Intel's Crescent Island AI inference accelerator will ship as the GPU D-Series family, promising 1.3 PFLOPS of FP8 compute and up to 480 GB of memory.

Intel has given Crescent Island a proper name: the Intel GPU D-Series. Announced at OCP 2026, it marks the first time the company has shared a theoretical peak performance target for the chip, 1.3 PFLOPS of FP8. Intel also disclosed the final memory bandwidth to expect from these AI inference accelerators: 1.5 TB/s. The new details add to the steady stream of information Intel has released about Crescent Island throughout 2026.

Crescent Island is the first chip built on Intel's Xe3P graphics architecture, though the chip lacks most or all of the graphics-specific functional units that would characterize a true GPU, so it is better described as a massively parallel computing architecture. Intel will release GPU D-Series products in an air-cooled PCIe form factor with a 350W TDP and up to 480GB of LPDDR5X memory.

For comparison, each of the 72 Rubin GPUs in Nvidia's Vera Rubin NVL72 rack-scale system can deliver a theoretical 17.5 PFLOPS of FP8 throughput. Those chips require liquid cooling and likely have a TGP measured in kilowatts. The GPU D-Series is clearly targeting a different performance envelope and infrastructure requirements than Rubin.

Intel GPU D-Series products are designed for traditional air-cooled rack servers scaling up to eight GPUs, similar to Nvidia's HGX systems. In such a configuration, Intel says the GPU D-Series could power inference on models with more than one trillion parameters. The company also highlights the chip's suitability for "large language models, long-context inference, and concurrent inference services," creating a foundation for "multi-tenant AI services, robotics, and other latency-sensitive workloads."

Distributed inference of this kind, in which multiple GPUs presumably act as one, will depend on a robust software stack. Intel is pointing to the "common Xe software foundation" it has already built for AI across its Xe Cores in integrated GPUs and discrete GPU products.

To bring Xe accelerators into the data center, Intel says it is working with "a broad ecosystem of OEMs, ODMs, neoclouds, independent software vendors, appliance providers, and AI-native companies." Customer sampling of GPU D-Series products is in progress, though Intel has not provided details of a broad launch.

It remains to be seen whether the GPU D-Series will give Intel a foothold in AI accelerators after the failure of the Data Center GPU Max family (code-named Ponte Vecchio) and its mothballed follow-on products, as well as a series of silicon acquisitions that did not coalesce into a coherent AI strategy. Still, in a market hungry for compute for inference, a product with a proven architectural foundation, targeted specifically at that workload, and deployable in an accessible thermal and electrical envelope could be the recipe Intel needs to gain traction as AI models advance and agentic workflows grow in importance.