Intel GPU D-Series
Intel's Xe3P inference accelerator (formerly Crescent Island): up to 480GB LPDDR5X, 1.3 PFLOPS FP8, 350W air-cooled PCIe.
This entry is marked for review. Check the original sources before relying on its details.
Overview
Intel named its Crescent Island inference GPU the GPU D-Series at the OCP Global Summit in mid-October 2026 (sources date it October 12 or 13). It is an Xe3P-based PCIe card with 1.5 TB/s memory bandwidth and 1.3 petaflops of FP8 compute at 350W, air-cooled. Intel's own branded card carries 160GB of LPDDR5X while partner-built cards reach 480GB; Intel says one card hosts models up to 300 billion parameters and eight cards scale past one trillion. Customer sampling has started; pricing and general availability are undisclosed, and the FP8 figure is a theoretical peak.
This is a sourced directory entry. Features and capabilities are reported in the linked sources; this entry does not document independent Aymorphic testing.
Key features
- Xe3P architecture, up to 32 Xe cores per press reports
- Up to 480GB LPDDR5X on partner cards; 160GB on Intel's branded card; 1.5 TB/s bandwidth
- 1.3 PFLOPS FP8 theoretical peak at 350W in an air-cooled PCIe form factor
- Single card hosts models up to 300B parameters; eight-card systems scale past 1T
- Customer sampling underway
Use cases
- Large-model inference where memory capacity matters more than peak bandwidth
- Air-cooled enterprise and edge data-center inference racks
Pricing
unknown; Intel has not disclosed pricing or general availability.
Pricing reflects the source record and may change. Confirm current terms with the provider.
Dates and sources
- Released
- 12 Oct 2026
- Record updated
- 23 Oct 2026
- Record last checked
- 23 Oct 2026