Intel has declared the latest improvements in AI infrastructure, including the Xeon 6 processors with Performance-cores (P-cores) and Gaudi 3 AI accelerators. These products are intended to meet the growing demand for efficient and cost-effective computing systems optimised for AI and high-performance computing (HPC) applications.
The Intel Xeon 6 CPUs aim to provide twice the performance of their predecessors, with increased core counts, greater memory bandwidth and integrated AI acceleration capabilities. This architecture is meant to support compute-intensive applications in a variety of contexts, including edge computing, data centres and cloud services.
Whereas, the Intel Gaudi 3 AI Accelerator complements the Xeon 6 by focussing on large-scale generative AI workloads. It includes 64 Tensor processing cores (TPCs) and eight matrix multiplication engines (MMEs) to accelerate deep neural network operations. The Gaudi 3 has 128 gigabytes (GB) of HBM2e memory for training and inference tasks, as well as 24 200 Gigabit Ethernet ports for scalable networking. Notably, it is compatible with the PyTorch framework and advanced Hugging Face transformer models.
Intel and IBM will collaborate to deploy Gaudi 3 AI accelerators as a service on IBM Cloud, with the goal of lowering total cost of ownership and improving performance for enterprise AI systems. This program is part of Intel's larger strategy to provide flexible and scalable AI deployments using its x86 infrastructure.
As per Intel 73 per cent of GPU-accelerated servers use Intel Xeon as the host CPU, demonstrating the importance of Xeon in the AI ecosystem. The company collaborates with key OEMs, such as Dell Technologies and Supermicro, to design customised systems that satisfy unique customer needs for successful AI adoption.
To address the complexity of moving generative AI solutions from prototypes to production, Intel is collaborating with its partners. These efforts are aimed at creating production-ready retrieval-augmented generation (RAG) solutions that include microservices optimised for both the Xeon and Gaudi AI platforms.
Intel's Tiber portfolio offers to streamline enterprise AI applications, addressing issues such as access, cost, complexity, security, efficiency and scalability. The Intel Tiber Developer Cloud now provides preview systems for Xeon 6 processors for technology evaluation. Select customers will also have early access to Gaudi 3 for verifying AI model deployments, with clusters set to release next quarter for larger-scale implementations.