Gemma on AMD Strix Halo

Can I run Gemma 3 12B on an Ryzen AI Max+ 395 128GB?

Yes. Gemma 3 12B at Q4_K_M uses 8.3 GB of the 96.0 GB available on a Ryzen AI Max+ 395 128GB, and runs at about 17.8 tokens per second. You can push the context to 128K.

Runs comfortably

8.3 GB of 96.0 GB · 9%
096 GB
Weights 7.0 GB
KV cache 0.8 GB
Runtime overhead 0.5 GB

Gemma 3 12B at Q4_K_M leaves 87.7 GB spare on a Ryzen AI Max+ 395 128GB. There is room to raise the context length or move up a quantisation level.

Generation17.8tok/s
Prompt processing861tok/s
Max context128Ktokens
KV per 1K tokens0GB

Every quantisation of Gemma 3 12B 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 22.7 GB 24.1 GB Runs comfortably 128K 5.76
INT8 / W8A8 8.50 11.8 GB 13.2 GB Runs comfortably 128K 10.8
Q8_0 (GGUF) 8.50 11.8 GB 13.2 GB Runs comfortably 128K 10.8
FP8 (E4M3) 8.00 11.4 GB 12.7 GB Runs comfortably 128K 11.2
Q6_K 6.56 9.3 GB 10.6 GB Runs comfortably 128K 13.6
Q5_K_M 5.67 8.1 GB 9.4 GB Runs comfortably 128K 15.5
Q5_K_S 5.52 7.8 GB 9.2 GB Runs comfortably 128K 15.9
AWQ 4-bit 4.25 7.4 GB 8.7 GB Runs comfortably 128K 16.8
GPTQ 4-bit 4.25 7.4 GB 8.7 GB Runs comfortably 128K 16.8
MXFP4 4.25 7.4 GB 8.7 GB Runs comfortably 128K 16.8
Q4_K_M 4.85 7.0 GB 8.3 GB Runs comfortably 128K 17.8
Q4_K_S 4.58 6.6 GB 7.9 GB Runs comfortably 128K 18.6
Q4_0 4.55 6.6 GB 7.9 GB Runs comfortably 128K 18.7
IQ4_XS 4.25 6.2 GB 7.5 GB Runs comfortably 128K 19.8
Q3_K_M 3.91 5.7 GB 7.1 GB Runs comfortably 128K 21.2
IQ3_M 3.70 5.5 GB 6.8 GB Runs comfortably 128K 22.2
IQ3_XXS 3.06 4.6 GB 6.0 GB Runs comfortably 128K 25.7
Q2_K 2.63 4.1 GB 5.4 GB Runs comfortably 128K 28.8
IQ2_XXS 2.06 3.3 GB 4.7 GB Runs comfortably 128K 34.2
IQ1_M 1.75 2.9 GB 4.3 GB Runs comfortably 128K 38.1

Also worth checking

Gemma 3 12B 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