AMD’s Ryzen AI Software 1.9 release extends the context length available to large language models on its NPU, beyond the 4K/16K limits of previous releases. It also adds support for several models, giving developers more options to try on AMD’s AI PC platform, though some are explicitly not performance-optimised.
AMD Watch analysis
What happened
AMD’s release notes list text support for Gemma-4-E2B, Gemma-4-E4B and Hy-MT2-1.8B. The release also adds functional, but not performance-optimised, support for Phi-4-14B, Qwen3.5-9B, Granite-4.1-8B and Gemma-4-12B.
Dynamic LoRA support is added for the text models Qwen2.5-3B, Qwen3-4B and Qwen3.5-4B. AMD also lists CNN/Transformer bug fixes and general improvements, without specifying them in this release summary.
Why it matters
Longer context gives developers more room to test what an NPU can handle with language-model workloads. The wider model list is useful too, but “functional” support is not the same as a performance claim: AMD specifically says the four larger models are not optimised for performance.
That makes this a practical update for people building or testing local AI applications on Ryzen AI hardware, rather than evidence that every listed model will run quickly or suit every workload.
Our read
There is real substance here: a stated expansion beyond earlier context limits, new model options and dynamic LoRA support. Developers should check the compatibility details for their hardware and treat the non-optimised models as experiments, not speed promises. The release notes give the menu; the benchmark is still the meal.
What to watch
- Which Ryzen AI hardware configurations support the extended context length.
- How the newly supported models perform on real workloads.
- Whether AMD provides more detail on the listed fixes and improvements.
Discussion spark: For local AI on Ryzen hardware, would you prioritise longer context, support for more models, or better performance for the models already available?
Sources and evidence
- Release Note: Version 1.9.0 (31 October 2026, 01:28 UTC)
Independent WittyWires tracker for public updates about AMD. Not affiliated with or endorsed by AMD; this is not an official account.