Can I run it?
Can I run gemma 4 12b it on AMD Ryzen AI Max+ 395?
Runs great
Runs great fully in unified memory on your AMD Ryzen AI Max+ 395 at BF16, 8K context, about 5.2–7.1 tok/s. Weights 22.2 GB plus 0.81 GB of KV cache — 23.5 GB of 48.0 GB usable. Up to 256K context still fits.
Best quantBF16
Working context8K
Max that fits256K
Speed5.2–7.1 tok/sEstimate from memory bandwidth, not a benchmark. Method.
Weights22.2 GB
KV cache0.81 GB
Total23.5 GB
Usable48.0 GB
gemma 4 12b it details · AMD Ryzen AI Max+ 395 page · How this estimate is made
Run commands
llama.cpp
llama-server -hf unsloth/gemma-4-12b-it-GGUF:BF16 -c 8192 -ngl 48
Ollama
ollama run gemma
LM Studio
lms get unsloth/gemma-4-12b-it-GGUF # Then load quant BF16
Assumptions
- Estimate from memory bandwidth, not a benchmark. Decode speed uses 55–75% of peak bandwidth.
- KV cache is FP16. Engine overhead assumes llama.cpp (0.3 GB + 1% of the weights). Ollama is closer to 0.9 GB + 3%, LM Studio closer to 1 GB + 3%.
- BF16 size is the real GGUF file size from Hugging Face.
- Unified memory uses 75% for the GPU by default, and always leaves about 5 GB for the OS.
- Speed is quoted at 8K context, not at the maximum context that still fits.
Larger models that also fit
- Can I run Qwen3.8 27B on AMD Ryzen AI Max+ 395? — Runs great, UD-Q8_K_L, 4.9–6.7 tok/s (estimate)
- Can I run Qwen3.6 35B A3B on AMD Ryzen AI Max+ 395? — Runs, UD-Q8_K_XL, 8.9–12 tok/s (estimate)
- Can I run Qwen3.8 35B A3B Distill on AMD Ryzen AI Max+ 395? — Runs, Q8_0, 9.1–12 tok/s (estimate)
Updated 2026-10-24. Profile on this page is the device default.