
Intel CEO Pat Gelsinger responded to Jensen Huang's assertions during his speech at Taiwan's Computex expo by stating that Moore's Law is alive and well. He emphasized that Intel will have a major role to play in the proliferation of AI as a leading provider of PC chips. Gelsinger showcased Intel's new Xeon 6 data-centre processors with more efficient cores and highlighted the company's Gaudi systems, which compile its chips into kits of multiple processors tailored to handle generative AI training. He also announced the pricing for Intel's Gaudi 2 and Gaudi 3 AI accelerators, undercutting Nvidia's offerings. Gelsinger's overall message was that Intel is not backing down from the challenges posed by competitors and is ready to compete in the era of artificial intelligence.

Intel has introduced the Intel® Gaudi® 3 AI accelerator, which is designed to bring performance, openness, and choice to enterprise generative AI (GenAI). The Gaudi 3 AI accelerator offers 4x more AI compute for BF16 and a 1.5x increase in memory bandwidth compared to its predecessor. It is purpose-built for high-performance, high-efficiency GenAI compute and features a heterogeneous compute engine comprised of 64 AI-custom and programmable Tensor Processor Cores (TPCs) and eight Matrix Multiplication Engines (MMEs).
The Gaudi 3 AI accelerator is manufactured on a 5nm process and offers significant advancements over its predecessor. It allows activation of all engines in parallel, enabling the acceleration needed for fast, efficient deep learning computation and scale. The accelerator provides 128GB of HBM2e memory and 24 200Gb Ethernet ports for flexible on-chip networking.
The Gaudi systems, which compile Intel's chips into kits of multiple processors tailored to handle generative AI training, are intended to be offered by partners like Dell Technologies and Inventec. One kit with eight Intel Gaudi 2 accelerators will sell for $65,000, while a more powerful kit of eight Intel Gaudi 3 accelerators will list at $125,000. Intel estimates that both offerings are more affordable than competitors' options.
Each Gaudi 3 cluster is composed of 8,192 accelerators, and Intel estimates it offers up to 40% faster time to train an AI model compared to an equivalent size cluster of Nvidia H100 GPUs. Intel also claims that Gaudi 3 would be up to two times faster than Nvidia's H100 in executing AI inference tasks, as measured in popular models like those made by Meta Platforms and Mistral. The Gaudi systems are aimed at accelerating the deployment of secure GenAI systems, enabled by retrieval-augmented generation.

Intel's new Xeon 6 data-centre processors, introduced at the Computex expo, feature more efficient cores that allow operators to reduce the space required for a given task to a third of what was needed with prior-generation hardware. The processors are designed for public and private clouds in situations where power efficiency and performance are critical1. The first wave of these processors is the Xeon 6 6700E series, based on Sierra Forest, featuring up to 144 E-cores per socket5. The processors promise 3:1 rack consolidation, 4.2x rack-level performance gains, and 2.6x performance per watt gains.