Qwen on AMD Strix Halo

Can I run Qwen3 4B on an Ryzen AI Max+ 395 128GB?

Yes. Qwen3 4B at Q4_K_M uses 3.9 GB of the 96.0 GB available on a Ryzen AI Max+ 395 128GB, and runs at about 46.5 tokens per second. You can push the context to 128K.

Runs comfortably

3.9 GB of 96.0 GB · 4%
096 GB
Weights 2.3 GB
KV cache 1.1 GB
Runtime overhead 0.5 GB

Qwen3 4B at Q4_K_M leaves 92.1 GB spare on a Ryzen AI Max+ 395 128GB. There is room to raise the context length or move up a quantisation level.

Generation46.5tok/s
Prompt processing2612tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of Qwen3 4B 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 7.5 GB 9.1 GB Runs comfortably 128K 16.6
INT8 / W8A8 8.50 3.9 GB 5.5 GB Runs comfortably 128K 30.0
Q8_0 (GGUF) 8.50 3.9 GB 5.5 GB Runs comfortably 128K 30.0
FP8 (E4M3) 8.00 3.7 GB 5.4 GB Runs comfortably 128K 31.0
Q6_K 6.56 3.1 GB 4.7 GB Runs comfortably 128K 36.7
Q5_K_M 5.67 2.7 GB 4.3 GB Runs comfortably 128K 41.4
Q5_K_S 5.52 2.6 GB 4.2 GB Runs comfortably 128K 42.4
AWQ 4-bit 4.25 2.5 GB 4.1 GB Runs comfortably 128K 43.2
GPTQ 4-bit 4.25 2.5 GB 4.1 GB Runs comfortably 128K 43.2
MXFP4 4.25 2.5 GB 4.1 GB Runs comfortably 128K 43.2
Q4_K_M 4.85 2.3 GB 3.9 GB Runs comfortably 128K 46.5
Q4_K_S 4.58 2.2 GB 3.8 GB Runs comfortably 128K 48.4
Q4_0 4.55 2.2 GB 3.8 GB Runs comfortably 128K 48.7
IQ4_XS 4.25 2.0 GB 3.7 GB Runs comfortably 128K 51.0
Q3_K_M 3.91 1.9 GB 3.5 GB Runs comfortably 128K 53.9
IQ3_M 3.70 1.8 GB 3.4 GB Runs comfortably 128K 55.9
IQ3_XXS 3.06 1.5 GB 3.2 GB Runs comfortably 128K 63.0
Q2_K 2.63 1.4 GB 3.0 GB Runs comfortably 128K 68.9
IQ2_XXS 2.06 1.1 GB 2.7 GB Runs comfortably 128K 78.6
IQ1_M 1.75 1.0 GB 2.6 GB Runs comfortably 128K 85.1

Also worth checking

Qwen3 4B on every GPU
Full hardware tableEverything that runs on a Ryzen AI Max+ 395 128GB
Full model tableWhy the KV cache is the part that bites
Guide