Intel is intensifying its efforts to embed AI in corporate edge networks, data-centered servers, and other types of devices. The company made announcements during its Intel Vision 2024 event regarding the development of Ethernet network interface cards (NICs) based on the upcoming Ultra Ethernet Consortium (UEC) standards, new Xeon 6 processors with superior performance abilities, and the Gaudi 3 AI accelerator, specifically designed for the processing of AI training and inference tasks in extensive AI generative environments.
According to Sachin Katti, senior vice president and general manager of the network and edge group at Intel, the company’s aim is to “deliver AI systems that extend from the API to enterprise IoT and edge to the data center, to simplify the adoption and effective utilization of AI for businesses, broadly speaking”.
Among the planned products are numerous Ethernet-powered Intel AI NIC and AI connectivity chiplets for XPU integration, Gaudi-based accelerator systems, and a diverse collection of soft and hard AI interconnect reference designs for Intel Foundry. Katti added this information to his statement.
The Ethernet-based systems will comply with the open network fabric requirements of the UEC, set to be published later in the year. Intel is one of the founding members of the UEC, a group now composed of over 50 vendors who are innovating to enhance the scalability, stability, and dependability of Ethernet networks to meet the high-performance networking demands of AI. The UEC specifications will emphasize a range of scalable Ethernet advancements like superior multi-path and packet delivery options, as well as modern congestion and telemetry features.
Intel, in collaboration with SAP, Red Hat, VMware, and others, declared a plan to establish an open platform that will expedite the enterprise implementation of secure AI systems that generates data. This strategy is aimed at providing enterprises with an open, standards-based networking connectivity when forming AI systems, allowing a secure, large-scale distribution of AI models, as mentioned by Katti.
Intel has also rolled out its Xeon 6 processors, which it claims will lead to considerable energy savings and will endow businesses with the capability to back AI learning and features like retrieval-augmented generation (RAG). The latter involves pulling data from an exterior knowledge base to create and support extensive language models, or LLMs.
Compared to the second-generation Intel Xeon processors, the new Xeon 6 ones boast a fourfold performance enhancement and almost three times the rack density, as per Intel’s statement.
Intel has set its sights on Nvidia and the extensive AI processing requirements. The company has launched the Gaudi 3 AI accelerator chip, which it alleges is, on average, 40% more power-efficient than the corresponding Nvidia H100 chips. The announcement was made by Intel CEO Pat Gelsinger at the Intel Vision event in Phoenix, Arizona.
“The Intel Gaudi 3 AI accelerator will power AI systems with up to tens of thousands of accelerators connected through the common standard of Ethernet,” Intel declared. For instance, each Intel Gaudi 3 accelerator integrates 24 200-gigabit Ethernet ports, allowing for flexible and open-standard networking.
The Intel Gaudi 3 offers 4x more AI computational power and a 1.5x boost in memory bandwidth compared to its predecessor, the Gaudi 2, enabling efficient scaling to cater to large compute clusters and prevent vendor lock-in by proprietary networking fabrics, as per Intel.
The proposition is that this accelerator can bring about a significant improvement in performance for AI training and inference models, offering businesses the freedom to choose the systems they deploy for scaling generative AI, stated Katti.
The Intel Gaudi 3 accelerator will be accessible to original equipment manufacturers in the second quarter of 2024 in industry-standard configurations of Universal Baseboard and open accelerator module (OAM). Among the vendors that will incorporate Gaudi 3 in servers and other hardware are Dell Technologies, Hewlett Packard Enterprise, Lenovo, and Supermicro. The general availability of Intel Gaudi 3 accelerators is slated for the third quarter of 2024.
