No one will say the opposite: Intel missed the wave of artificial intelligence that has swept the semiconductor industry over the last ten years, despite a certain number of efforts in this area. The company is now working hard to catch up on the subject in the face of Nvidia. With the Gaudi accelerators, created by Habana Labs, with its GPUs for data centers, and of course with the eternal Xeon. On the occasion of the WAICF event which took place on February 8 and 9 in Cannes, Intel launched a large-scale seduction operation around the contribution of its CPUs to AI.
Position yourself in the inference market
According to Lisa Spelman, corporate VP and GM responsible for Xeon products and solutions, “Intel chips offer better performance for inference”. A comment which is not insignificant, inference now represents the majority of calculation needs compared to model training, a sign that these technologies have reached maturity and are now being put into production at scale.
It must also be said that the battle in data centers is also against AMD and its Epyc CPUs, whose market shares have increased rapidly over the last five years. If Intel remains dominant with around 75% of the market, it cannot be complacent. Positioning yourself as a leader in inference therefore makes sense.
A partnership strategy
Lisa Spelman gave an overview of Intel's commercial roadmap, which includes increasing partnerships with customers like OVHcloud, which are launching Xeon-based AI services. “OVH does this to provide localized support and to respond to certain challenges, particularly in terms of data regulation,” she specifies. On this subject, Thierry Souche, technical director of the French supplier, specifies that OVHcloud servers will be available with 5th generation Xeon processors later this year.
Other partner, Iterate.ai is a company focused on scaling generative AI in the enterprise. It has a low-code solution called Interplay. Currently delivered on approximately 4000 edge servers running on Xeon, it also offers services for vertical industries. Taking the example of fast food, Lisa Spelman looks at how Iterate.ai and its Interplay solution – which runs on Xeon – can improve the fast food drive-thru experience.
Iterate.ai uses a large language model coupled with voice recognition to help their customers improve the accuracy of their orders. While voice recognition can be used as a secondary source to assist on-site employees, it can also help fast food restaurants cope with periods of staffing shortages and peak traffic periods.
The Netflix case
Among the prestigious clients, we find Netflix. It uses artificial intelligence in many ways: recommendation engines, suggesting what you should watch next based on your history, etc. The Californian giant is also using AI to improve customer experience and better manage content delivery and network costs. Using Xeon processors would have doubled its performance compared to another solution.
However, the entertainment giant also uses GPUs, which Lisa Spelman admits: “There is a portion of their AI workloads that they use GPUs for”before adding “But our goal is to enable them to use their Xeon infrastructure to its full capacity and scale it up before they have to invest in an expensive GPU solution.”
Analysis and monitoring of athletes with Xeon and Gaudi
Intel has also tied up with a company called ai.io, specializing in sports performance monitoring, which uses Xeon CPUs and Gaudi accelerators to improve the tracking of Olympic athletes in 3D. Intended for high-level athletes, the solution generates complex analyzes to enable them to improve their performance and adapt their training dynamics.
The company is currently working with many athletes preparing for the upcoming Olympics, and its goal is to be able to provide them with a real-time 3D response. Since rendering can take a long time, ai.io relies on Gaudi accelerators, “allowing them to deliver this capability at a lower total cost of ownership than they had previously scaled and scaled on GPUs.”
AI PCs, the future of office computing?
Lisa Spelman also made a point of recalling the firm's investments in the development of the “AI PC” category. This relatively vague marketing term refers to personal computers with better computing capabilities for AI. To do this, Intel can count on the Movidius IP, now integrated into its CPUs since Meteor Lake. The entire “Wintel” ecosystem is on board, with Microsoft of course, but also the major PC manufacturers. Intel estimates that by 2028, around 70% of PCs will be “AI PCs”.
Selected for you