Home / Technology / Intel GPU Shatters Memory Record for AI
Intel GPU Shatters Memory Record for AI
19 Oct
Summary
- Intel's new GPU boasts 480GB of LPDDR5X memory.
- It surpasses Nvidia's Blackwell Ultra B300 in capacity.
- The D-series supports AI models up to 300 billion parameters.
Intel has unveiled its D-series GPU, featuring an unprecedented 480GB of LPDDR5X memory. This significant capacity aims to power complex AI inference tasks and positions the D-series as the new leader in data center GPU memory, surpassing Nvidia's Blackwell Ultra B300.
The Intel D-series prioritizes memory capacity and power efficiency over traditional gaming graphics card design. It supports artificial intelligence models with up to 300 billion parameters on a single card, and systems with eight GPUs can handle models exceeding one trillion parameters. This capability is crucial for enterprise AI services and robotics.
Designed for seamless integration into existing data centers, the D-series operates within a 350W power envelope and uses a standard PCIe form factor with air cooling. Intel also plans to integrate these accelerators with its Xeon processors for coordinated task management. The company has not yet announced pricing or a specific release date.
