Qwen on AMD Strix Halo

Can I run Qwen3 8B on an Ryzen AI Max+ 395 128GB?

Yes. Qwen3 8B at Q4_K_M uses 6.4 GB of the 96.0 GB available on a Ryzen AI Max+ 395 128GB, and runs at about 25.3 tokens per second. You can push the context to 128K.

Runs comfortably

6.4 GB of 96.0 GB · 7%
096 GB
Weights 4.7 GB
KV cache 1.1 GB
Runtime overhead 0.5 GB

Qwen3 8B at Q4_K_M leaves 89.6 GB spare on a Ryzen AI Max+ 395 128GB. There is room to raise the context length or move up a quantisation level.

Generation25.3tok/s
Prompt processing1282tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of Qwen3 8B 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 15.3 GB 16.9 GB Runs comfortably 128K 8.49
INT8 / W8A8 8.50 7.8 GB 9.5 GB Runs comfortably 128K 16.0
Q8_0 (GGUF) 8.50 7.8 GB 9.5 GB Runs comfortably 128K 16.0
FP8 (E4M3) 8.00 7.6 GB 9.3 GB Runs comfortably 128K 16.4
Q6_K 6.56 6.3 GB 7.9 GB Runs comfortably 128K 19.6
AWQ 4-bit 4.25 5.8 GB 7.4 GB Runs comfortably 128K 21.2
GPTQ 4-bit 4.25 5.8 GB 7.4 GB Runs comfortably 128K 21.2
MXFP4 4.25 5.8 GB 7.4 GB Runs comfortably 128K 21.2
Q5_K_M 5.67 5.4 GB 7.1 GB Runs comfortably 128K 22.4
Q5_K_S 5.52 5.3 GB 6.9 GB Runs comfortably 128K 23.0
Q4_K_M 4.85 4.7 GB 6.4 GB Runs comfortably 128K 25.3
Q4_K_S 4.58 4.5 GB 6.1 GB Runs comfortably 128K 26.4
Q4_0 4.55 4.5 GB 6.1 GB Runs comfortably 128K 26.5
IQ4_XS 4.25 4.2 GB 5.9 GB Runs comfortably 128K 27.9
Q3_K_M 3.91 4.0 GB 5.6 GB Runs comfortably 128K 29.6
IQ3_M 3.70 3.8 GB 5.4 GB Runs comfortably 128K 30.7
IQ3_XXS 3.06 3.3 GB 4.9 GB Runs comfortably 128K 34.8
Q2_K 2.63 2.9 GB 4.6 GB Runs comfortably 128K 38.3
IQ2_XXS 2.06 2.5 GB 4.1 GB Runs comfortably 128K 44.0
IQ1_M 1.75 2.2 GB 3.9 GB Runs comfortably 128K 48.0

Also worth checking

Qwen3 8B 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