Intel shared new technical details on Intel GPU D-series AI inference today at the Open Compute Project Global Summit. The chip, previously introduced as Crescent Island, is built for sustained inference work that strains memory, power, and system budgets.
Intel says the design pairs up to 480GB of LPDDR5X memory with 1.5TB/s of memory bandwidth and 1.3 petaflops of FP8 compute. For teams running large models, that combination points to fewer memory stalls and more room to keep data close to the GPU during long inference runs.
Xe3P memory and throughput
The Intel GPU D-series is built on Xe3P, which puts the emphasis on feeding inference workloads rather than chasing headline compute alone. That matters because agentic AI turns inference into a multi-step workflow spanning retrieval, reasoning, orchestration, tool use and execution.
Intel said open models are expanding the options enterprises have for deploying AI. “Open models are expanding the options enterprises have for deploying AI. To realize their full potential, organizations need infrastructure that can match different workloads with the right intelligence, right compute resources and right economics. That's what we mean by 'put every prompt in its place.”
Open models and rack economics
Intel said the design is intended to improve the economics of sustained, memory-intensive AI inference. That is the practical test here, because memory capacity and bandwidth can decide whether a system keeps up without forcing more hardware into the rack.
Intel also said it sees growing demand for more flexible, cost- and power-efficient infrastructure, while x86 CPUs remain central to agentic AI workflows. The company framed that need against power constraints and memory bottlenecks, which is why the 480GB figure and the 133.45 kW rack power number sit in the same story.
What Intel did not say
Intel has not given a deployment date for Intel GPU D-series AI inference. Buyers who want to plan around the hardware still have specs, but they do not yet have a release window to match them against procurement cycles.
