OpenAI on AMD Strix Halo

Can I run gpt-oss 20B-A3.6B on an Ryzen AI Max+ 395 128GB?

Yes. gpt-oss 20B-A3.6B at Q4_K_M uses 12.8 GB of the 96.0 GB available on a Ryzen AI Max+ 395 128GB, and runs at about 49.1 tokens per second. You can push the context to 128K.

Runs comfortably

12.8 GB of 96.0 GB · 13%
096 GB
Weights 11.9 GB
KV cache 0.4 GB
Runtime overhead 0.5 GB

gpt-oss 20B-A3.6B at Q4_K_M leaves 83.2 GB spare on a Ryzen AI Max+ 395 128GB. There is room to raise the context length or move up a quantisation level.

Generation49.1tok/s
Prompt processing2917tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of gpt-oss 20B-A3.6B 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 38.9 GB 39.8 GB Runs comfortably 128K 16.5
INT8 / W8A8 8.50 20.4 GB 21.3 GB Runs comfortably 128K 30.3
FP8 (E4M3) 8.00 19.5 GB 20.3 GB Runs comfortably 128K 31.6
Q6_K 6.56 16.0 GB 16.8 GB Runs comfortably 128K 37.9
Q5_K_M 5.67 13.8 GB 14.7 GB Runs comfortably 128K 43.1
Q5_K_S 5.52 13.4 GB 14.3 GB Runs comfortably 128K 44.2
AWQ 4-bit 4.25 11.9 GB 12.8 GB Runs comfortably 128K 49.0
GPTQ 4-bit 4.25 11.9 GB 12.8 GB Runs comfortably 128K 49.0
MXFP4 4.25 11.9 GB 12.8 GB Runs comfortably 128K 49.0
Q4_K_M 4.85 11.9 GB 12.8 GB Runs comfortably 128K 49.1
Q4_K_S 4.58 11.3 GB 12.1 GB Runs comfortably 128K 51.5
Q4_0 4.55 11.2 GB 12.1 GB Runs comfortably 128K 51.8
IQ4_XS 4.25 10.5 GB 11.4 GB Runs comfortably 128K 54.7
Q3_K_M 3.91 9.7 GB 10.6 GB Runs comfortably 128K 58.4
IQ3_M 3.70 9.2 GB 10.1 GB Runs comfortably 128K 60.9
IQ3_XXS 3.06 7.8 GB 8.6 GB Runs comfortably 128K 70.3
Q2_K 2.63 6.8 GB 7.7 GB Runs comfortably 128K 78.4
IQ2_XXS 2.06 5.5 GB 6.4 GB Runs comfortably 128K 92.6
IQ1_M 1.75 4.8 GB 5.6 GB Runs comfortably 128K 103

Also worth checking

gpt-oss 20B-A3.6B 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