Intel Introduced its New Graphics Card with 480 GB Memory

Intel broke a significant capacity record in data center hardware with the new Intel GPU D-series, developed specifically for artificial intelligence inference processes. This new hardware, previously known as Crescent Island, took over the title of the data center GPU with the highest memory in the industry with 480GB LPDDR5X memory capacity from Nvidia's Blackwell Ultra B300 model.

In this new accelerator, the company chose to use LPDDR5X memory technology, which is preferred in laptops and energy-saving devices, instead of GDDR7, which is widely used in modern gaming cards. This strategic choice makes it clear that Intel is focusing on high memory capacity and energy efficiency, which are critical for AI inference, rather than gaming performance.

High Memory Capacity and Performance

Intel GPU D-series can support large language models with up to 300 billion parameters on a single card. When a system with eight GPUs is installed, this capacity exceeds one trillion parameters, allowing the processing of complex artificial intelligence models.

Among the technical specifications of the hardware, the 1.3 petaflops performance value for FP8 calculations stands out. In addition, 1.5TB/s memory bandwidth supports the overall speed of the system by allowing data to be moved quickly between the memory and the processor. These capabilities are designed for high-performance applications such as long scripts, concurrent inference requests, and enterprise AI services.

Intel states that this hardware provides a great advantage, especially in cases where artificial intelligence systems need to keep large model weights and related data accessible during inference. However, the company emphasizes that memory capacity alone is not sufficient and performance also depends on factors such as bandwidth, computing resources and software optimizations.

Easy Integration for Data Centers

Intel designed its new accelerator to easily fit into existing data centers. 350W power consumption and standard PCIe form factor allow installation without the need for special cooling systems or completely new rack infrastructures.

The company plans to optimize these GPUs to work with its own Xeon processors. Thanks to this integration, CPUs increase the efficiency of the system by coordinating data movement and workflows while managing tasks that do not require GPU acceleration. Intel states that this approach is ideal for multi-stage artificial intelligence systems such as retrieval, reasoning, orchestration and the use of external tools.

Intel states that AI inference has become a defining workload and businesses need infrastructures that deliver the right performance at the right cost. No official statement has been made yet about the general sales date or price of the product. What impact do you think this high-capacity GPU solution from Intel will have on the artificial intelligence infrastructure market?