OpenAI on AMD Strix Halo

Can I run gpt-oss 120B-A5.1B on an Ryzen AI Max+ 395 128GB?

Yes. gpt-oss 120B-A5.1B at Q4_K_M uses 67.1 GB of the 96.0 GB available on a Ryzen AI Max+ 395 128GB, and runs at about 34.6 tokens per second. You can push the context to 128K.

Runs comfortably

67.1 GB of 96.0 GB · 70%
096 GB
Weights 66.0 GB
KV cache 0.6 GB
Runtime overhead 0.5 GB

gpt-oss 120B-A5.1B at Q4_K_M leaves 28.9 GB spare on a Ryzen AI Max+ 395 128GB. There is room to raise the context length or move up a quantisation level.

Generation34.6tok/s
Prompt processing2059tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of gpt-oss 120B-A5.1B 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 16.00 217.6 GB 218.6 GB Won't fit — 4.83
INT8 / W8A8 8.50 115.3 GB 116.4 GB Won't fit — 14.4
FP8 (E4M3) 8.00 108.8 GB 109.8 GB Won't fit — 16.7
Q6_K 6.56 89.2 GB 90.3 GB Fits, but tight 90K 26.6
Q5_K_M 5.67 77.1 GB 78.2 GB Runs comfortably 128K 30.2
Q5_K_S 5.52 75.1 GB 76.1 GB Runs comfortably 128K 31.0
Q4_K_M 4.85 66.0 GB 67.1 GB Runs comfortably 128K 34.6
Q4_K_S 4.58 62.4 GB 63.5 GB Runs comfortably 128K 36.3
Q4_0 4.55 62.0 GB 63.1 GB Runs comfortably 128K 36.5
AWQ 4-bit 4.25 59.4 GB 60.4 GB Runs comfortably 128K 37.9
GPTQ 4-bit 4.25 59.4 GB 60.4 GB Runs comfortably 128K 37.9
MXFP4 4.25 59.4 GB 60.4 GB Runs comfortably 128K 37.9
IQ4_XS 4.25 58.0 GB 59.0 GB Runs comfortably 128K 38.7
Q3_K_M 3.91 53.4 GB 54.4 GB Runs comfortably 128K 41.5
IQ3_M 3.70 50.6 GB 51.6 GB Runs comfortably 128K 43.4
IQ3_XXS 3.06 41.9 GB 43.0 GB Runs comfortably 128K 50.5
Q2_K 2.63 36.1 GB 37.2 GB Runs comfortably 128K 56.8
IQ2_XXS 2.06 28.5 GB 29.5 GB Runs comfortably 128K 68.0
IQ1_M 1.75 24.3 GB 25.4 GB Runs comfortably 128K 76.1

Also worth checking

gpt-oss 120B-A5.1B 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