Tag
PyTorch is reportedly adding a fast Thunderbolt communication backend for distributed model training on Macs.
Fabien Sanglard shares his experience with a dock that finally wakes up reliably after switching to an ASUS ROG Swift monitor, solving a longstanding Thunderbolt wake-from-sleep frustration.
Experimental implementation of RDMA over USB4 demonstrated using Thunderbolt and Strix Halo, potentially enabling high-speed data transfer with any USB4 host.
Mozilla's for-profit subsidiary MZLA Technologies launched Thunderbolt AI, an open-source, self-hosted AI client that runs entirely on user infrastructure, supports any model, and ensures data privacy by keeping data inside the organization's own systems.
A user showcases a DIY cluster of M5 Max MacBooks connected via Thunderbolt 5, highlighting the aggregate compute power and connectivity challenges.
MLX's RDMA-over-Thunderbolt implementation for macOS is now available as a standalone library, enabling high-speed Mac clusters for local AI workloads.
A demo shows Qwen3.6 35B vision model running across two M5 Max MacBook Pros connected via RDMA over Thunderbolt 5, achieving near-instant responses with prefix caching. The model correctly identifies Apple Park but misidentifies a person in the image.