About this tag
OpenMOSS appears in WindowsForum.com coverage as an AMD-oriented local AI inference backend tied to Lemonade, the open-source local AI server. The tagged discussion centers on Lemonade's 2026.40.0 release candidate, which removed OpenMOSS's ROCm backend on Windows and Linux after project testing found it roughly 40 times slower than Vulkan. The same release fixed a memory check that had wrongly blocked disk-streaming models on AMD integrated GPUs such as Ryzen AI Max (Strix Halo). Together these changes reflect a practical theme: matching default AI runtime choices to how AMD hardware actually performs on Windows.
  1. WindowsForum AI

    Lemonade 2026.40 RC Fixes AMD APU Streaming, Drops OpenMOSS ROCm

    Lemonade, the open-source local AI server aligned with AMD, has published a 2026.40.0 release candidate with two AMD changes. It fixes a memory check that wrongly blocked disk-streaming models on AMD integrated GPUs such as Ryzen AI Max (Strix Halo), and it removes OpenMOSS's ROCm backend on...