of over 125% and packs 1.47 petabytes of storage into a 2u rack with 24 solid-state flash drives. It features a new parallel file system called Project Lightning that can reach 97% network saturation while serving thousands of graphic processing units, said Varun Chhabra, senior vice president of product marketing for Dell’s infrastructure solutions group. The storage server is the first to be certified for use with Nvidia Corp.’s DGX SuperPod infrastructure for large-scale AI training and inferencing.

An expanded Dell networking portfolio coming in August is led by the PowerSwitch Z9864F-ON, an Ethernet switch that the company said doubles the performance of the current PowerSwitch generation. It uses Broadcom Ltd.’s Tomahawk 5 chipset, a technology that uses Remote Direct Memory Access over Converged Ethernet to offload data movement and availability of CPU resources to the application. Dell’s PowerEdge XE9680 servers now support Broadcom’s 400G PCIe Gen 5.0 Ethernet Adapters.

The switch uses the Software for Open Networking in the Cloud open-source network operating system to simplify and standardize deployment, orchestration and lifecycle management.

“GPUs are getting larger and more demanding, and the network has to keep up,” Chhabra said.

To protect it all, the new Dell Solution for AI Data Protection service “encompasses backup, recovery and cyber resilience for AI data that helps safeguard crucial training data, models and output data,” Chhabra said. The reference design will be available in the first half of 2024.

AI partnerships expand

Dell is also adding and extending partnership deals aimed at AI development. The deal with Hugging Face Inc. will set up a development hub on Hugging Face’s development platform that enables organizations to train and deploy open customized large language models on-premises on Dell infrastructure. “Dell will be Hugging Face’s preferred on-prem infrastructure provider to support enterprise adoption of tailored open-source gen AI data sets, libraries and models,” Chhabra said. A set of services will be available to help customers get started with the LLMs.

An ongoing collaboration with Meta Platforms Inc. is aimed at simplifying the deployment of Meta’s open-source Llama 3 models on Dell infrastructure. The Dell Solution for Microsoft Azure AI Services speeds the deployment of AI services built on the Dell Apex Cloud Platform for Microsoft Corp.’s Azure cloud.

Expanded Dell AI Professional Services now cover development of Microsoft Copilots and strategic counsel on tool and model selection on Hugging Face.

Built for Nvidia

The Dell AI Factory with Nvidia, introduced two months ago, integrates Dell’s AI portfolio with the Nvidia AI Enterprise software platform, which runs the GPU maker’s Tensor Core GPUs, Spectrum-X Ethernet networking fabric and Bluefield data processing units.

Chhabra called the Nvidia co-offering “a unique instantiation of the Dell AI factory that includes Dell’s infrastructure, services and software with Nvidia’s advanced AI infrastructure and software suite, all underpinned by high-speed Nvidia networking fabric.”

A new rack server, the PowerEdge XE9680L (pictured), supports eight Nvidia Blackwell GPUs in a 4u form factor with additional PCI Express fifth generation slots and double the north/south network expansion capacity of previous PowerEdge servers, including 12 PCIe slots and full support for 400G ethernet, Infiniband and Bluefield networking. It features direct liquid cooling, which applies coolant directly to CPUs and GPUs. Availability is planned for the second half of 2024.

Dell is also announcing a series of dense and energy-efficient turnkey rack-scale servers that include an air-cooled design supporting 64 GPUs in a rack and a liquid-cooled option containing 72 Nvidia Blackwell GPUs, both in single-rack configurations. NativeEdge is a new orchestration platform that automates the delivery of Nvidia AI Enterprise software for edge deployments. It enables users to analyze edge data using deployment blueprints provided by Dell professional services that include Nvidia’s Metropolis video analytics, Riva speech and translation capabilities and NIM inference microservices.

The blueprints and automated deployment capabilities will be available in the second half of 2024. Nvidia NIM microservices are available today.

New Accelerator Services for RAG on Precision AI Workstations helps shorten the large language model development cycle using retrieval-augmented generation on a Dell Precision workstation with Nvidia’s AI Workbench development platform.

Broad storage enhancements

Outside the AI realm, Dell is announcing improvements to the PowerStore storage that enhance performance, efficiency, resiliency and multicloud data mobility. The company is also expanding its Apex portfolio of composable infrastructure with new AIOps features and enhanced multicloud and Kubernetes storage management.

PowerStore servers now feature quad-level cell storage, a type of flash memory that stores four bits of data per memory cell. Customers can start with as few as 11 QLC drives and scale up to 5.9 petabytes of effective capacity per appliance. Intelligent load-balancing capabilities save costs and improve workload placement across mixed triple-level and QLC clusters. Performance boosts of up to 66% are possible through new data-in-place higher model appliance upgrades.

Software updates Dell is making available at no additional cost to current customers can boost mixed workload performance up to 30%. Native synchronous replication for block and file workloads and native metro replication for Windows, Linux and VMware environments enhance data protection and other software enhancements provide up to 20% better data reduction and 28% more effective terabytes per watt.

The PowerStore enhancements are part of a new integrated offering called PowerStore Prime for Apex customers that combines updated PowerStore systems with programs that better protect existing storage investments. These include a 5:1 data reduction guarantee, access to live support, flexible technology upgrade options, capacity trade-in and storage advisory services.

AIOps-as-a-service

Apex AIOps software-as-a-service optimizes Dell infrastructure health and service availability with full-stack observability and incident management. Infrastructure observability resolves issues up to 10 times faster than traditional approaches through AI-powered health, cybersecurity and sustainability recommendations, Dell said. A generative AI-powered AIOps Assistant responds to infrastructure questions in natural language and provides detailed problem-solving recommendations.

Apex Navigator SaaS has been expanded to include Kubernetes storage management and additional support for Apex Storage for Public Cloud offers. The Apex Navigator for Kubernetes simplifies Kubernetes storage management on PowerFlex storage servers with support for PowerScale servers and Red Hat Inc.’s OpenShift planned. Apex Navigator for Multicloud Storage supports Apex File Storage for the Amazon Web Services Inc. cloud. Azure support is planned for later this year.

Image: Dell