# [For 780M] Vulkan inference 12.9% faster: MMVQ tuning for AMD 780M/Strix in llama.cpp, Q4_0 ROCmFP4 models arriving one after another

Digest AI · Hardware & Compute · published 2026-09-26T04:34:00Z

Canonical: https://digestai.news/story/for-780m-vulkan-inference-12-9-faster-mmvq-tuning-for-amd-780m-strix-i

## Summary

By setting load‑mode to auto, the build now skips memory‑mapped I/O on iGPUs, preventing the model from being loaded twice and halving usable memory.

New ROCmFP4 quantized models for Strix/Halo are appearing on HuggingFace, including Laguna‑S 2.1 variants and large‑scale MoE models. A known bug with Sliding Window Attention and the preserve‑thinking flag doubles generation time; users should avoid that combination.

## Why it matters

Developers can adopt the changes immediately, reducing memory errors and improving throughput for applications that rely on Vulkan-based inference.

## Sources

1. [For 780M Vulkan inference 12.9% faster: MMVQ tuning for AMD 780M/Strix in llama.cpp, Q4_0 ROCmFP4 models arriving one after another](https://note.com/samehadaonsen/n/n43d8cbad039a?hl=en) (note.com, 2026-09-26)

Part of the developing story: [Open-Source AI Surge Local Models Faster Inference](https://digestai.news/thread/kdnuggets-lists-seven-open-source-chatgpt-alternatives-that-run-locally) (4 stories)

## Cite

Digest AI, "[For 780M] Vulkan inference 12.9% faster: MMVQ tuning for AMD 780M/Strix in llama.cpp, Q4_0 ROCmFP4 models arriving one after another", 26 September 2026, https://digestai.news/story/for-780m-vulkan-inference-12-9-faster-mmvq-tuning-for-amd-780m-strix-i

---

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/for-780m-vulkan-inference-12-9-faster-mmvq-tuning-for-amd-780m-strix-i.json