A New Benchmark in Memory Capacity
During the recent Open Compute Project Global Summit, Intel officially announced the launch of its new D-series graphics processor. Designed specifically to handle the intensive demands of AI inference, the card boasts a massive 480GB of memory, setting a new industry standard for data center GPUs. This development notably surpasses the 288GB of HBM3e memory found in Nvidia's Blackwell Ultra B300.
Previously referred to by the codename "Crescent Island," the D-series integrates specialized hardware architectures tailored for AI tasks. By opting for LPDDR5X memory—a standard often reserved for laptops and energy-efficient devices—rather than the high-bandwidth GDDR7 used in consumer gaming cards, Intel has signaled a clear shift toward memory density and power economy.
Scalability and Large Language Models
The technical configuration of the D-series is optimized for massive AI models. Intel reports that a single card is capable of supporting large language models (LLMs) with as many as 300 billion parameters. Furthermore, by deploying an eight-GPU system, organizations can run models that exceed one trillion parameters. These capabilities are designed to assist with complex tasks, such as handling extensive prompts, managing multiple concurrent requests, and powering enterprise-grade robotics.
"Inference is becoming the defining AI workload, and enterprises need infrastructure that delivers the right performance at the right cost," noted Anil Nanduri, vice president of AI Products & GTM for Intel's Data Center Group. "With Intel GPU D-series, we're focused on making high-capacity inference practical to deploy and economical to scale."
Performance Metrics and Infrastructure Integration
While memory capacity is a primary feature, the D-series also balances raw processing power. The accelerator is rated at 1.3 petaflops for FP8 computation—a format frequently utilized to streamline AI operations—and offers a memory bandwidth of 1.5TB/s.
To ensure ease of adoption, Intel designed the hardware with a 350W power envelope and a standard PCIe form factor. These choices are intended to allow data centers to integrate the new hardware into existing infrastructure without the need for specialized cooling systems or total rack overhauls.
Strategic Synergy with Xeon Processors
The company also aims to pair the D-series with its Xeon processors. This CPU-GPU combination is intended to streamline data movement and task coordination, ensuring that workloads are appropriately divided between the processors that handle logic and those optimized for acceleration.
"Open models are expanding the options enterprises have for deploying AI," stated Nicolas Dubé, corporate vice president of the Data Center Systems and Solutions Group at 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.'"
At the time of this report, Intel has not yet disclosed pricing details or a firm release date for the general market.