Intel Details Crescent Island GPU and New Xeon Rack Designs

Intel said on 12 October 2026, at the Open Compute Project (OCP) Global Summit, that its Intel GPU D-series, formerly code-named Crescent Island, combines up to 480GB of LPDDR5X memory, 1.5TB/s of memory bandwidth and 1.3 petaflops of FP8 compute within a 350-watt power envelope, and that customer sampling is underway.

It also showed new Xeon-based rack designs with partners, and says select Xeon 6+ configurations scale beyond 46,000 cores per rack at 133.45 kW of total rack power.

What Intel Said About GPU D-Series

Intel says the GPU D-series is designed for AI inference and built on its Xe3P architecture, to improve the economics of sustained, memory-intensive inference.

According to Intel, it supports models with up to 300 billion parameters on a single card and more than one trillion parameters in an eight-GPU system. Intel names target uses from multi-tenant AI services to robotics and other latency-sensitive workloads.

The release’s footnote adds a detail on memory: the Intel branded PCIe card has 160GB of LPDDR5X, and the design lets partners build their own branded cards with options up to 480GB.

Intel says the card uses a standard PCIe form factor and an air-cooled design, so customers can scale inference without specialized rack or cooling infrastructure.

Anil Nanduri, vice president, AI Products and GTM at Intel’s Data Center Group, said: “Inference is becoming the defining AI workload, and enterprises need infrastructure that delivers the right performance at the right cost.”

The Xeon Rack Designs

Intel says it showcased Xeon-based rack designs developed with ecosystem partners, many of them built on OCP’s Open Rack v3 (ORv3) standard.

In Intel’s description, agentic workflows place more demand on general-purpose compute to orchestrate workloads, process data, manage I/O and execute tasks that keep accelerators productive. It says performance is increasingly measured by the number of concurrent agents a rack can sustain at target latency.

Nicolas Dubé, corporate vice president of Intel’s Data Center Systems and Solutions Group, said: “Open models are expanding the options enterprises have for deploying AI.”

Software and OCP Contributions

Intel says its open software strategy relies on native upstream integration with industry-standard AI frameworks, inference runtimes and deployment environments across CPUs and GPUs, and that GPU D-series shares a common Xe software foundation with other Intel GPUs.

It lists Day 0 support for Meta’s Muse Glimmer among recent model work, and says its AI for Enterprise Agent Toolkit is deployable on Xeon processors today.

Intel says it has made 116 contributions to OCP since 2011. This year it is showcasing its contribution to the Open Adapter Card workgroup, which defines a standardized sub-rack accelerator card form factor.

Where This Sits This Week

Intel’s details came on the same OCP day as other rack and chip news we covered: NVIDIA’s open Vera Rubin rack designs, AMD’s Helion agent rack blueprint, Marvell’s new CXL memory expanders, and Microsoft’s Maia 300 accelerator.

The Structural Read

As we read it, the GPU D-series pitch is about where the card fits: Intel says it uses a standard PCIe form factor and air cooling, so customers can scale inference without specialized rack or cooling infrastructure.

As we read it, the memory detail matters most: Intel puts LPDDR5X on the card, up to 480GB on partner cards and 160GB on its own, and says it supports models with up to 300 billion parameters on one card.

As we read it, the Xeon racks carry Intel’s argument for CPUs in agentic AI: Intel says agentic workflows place more demand on general-purpose compute to orchestrate workloads, process data and execute tasks.

Three Implications

ENTERPRISE DATA CENTERS Intel says the GPU D-series is air-cooled in a standard PCIe form factor, designed to simplify deployment within existing data center environments.

SYSTEM BUILDERS Intel says partners can build their own branded cards with memory options up to 480GB, and that many of the showcased Xeon racks use OCP’s Open Rack v3 standard.

MODEL DEVELOPERS Intel says it supports emerging open models on GPU D-series and Xeon, listing Day 0 support for Meta’s Muse Glimmer.

The Business Engineer Lens

This story maps onto the Business Engineer framework The State of the Inference Economy.

The framework puts it this way: “Inference is a revenue engine: continuous, distributed across millions of applications and agents, incurring marginal cost with every query.”

As we read it, Intel is pitching the GPU D-series at that marginal cost: it says the card is designed to improve the economics of sustained, memory-intensive inference within a 350-watt envelope.

What Is Not Established

Intel gives no price or general availability date for GPU D-series, and names no customer; it says customer sampling is underway. The 46,000-core figure applies to select Xeon 6+ configurations, and the rack-level claims are Intel’s own. The 480GB figure is the most a partner card can carry; Intel’s own card has 160GB.

The Bottom Line

At OCP, Intel put numbers on its GPU D-series inference card and showed Xeon rack designs, many of them built on OCP’s Open Rack v3 standard. The specifications and claims are Intel’s own, and the GPU is at the customer sampling stage.