Summary
NetworkChuck tests WiCi, a sponsored early-engineering-model device that shares an NVIDIA RTX 5090 across multiple devices over Wi-Fi, making the GPU appear as a local card on a laptop. The video puts it through gaming with Cyberpunk, Blender rendering, image and video generation, and large-model local inference.
A major part of the video explains why this is hard. A GPU on PCIe offers roughly 512 gigabits per second with sub-microsecond response, while Wi-Fi manages about one gigabit per second at around 1.5 milliseconds, a gap that matters because games and AI workloads exchange instructions thousands of times per second.
The video then tackles the memory problem behind running very large models locally. It compares VRAM at about 1.8 terabytes per second, DDR5 system RAM at roughly 90 gigabytes per second, and SSDs at about 14 gigabytes per second, and works through why running a roughly 750-billion-parameter DeepSeek model on a 32 GB card would normally be impractical. It then shows how the WiCi approach tries to work around those limits.
Viewers interested in local AI, home labs, and the practical limits of GPU memory and networking will find a clear walkthrough of the tradeoffs.
📺 Source: NetworkChuck · Published October 30, 2026
🏷️ Format: Review
