Qwen on AMD Strix Halo

Can I run Qwen3 30B-A3B on an Ryzen AI Max+ 395 128GB?

Yes. Qwen3 30B-A3B at Q4_K_M uses 18.5 GB of the 96.0 GB available on a Ryzen AI Max+ 395 128GB, and runs at about 42.2 tokens per second. You can push the context to 128K.

Runs comfortably

18.5 GB of 96.0 GB · 19%
096 GB
Weights 17.3 GB
KV cache 0.8 GB
Runtime overhead 0.5 GB

Qwen3 30B-A3B at Q4_K_M leaves 77.5 GB spare on a Ryzen AI Max+ 395 128GB. There is room to raise the context length or move up a quantisation level.

Generation42.2tok/s
Prompt processing3144tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of Qwen3 30B-A3B on a Ryzen AI Max+ 395 128GB

Highlighted row is the highest quality that still fits at 8K context.

QuantisationbpwWeights TotalVerdictMax ctxtok/s
FP16 / BF16 16.00 56.9 GB 58.1 GB Runs comfortably 128K 16.4
INT8 / W8A8 8.50 30.1 GB 31.3 GB Runs comfortably 128K 28.0
Q8_0 (GGUF) 8.50 30.1 GB 31.3 GB Runs comfortably 128K 28.0
FP8 (E4M3) 8.00 28.4 GB 29.7 GB Runs comfortably 128K 29.2
Q6_K 6.56 23.3 GB 24.6 GB Runs comfortably 128K 34.0
Q5_K_M 5.67 20.2 GB 21.4 GB Runs comfortably 128K 37.9
Q5_K_S 5.52 19.6 GB 20.9 GB Runs comfortably 128K 38.6
Q4_K_M 4.85 17.3 GB 18.5 GB Runs comfortably 128K 42.2
Q4_K_S 4.58 16.3 GB 17.6 GB Runs comfortably 128K 43.9
Q4_0 4.55 16.2 GB 17.5 GB Runs comfortably 128K 44.1
AWQ 4-bit 4.25 16.0 GB 17.2 GB Runs comfortably 128K 44.6
GPTQ 4-bit 4.25 16.0 GB 17.2 GB Runs comfortably 128K 44.6
MXFP4 4.25 16.0 GB 17.2 GB Runs comfortably 128K 44.6
IQ4_XS 4.25 15.2 GB 16.4 GB Runs comfortably 128K 46.1
Q3_K_M 3.91 14.0 GB 15.2 GB Runs comfortably 128K 48.6
IQ3_M 3.70 13.3 GB 14.5 GB Runs comfortably 128K 50.3
IQ3_XXS 3.06 11.1 GB 12.3 GB Runs comfortably 128K 56.2
Q2_K 2.63 9.6 GB 10.8 GB Runs comfortably 128K 61.1
IQ2_XXS 2.06 7.6 GB 8.8 GB Runs comfortably 128K 69.0
IQ1_M 1.75 6.5 GB 7.7 GB Runs comfortably 128K 74.3

Also worth checking