Intel Unveils New Data Center GPU for Inference, Dubbed ‘Crescent Island’ October 16, 2025 by Alex Woodie 
Intel unveiled a new data center GPU at the OCP Global Summit this week. Dubbed “Crescent Island,” the GPU will utilize the Xe3P graphics architecture, low-power LPDDR5X memory, and will target AI inference workloads, with energy efficiency as a primary characteristic.
As the focus of the Gen AI revolution shifts from model training to inference and agentic AI, chipmakers have responded with new chip designs that are optimized for inference workloads. Instead of cranking out massive AI accelerators that have tons of number-crunching horsepower–and consume heaps of energy and produce gobs of heat that must be removed with fans or liquid cooling–chipmakers are looking to build processors that get the job done as efficiently and cost-effectively as possible.
That’s the backdrop for Intel’s latest GPU, Crescent Island, which is due in the second half of 2026. The new GPU will feature 160GB of LPDDR5X memory, utilize the Xe3P microarchitecture, and will be optimized for performance-per-watt, the company says. Xe3P is a new, performance-oriented version of the Xe3 architecture used in Intel’s Panther Lake CPUs.
“AI is shifting from static training to real-time, everywhere inference–driven by agentic AI,” said Sachin Katti, CTO of Intel. “Scaling these complex workloads requires heterogeneous systems that match the right silicon to the right task, powered by an open software stack. Intel’s Xe architecture data center GPU will provide the efficient headroom customers need —and more value—as token volumes surge.”
Intel’s Ponte Vecchio GPU, circa 2022
Intel launched its Intel Xe GPU microarchitecture initiative back in 2018, with details emerging in 2019 at its HPC Developer Conference (held down the street from the SC19 show in November 2019). The goal was to compete against Nvidia and AMD GPUs for both data center (HPC and AI) and desktop (gaming and graphics) use cases. It has launched a series of Xe (which stands for “exascale for everyone”) products over the years, including discrete GPUs for graphics, integrated GPUs embedded onto the CPU, and data center GPUs used for AI and HPC workloads.
Its first Intel Xe data center GPU was the Ponte Vecchio, which utilized the Xe-HPC microarchitecture and the Embedded Multi-Die Interconnect Bridge (EMIB) and Foveros die stacking packaging on an Intel 4 node, which was its 7-nanometer technology. Ponte Vecchio also used some 5 nm components from TSMC.
You will remember that Argonne National Laboratory’s Aurora supercomputer, which was the second fastest supercomputer ever built when it debuted two years ago, was built using six Ponte Vecchio Max Series GPUs alongside every one Intel Xeon Max Series CPU in an HPE Cray EX frame using an HPE Slingshot interconnect. Aurora featured a total of 63,744 of the Xe-HPC Ponte Vecchio GPUs across more than 10,000 nodes, delivering 585 petaflops in November 2023. It officially became the second supercomputer to break the exascale barrier in June 2024, and it currently sits in the number three slot on the Top500 list.
When Aurora was first revealed back in 2015, it was slated to pair Intel’s Xeon Phi accelerators alongside Xeon CPUs. However, when Intel killed Xeon Phi in 2017, it forced the computer’s designers to go back to the drawing board. The answer came when Intel announced Ponte Vecchio in 2019.
Intel’s new Crescent Lake GPU will feature LPDDR5X memory
It’s unclear exactly how Crescent Lake, which is the successor to Ponte Vecchio, will be configured, and whether it will be delivered as a pair of smaller GPUs or one massive GPU. The performance characteristics of Crescent Island will also be something to keep an eye on, particularly in terms of memory bandwidth, which is the sticking point in a lot of AI workloads these days.
The use of LVDDR5X memory, which is usually found in PCs and smartphones, is an interesting choice for a data center GPU. LVDDR5X was released in 2021 and can apparently reach speeds up to 14.4 Gbps per pin. Memory makers like Samsung and Micron offer LVVDR5X memory in capacities up to 32GB, so Intel will need to figure out a way to connect a handful of DIMMs to each GPU.
Both AMD and Nvidia are using large amounts of the latest generation of high bandwidth memory (HBM) in their next-gen GPUs due in 2026, with AMD MI450 offering up to 432GB of HBM4 and Nvidia using up to 1TB of HBM4 memory with its Rubin Ultra GPU.
HBM4 has advantages when it comes to bandwidth. But with rising prices for HBM4 and tighter supply chains, perhaps Intel is on to something by using LVDDR5X memory–particularly with power efficiency and cost being such big factors in AI success.
This article first appeared on our sister publication, HPCwire.
Related
Alex Woodie has written about IT as a technology journalist for more than a decade. He brings extensive experience from the IBM midrange marketplace, including topics such as servers, ERP applications, programming, databases, security, high availability, storage, business intelligence, cloud, and mobile enablement. He resides in the San Diego area.
