macOS 26.2 enables fast AI clusters with RDMA over Thunderbolt (developer.apple.com)

Article summary

The article discusses the new macOS 26.2 update, which enables fast AI clusters with RDMA over Thunderbolt, allowing for faster tensor parallelism in MLX. This update is expected to improve the performance of AI models on Macs. However, the article's details are not available, and the discussion revolves around the implications of this update and related topics. The update is seen as beneficial for AI model training and decoding, but its limitations and potential issues are also discussed.

Main themes

• macOS update
• AI clusters
• Tensor parallelism
• MLX performance
• HDR support
• Linux and BSD support

What commenters say

• The new macOS update enables fast AI clusters with RDMA over Thunderbolt, which can significantly improve AI model training and decoding performance.
• The update's benefits are limited by the lack of proper HDR support on non-Apple monitors, which can result in washed-out colors and grey blacks.
• Some argue that the update's focus on AI performance is more important than gaming features like HDR support, while others prioritize the latter.
• The use of tensor parallelism in the update can lead to significant speedups in AI model training, but it also introduces latency challenges.
• The lack of official Linux and BSD support is a major limitation for large-scale deployments of the update.
• Some commenters believe that Apple's closed ecosystem and limited hardware support will hinder the adoption of the update, while others see potential benefits in the update's performance advantages.
• There are differing opinions on the importance of CUDA compatibility and the potential for Apple to adopt industry-standard software and hardware support.
• The update's potential to enable new use cases, such as speculative decoding and distributed model training, is seen as a significant advantage by some commenters.