The timing wasn't an accident. Training large language models demands enormous parallelism, exactly what GPUs are built for, so as generative AI captured attention, the GPU became shorthand for AI infrastructure itself.
With its new Xeon 6 portfolio, Intel is providing organizations with a broad family of socket-compatible processors specifically designed to accommodate diverse requirements across the entire data center, from high-performance AI inference to densely packed, energy-efficient cloud computing.
CPUs Play a Critical Role in Enterprise AI
GPUs will remain instrumental for many AI workloads, especially those involving large-scale model training. But training is not necessarily the only — or even the most significant — aspect of AI activities for many organizations.
Rather than creating their own proprietary foundational models, many enterprises are choosing to use pretrained models as a starting point for their AI initiatives, then fine-tuning those models to suit their specific business requirements. Once those models go live in production environments, inference typically constitutes the largest part of their operational lifecycle. Many of these applications use smaller, specialized models that are accessed by a relatively small number of employees.
For instance, a legal department may use a retrieval-augmented generation application to analyze large volumes of contract documents and answer questions related to their content. Similarly, HR departments and customer service centers may use similar types of targeted AI tools. Such workloads present opportunities for organizations to leverage their existing Xeon infrastructure as opposed to automatically investing in dedicated GPU systems, Bartley says: “Most of our customers start with internal use cases, then expand to customer-facing ones as they get comfortable.”
Click the banner below to learn how to turn complexity into a business advantage.
