Intel Corp. today debuted a line of central processing units based on its latest manufacturing process and shared new details about its next-generation graphics cards.

Both product families are geared toward artificial intelligence data centers. Intel is also prioritizing the AI infrastructure market with a new series of Ethernet chips that it announced in conjunction.

The Xeon 6+ product series comprises six CPUs based on Intel’s Darkmont core design, which prioritizes power efficiency over performance. The most advanced chip in the lineup, the Xeon 6990E+, features 288 cores. They’re backed by an L3 cache that is several times larger than the one in Intel’s previous-generation CPU. The company says the processor provides 30% higher power efficiency and performance per thread than Advanced Micro Devices Inc.’s comparable Epyc 9965.

The Xeon 6+ series is the first data center CPU series to use Intel’s latest Intel 18A process. One of the node’s main selling points is a technology called RibbonFET that enables engineers to closely align a processor’s design with workload requirements.

Transistors include a component called a channel that conducts electricity. Intel’s RibbonFET technology implements the channel not as a single structure, but rather as several thin sheets suspended above one another inside a rectangular frame. Engineers can adjust the width of the sheets to change the transistor’s power efficiency and speed. That customizability makes it possible to tailor a RibbonFET to specific use cases.

Hardware makers can include up to two Xeon 6990E+ chips in a server for a total of 576 cores. A new software tool called Application Energy Telemetry, or AET, makes it possible to monitor the power usage of the workloads that run on the CPUs. Customers can use that information to optimize their applications.

“As AI becomes more agentic, the constraints shift to orchestration, concurrency, and data movement,” said Kevork Kechichan, executive vice president and general manager of Intel’s Data Center Group. “That shift reinforces a core reality: The CPU remains the control plane for the modern AI infrastructure.”

Intel announced Xeon 6+ alongside its next data center GPU series, which is codenamed Crescent Island. The chips in the series will ship as part of air-cooled accelerator cards that can be attached to servers via a PCIe port.

Many data center GPUs feature HBM memory, a RAM variety that is pricier and significantly faster than the kind used in consumer devices. Crescent Island chips, by contrast, will include a type of memory called LPDDR5x that is most commonly found in smartphones. It trades off some of HBM’s speed for a significant reduction in power use.

Crescent Island chips will feature up to 480 gigabytes of LPDDR5x and 350-watt power footprint. The GPUs’ cores will be based on a new architecture, Xe 3P, that Intel says is more performant than its current design. The architecture will enable AI models to store data in multiple floating point formats including MXFP4, which helps reduce the memory requirements of language models.

Xeon 6+ and Crescent Island are rolling out alongside several other additions to Intel’s chip portfolio. 

The company today introduced a 12-core CPU called Xeon 6300 processor that is designed for entry-level servers. In addition, Intel debuted a line of Ethernet controllers and adapters called Ethernet E835 that can be used to connect servers to data center networks. The chipmaker says that the most advanced device in the product series provides up to 1.9 times better performance per watt than comparable Nvidia Corp. silicon.

Photo: Thomas Cloer/Flickr