Intel Unveils Gaudi 3 AI Accelerator and Xeon 6 Processors at Vision 2024 Event

Intel CEO Pat Gelsinger unveiled the new Gaudi 3 AI accelerator and a new family of Xeon 6 processors, both offering generational performance and efficiency advantages over their predecessors and competing favorably with rivals. Gaudi 3 features 64 5th-Gen Tensor Processors Cores (TPC) and 8 Matrix Math Engines (MME), doubling the network and memory bandwidth of its predecessor, resulting in increased AI workload throughput. Intel's Gaudi 3 AI accelerator can deliver an average 50% improvement in inferencing performance and a 40% improvement in power efficiency versus NVIDIA’s H100 while claiming to do so at a fraction of the cost. The variety of form factors and configurations for Gaudi 3, in conjunction with built-in industry-standard Ethernet networking and Intel’s open software tools, allows partners to scale systems for specific needs, from 1 node up to 1024 node clusters.

Key Takeaways

• Intel unveils the Gaudi 3 AI accelerator and a new family of Xeon 6 processors aimed at advancing its AI ambitions in data centers and enterprise, boasting generational performance and efficiency advantages.
• The Gaudi 3 AI accelerator offers significant improvements in performance and efficiency over its predecessor, providing 2x FP8 and 4x BF16 compute performance, 1.5x the memory bandwidth of Gaudi 2, and more network bandwidth, leading to increased throughput for AI workloads.
• Intel Gaudi 3 will be available in various form factors, from Mezzanine cards to PCIe add-in cards, allowing partners to scale systems from 1 to 1024 node clusters with up to 8,192 accelerators.
• Intel also announced a plan to develop an open platform for enterprise AI, and a new family of Xeon processors comprising Sierra Forest and Granite Rapids architectures.
• The Gaudi 3 AI accelerator is projected to deliver a 50% faster time-to-train and outperform NVIDIA's H100 and H200 by averages of about 50% and 30% in inference throughput, while also achieving big efficiency wins.

News Content

At the Vision 2024 event in Phoenix, Arizona, Intel unveiled new hardware and software initiatives to advance its AI ambitions in data centers and the enterprise. The CEO, Pat Gelsinger, introduced the Gaudi 3 AI accelerator and a new family of Xeon 6 processors, which promise generational performance and efficiency advantages. The Gaudi 3 features advancements such as a 5nm process node, increased FP8 and BF16 compute performance, and improved network and memory bandwidth, delivering notable improvements in inferencing performance and power efficiency compared to NVIDIA’s H100.

Moreover, Intel announced a plan to develop an open platform for enterprise AI and unveiled a new family of Xeon processors targeting the edge and cloud with simplified branding. The Sierra Forest-based Intel Xeon 6 processors, featuring E-cores, will launch this quarter and target edge and high-density scale out workloads, while the Granite Rapids-based processors will target compute-intensive workloads. The Gaudi 3 AI accelerator will be available to OEMs in the second quarter of this year, with volume production scheduled for Q3 and Q4, indicating Intel's robust commitment to advancing its offerings in the AI market.

Analysis

Intel's strategic unveiling at the Vision 2024 event signals its strong commitment to advancing AI capabilities in data centers and the enterprise. The introduction of the Gaudi 3 AI accelerator and new Xeon 6 processors reflects Intel's response to the increasing demand for high-performance and efficient AI solutions. This move is aimed at challenging competitors like NVIDIA and expanding Intel's presence in the AI market. In the short term, these advancements will likely enhance Intel's competitive position and drive industry innovation. In the long term, they could establish Intel as a dominant player in the AI hardware and software space, potentially reshaping the industry landscape.

Do You Know?

• Gaudi 3 AI Accelerator and Xeon 6 Processors:

  • Intel unveiled the Gaudi 3 AI accelerator and a new family of Xeon 6 processors, highlighting their significant advancements in performance and efficiency for AI-related tasks in data centers and the enterprise.
  • The Gaudi 3 features a 5nm process node, increased FP8 and BF16 compute performance, and improved network and memory bandwidth, resulting in notable improvements in inferencing performance and power efficiency compared to competitors like NVIDIA’s H100.

• Open Platform for Enterprise AI:

  • Intel announced a plan to develop an open platform for enterprise AI, signaling a strategic move towards fostering collaboration and innovation in the AI space.
  • This initiative aims to provide a flexible and accessible environment for the development and deployment of AI solutions tailored to enterprise needs.

• Sierra Forest and Granite Rapids Based Xeon Processors:

  • Intel introduced new families of Xeon processors targeting edge and cloud computing with simplified branding, including the Sierra Forest-based processors featuring E-cores for edge and high-density scale out workloads, and the Granite Rapids-based processors focusing on compute-intensive workloads.