'Inference Is Becoming The Defining AI Workload': Intel GPU D-series Launched With The World’s Largest 480GB Capacity And Air-cooled 350W Design

Intel’s GPU D-series sets a new GPU memory record with 480GB of LPDDR5X
Intel chose LPDDR5X instead of GDDR7 for its high-capacity AI accelerator
The D-series supports models containing up to 300 billion parameters

At the recent Open Compute Project Global Summit, Intel unveiled a graphics processor with up to 480GB of memory, an unusually large capacity designed for demanding AI inference workloads.

The GPU, called Intel GPU D-series and previously known by its Crescent Island codename, also hosts specialized computing hardware.

The Intel D-series has now dethroned Nvidia‘s Blackwell Ultra B300, which offers 288GB of HBM3e memory, as the data center GPU with the highest memory capacity.

Intel prioritizes memory capacity

Rather than using GDDR7 memory associated with modern graphics cards, Intel has chosen LPDDR5X, a memory technology commonly found in laptops and other power-conscious devices.

The configuration suggests that Intel is prioritizing memory capacity and power efficiency for AI inference instead of designing a conventional gaming graphics card.

Intel says the accelerator can support large language models containing up to 300 billion parameters on one card, while an eight-GPU system can accommodate models exceeding one trillion parameters.

These capabilities are intended for lengthy prompts, simultaneous inference requests, enterprise AI services, and robotics applications requiring responsive processing.

“Inference is becoming the defining AI workload, and enterprises need infrastructure that delivers the right performance at the right cost,” said Anil Nanduri, vice president, AI Products & GTM, Intel Data Center Group.

“With Intel GPU D-series, we’re focused on making high-capacity inference practical to deploy and economical to scale.”

The 480GB capacity is particularly relevant when AI systems must keep substantial model weights and associated data accessible during inference.

However, memory capacity alone does not establish how quickly a GPU will serve requests, since performance also depends on bandwidth, computing resources, software, and workload characteristics.

Intel rates the D-series at 1.3 petaflops for FP8 computation, a numerical format commonly used to accelerate AI operations while reducing memory requirements compared with higher-precision formats.

Its 1.5TB/s memory bandwidth provides a separate measure of how quickly data can move between memory and the processor.

Intel wants the accelerator to fit existing data centers

The company specifies a 350W power envelope and a standard PCIe form factor, alongside air cooling and software compatibility across its GPU platforms.

These characteristics are intended to simplify deployment in existing data centers without requiring specialized cooling arrangements or entirely new rack infrastructure.

Intel also plans to connect the accelerator with its Xeon processors, allowing CPUs to coordinate tasks, manage data movement, and handle workloads that do not require GPU acceleration.

This approach addresses AI systems increasingly built around multiple stages, including retrieval, reasoning, orchestration, external tools, and execution.

“Open models are expanding the options enterprises have for deploying AI,” said Nicolas Dubé, corporate vice president, Data Center Systems and Solutions Group, Intel.

“To realize their full potential, organizations need infrastructure that can match different workloads with the right intelligence, right compute resources and right economics. That’s what we mean by ‘put every prompt in its place.”

As of the time of writing, Intel has not revealed the public retail price or a specific general-availability date for the GPU D-series.