Microsoft Brings Phi Silica AI Model to Nvidia RTX GPUs in Early Developer Preview
Summary
Microsoft is expanding its Phi Silica AI model to Nvidia RTX 30-series and newer GPUs with at least 6GB of VRAM in an early developer preview, bringing local AI capabilities to more Windows users beyond Copilot+ PCs, though GPU-based execution lacks some NPU-exclusive features like prompt compression and speculative decoding.
Key Points
- Microsoft is testing Phi Silica small language model support on Nvidia RTX 30-series or newer GPUs with at least 6GB of VRAM, expanding local AI capabilities beyond Copilot+ PCs with NPUs.
- Access is currently restricted to developers who have the Experimental Channel, Developer Mode, Windows App SDK 2.2.2-experimental9 or later, and up-to-date GPU drivers, making this a developer preview rather than a consumer feature.
- GPU-based Phi Silica execution lacks NPU-exclusive features like prompt compression and speculative decoding, meaning Nvidia RTX users do not achieve full Copilot+ PC parity, with AMD GPU support still listed as coming later.