Intel Gaudi 3 And Dell PowerEdge Servers Deliver AI ‘Game-Changer,’ Michael Dell Says
‘Gaudi 3 strengthens our partnership by addressing customer infrastructure needs, reducing [total cost of ownership] and easing deployment. And it’s all through an open ecosystem featuring AI frameworks optimized for Dell and Gaudi 3 along with scalable Ethernet-based AI fabrics,’ Michael Dell said at the Intel Vision event Tuesday.
“Gaudi 3 strengthens our partnership by addressing customer infrastructure needs, reducing [total cost of ownership] and easing deployment. And it’s all through an open ecosystem featuring AI frameworks optimized for Dell and Gaudi 3 along with scalable Ethernet-based AI fabrics,” Michael Dell said at the Intel Vision event Tuesday.
At the event, Michael Dell said the company plans to integrate Gaudi 3 chips into its purpose-built AI system, the PowerEdge XE9680. The Gaudi 3 version of Dell’s server will be accessible within Intel Developer Cloud prior to launch.
Emphasizing that Dell is “working hard to meet the demand and make AI adoption really easy for these enterprise customers,” the company’s founder finished his brief talk at Intel Vision with a plea for Intel CEO Pat Gelsinger.
“What we need is more Gaudi 3 in volume. That would be great,” Dell said.
[RELATED: Intel’s Gaudi 3 AI Chip Targets Nvidia H100, H200; Scales To 8,000-Chip Clusters]
Intel at the event said its Gaudi 3 processor can beat rival Nvidia’s H-built chips, delivering 50 percent better inferencing and 40 percent better power than an H-100 “at a fraction of the cost.”
Deania Davidson, Dell’s director of product planning and management for Dell’s AI-optimized compute portfolio, called the partnership “a watershed moment.”
“The collaboration between Dell and Intel, crystallized in the Dell PowerEdge XE9680 with the Intel Gaudi3 AI accelerator, is a watershed moment in AI computing,” Davidson wrote in a blog post. “This evolution caters to a more diversified range of workloads, providing developers and enterprise professionals with options for pushing the limits of GenAI acceleration.”
Gary McConnell, CEO of VirtuIT, a Dell Platinum partner based in Nanuet, N.Y., is celebrating the upside of having Intel challenge Nvidia for GPU dominance.
“Two things stand out: more competition for Nvidia as Intel works through their road map of chips. The continued competition drives customer value faster,” he told CRN. “The networking piece introduces more simplicity, which we expect to continue to see as architectures are pressured do more with AI. Any time we can access data more efficiently and do so with a simplified architecture, it's a win for customers.”
Dell wrote that the PowerEdge XE9680, augmented with the Gaudi3 accelerator, introduces novel networking capabilities directly integrated into the accelerators via six OSFP 800-GbE ports.
“These links allow for direct connections to an external accelerator fabric without the need for external NICs to be placed in the system. This not only simplifies the infrastructure but also aims to lower the total cost of ownership and complexity of an infrastructure,” wrote Davidson.
Dell also said Intel Gaudi 3’s specialized media decoders are designed for AI vision applications. When combined with its PowerEdge server, they are capable of handling extensive pre-processing tasks, such as streamlining video-to-text conversions and enhancing performance for enterprise AI applications.
Davidson called the collaboration a “no-compromise AI acceleration solution”
“This partnership promises to empower technology leaders with advanced tools for innovation, pushing the boundaries of AI development and setting new standards for computational excellence and efficiency,” she wrote.
Intel is taking aim at Nvidia’s H100, which is also used by Dell in the same PowerEdge 9680 server, featuring the same eight-way architecture.
Intel said Gaudi 3 delivers four times more power and a 1.5-time increase in memory bandwidth over previous versions and gives users a “significant leap in AI training and inferencing.”
When it comes to training the models that are at the heart of generative AI systems, Gaudi 3 is projected to deliver 50 percent faster times on the Llama2 models with 7 billion and 13 billion parameters and on GPT-3, which has 175 billion parameters.