The Pulse
Intel’s D-Series GPU Can Reach 480GB in Partner Cards
Intel detailed its D-series inference GPU at the OCP Global Summit on October 13, describing designs with up to 480GB of LPDDR5X memory. The Intel-branded PCIe card has 160GB; partner-built cards can offer configurations up to 480GB.
AI.info Team · Written with AI ·
Intel’s D-series inference GPU can be configured with up to 480GB of LPDDR5X memory—but that capacity does not belong to the Intel-branded card. The company’s October 13 announcement says its own PCIe card has 160GB, while original-design manufacturers can build partner-branded versions with options reaching 480GB. The distinction puts a qualification on the largest number in Intel’s pitch for agentic AI infrastructure.
480GB is a partner-built option
Intel shared new technical details at the Open Compute Project Global Summit for the GPU, formerly code-named Crescent Island and now called the Intel GPU D-series. Intel says the design combines up to 480GB of LPDDR5X memory, 1.5 terabytes per second of memory bandwidth and 1.3 petaflops of FP8 compute within a 350-watt power envelope. Its footnote clarifies that the Intel-branded PCIe card carries 160GB; the higher-capacity designs are for partners to build under their own brands.
That matters because memory capacity is a practical limit for inference workloads: models and their working context must fit within available memory, or be split across accelerators. Intel says one card can support models with up to 300 billion parameters, and an eight-GPU system can support more than one trillion parameters. The announcement does not specify the model precision or other conditions behind those capacity claims.
Intel targets long-context and concurrent inference
The D-series is designed for large language models, long-context workloads and concurrent inference services, with potential applications including multi-tenant AI services and robotics. Intel says the card uses a standard PCIe form factor, air cooling and a shared Xe software foundation, so it can fit into existing data-center environments without specialized rack or cooling infrastructure.
“Inference is becoming the defining AI workload, and enterprises need infrastructure that delivers the right performance at the right cost,” said Anil Nanduri, Intel’s vice president of AI Products and go-to-market in the Data Center Group. He said the company is focusing on making high-capacity inference practical to deploy and economical to scale.
Xeon and software fill out Intel’s pitch
Intel framed the GPU as one part of a system that pairs accelerators with Xeon CPUs. The company argues that agentic workflows—retrieval, reasoning, orchestration, tool use and execution—also require general-purpose processors to move data and run tasks around the accelerator. At the summit, Intel showed Xeon-based rack designs; it said select Xeon 6+ configurations exceed 46,000 cores per rack at 133.45 kilowatts of total rack power.
The software pitch centers on compatibility with existing AI frameworks and inference runtimes, plus portability across Intel GPU platforms. Intel also said its AI for Enterprise Agent Toolkit is deployable on Xeon processors today, while its provider-based architecture is designed to extend to Intel GPUs. Those are company-described capabilities; the announcement did not provide independent benchmarks for the D-series’ performance or economics.
Sampling has begun, but pricing is absent
Intel says customer sampling is underway and that it is working with original equipment manufacturers, original-design manufacturers, neoclouds, software vendors and AI-focused companies on validated platforms and deployment-ready systems. The announcement gives no price or broader availability date. For now, the headline 480GB capacity describes the ceiling for partner-built cards—not the memory in Intel’s own 160GB-branded model.