Gemma on AMD Strix Halo
Can I run Gemma 3 27B on an Ryzen AI Max+ 395 128GB?
Yes. Gemma 3 27B at Q4_K_M uses 17.2 GB of the 96.0 GB available on a Ryzen AI Max+ 395 128GB, and runs at about 8.21 tokens per second. You can push the context to 128K.
Runs comfortably
17.2 GB of 96.0 GB · 18%096 GB
Weights
15.6 GB
KV cache
1.1 GB
Runtime overhead
0.5 GB
Gemma 3 27B at Q4_K_M leaves 78.8 GB spare on a Ryzen AI Max+ 395 128GB. There is room to raise the context length or move up a quantisation level.
Generation8.21tok/s
Prompt processing383tok/s
Max context128Ktokens
KV per 1K tokens0GB
Every quantisation of Gemma 3 27B on a Ryzen AI Max+ 395 128GB
Highlighted row is the highest quality that still fits at 8K context.
| Quantisation | bpw | Weights | Total | Verdict | Max ctx | tok/s |
|---|---|---|---|---|---|---|
| FP16 / BF16 | 16.00 | 51.0 GB | 52.7 GB | Runs comfortably | 128K | 2.59 |
| INT8 / W8A8 | 8.50 | 26.8 GB | 28.4 GB | Runs comfortably | 128K | 4.87 |
| Q8_0 (GGUF) | 8.50 | 26.8 GB | 28.4 GB | Runs comfortably | 128K | 4.87 |
| FP8 (E4M3) | 8.00 | 25.5 GB | 27.1 GB | Runs comfortably | 128K | 5.11 |
| Q6_K | 6.56 | 20.9 GB | 22.6 GB | Runs comfortably | 128K | 6.19 |
| Q5_K_M | 5.67 | 18.1 GB | 19.7 GB | Runs comfortably | 128K | 7.12 |
| Q5_K_S | 5.52 | 17.6 GB | 19.2 GB | Runs comfortably | 128K | 7.31 |
| Q4_K_M | 4.85 | 15.6 GB | 17.2 GB | Runs comfortably | 128K | 8.21 |
| AWQ 4-bit | 4.25 | 15.5 GB | 17.1 GB | Runs comfortably | 128K | 8.26 |
| GPTQ 4-bit | 4.25 | 15.5 GB | 17.1 GB | Runs comfortably | 128K | 8.26 |
| MXFP4 | 4.25 | 15.5 GB | 17.1 GB | Runs comfortably | 128K | 8.26 |
| Q4_K_S | 4.58 | 14.8 GB | 16.4 GB | Runs comfortably | 128K | 8.65 |
| Q4_0 | 4.55 | 14.7 GB | 16.3 GB | Runs comfortably | 128K | 8.70 |
| IQ4_XS | 4.25 | 13.8 GB | 15.4 GB | Runs comfortably | 128K | 9.24 |
| Q3_K_M | 3.91 | 12.7 GB | 14.4 GB | Runs comfortably | 128K | 9.94 |
| IQ3_M | 3.70 | 12.1 GB | 13.7 GB | Runs comfortably | 128K | 10.4 |
| IQ3_XXS | 3.06 | 10.2 GB | 11.8 GB | Runs comfortably | 128K | 12.3 |
| Q2_K | 2.63 | 8.9 GB | 10.5 GB | Runs comfortably | 128K | 13.9 |
| IQ2_XXS | 2.06 | 7.1 GB | 8.8 GB | Runs comfortably | 128K | 17.0 |
| IQ1_M | 1.75 | 6.2 GB | 7.8 GB | Runs comfortably | 128K | 19.2 |