# Lemonade fixes AMD APU model streaming and drops OpenMOSS ROCm backend

Digest AI · Agents & Tools · published 2026-09-23T20:16:51Z

Canonical: https://digestai.news/story/lemonade-fixes-amd-apu-model-streaming-and-drops-openmoss-rocm-backend

## Summary

The Lemonade open‑source SDK released version 2026.39.1 and a 2026.40 release candidate on Tuesday. The new 2026.40 RC changes how streaming models are handled on AMD integrated GPUs by allocating memory from the APU GTT pool instead of a fixed vRAM carve‑out. This adjustment lets models such as DeepSeek‑V4‑Flash‑IQ2XXS‑DS4 run on AMD Ryzen AI Max (Strix Halo) where they previously failed.

In the same update the OpenMOSS back‑end for ROCm was removed on both Windows and Linux because its performance was reported to be around 40x slower than the Vulkan back‑end on identical hardware. The merge request also added a launch agent for JetBrains’ Junie and carried forward other enhancements like configurable vRAM auto‑eviction introduced in 2026.39.1. The changes are documented on GitHub.

## Key points

- Lemonade 2026.40 RC adds APU GTT‑based streaming support for AMD integrated GPUs.
- OpenMOSS ROCm backend removed because it runs around 40x slower than Vulkan on the same hardware.
- New launch agent for JetBrains’ Junie added; 2026.39.1 introduced configurable vRAM auto‑eviction.

## Why it matters

Enabling streaming on AMD APUs expands local AI serving to more low‑power devices, while dropping a poorly performing ROCm backend avoids wasted resources.

## Sources

1. [Lemonade fixes AMD APU model streaming, drops OpenMOSS ROCm as ~40x slower than Vulkan](https://phoronix.com/news/Lemonade-2026.40-RC) (phoronix.com, 2026-09-23)

## Cite

Digest AI, "Lemonade fixes AMD APU model streaming and drops OpenMOSS ROCm backend", 23 September 2026, https://digestai.news/story/lemonade-fixes-amd-apu-model-streaming-and-drops-openmoss-rocm-backend

---

Written by Digest AI's editorial model from the linked sources; the sources are the record. Headlines, digests and key points are written by Digest AI and may be quoted with a link to the story page. Linked articles belong to their publishers. Terms: https://digestai.news/terms#reuse
JSON: https://digestai.news/story/lemonade-fixes-amd-apu-model-streaming-and-drops-openmoss-rocm-backend.json