Intel® GPU D270 │ Inference-Optimized Data Center GPU
Intel’s D-series pairs an inference-focused design with up to 480 GB of LPDDR5X memory in an air-cooled card.
TL;DR
- Intel describes the D270 as an inference-optimized data-center GPU with up to 480 GB of LPDDR5X memory.
- The product page lists a 350 W air-cooled design, 32 Xe3P cores and 256 XMX matrix engines.
- The specifications are vendor claims, and independent workload results and broad availability remain open.
Intel’s product page identifies the D270 as an inference-optimized data-center GPU in its D-series. The company lists up to 480 GB of LPDDR5X memory per card, alongside a 350 W air-cooled design. [1] [1]
Intel says the device contains 32 Xe3P cores and 256 XMX matrix engines, and frames it for long-context and agentic workloads. Those are manufacturer specifications and positioning, rather than independent measurements of performance or cost. [1] [1]
StorageReview reports that partner-built cards may carry up to 480 GB, while Intel’s own reference card is listed at 160 GB, and describes the product as targeting inference rather than graphics. This makes the memory configuration a central part of the launch, but does not establish how deployments will perform across model sizes or serving conditions. [2] [2]
The published specifications do not settle price, general availability or comparative throughput in production workloads. Intel’s stated positioning is clear; customer adoption and independent testing remain open. [1, 2] [1] [2]
Why it matters
The D270 is Intel’s attempt to compete in inference infrastructure through memory capacity and conventional air-cooled deployment. Whether that changes data-center economics depends on independent performance, price and availability evidence that is not yet established.
Editor's note
Headline follows Intel’s product-page title. Technical capabilities are vendor specifications; no independent benchmark or confirmed general-availability date was found.