Intel has shaken up the artificial intelligence hardware landscape by revealing a new data center graphics processor boasting an unprecedented 480GB of memory. Codenamed Crescent Island and formally designated as the Intel GPU D-series, the accelerator was showcased at the recent Open Compute Project Global Summit. The staggering memory capacity allows the new hardware to eclipse previous record-holders like Nvidia's Blackwell Ultra B300, which maxes out at 288GB of HBM3e memory.
Rather than relying on ultra-fast gaming-grade GDDR7, Intel opted for LPDDR5X technology, traditionally utilized in power-conscious consumer devices like laptops. This design choice underscores a strategic pivot toward maximizing memory capacity and energy efficiency specifically for AI inference tasks. According to Intel, a single D-series card can natively support large language models containing up to 300 billion parameters, while an eight-GPU cluster can scale to accommodate models exceeding one trillion parameters.
Under the hood, the accelerator delivers 1.3 petaflops of FP8 computation capability paired with 1.5TB/s of memory bandwidth, operating within a standard 350W power envelope. By utilizing air cooling and a conventional PCIe form factor, Intel designed the hardware to slot seamlessly into existing enterprise server racks without necessitating expensive infrastructure overhauls or exotic liquid cooling solutions. The company also intends for the accelerators to pair directly with its Xeon processors to optimize workflow coordination.
Enterprise adoption of open models continues to accelerate, driving demand for infrastructure that balances massive memory footprints with practical deployment costs. While Intel has outlined the technical architecture and power specifications, the company has not yet disclosed official retail pricing or a definitive general-availability timeline for the GPU D-series.
Intel Unveils 480GB AI GPU Powered By LPDDR5X Memory
Comments