Intel Unveils 480GB GPU D-Series for High-Capacity AI Inference

Intel has introduced its new GPU D-series, an AI accelerator featuring a massive 480GB of LPDDR5X memory, surpassing the capacity of current industry leaders like Nvidia's Blackwell B300. Designed specifically for demanding AI inference workloads, the D-series can support models with up to 300 billion parameters on a single card, or over one trillion parameters in an eight-GPU configuration. By utilizing LPDDR5X instead of GDDR7, Intel aims to prioritize memory density and power efficiency, allowing the hardware to fit into standard 350W PCIe data center slots without requiring specialized cooling. The GPU offers 1.3 petaflops of FP8 performance and 1.5TB/s of memory bandwidth. Intel intends for this hardware to simplify the deployment of large-scale AI models in enterprise environments, providing a cost-effective solution for complex tasks like robotics, reasoning, and large-scale data orchestration. Pricing and availability details have not yet been disclosed.