Can I run it?

Can I run gemma 4 31B it on GeForce RTX 3070?

Partial offload

Partial offload on your GeForce RTX 3070: Q4_K_M at 4K context, 18 of 60 layers on the GPU and 11.8 GB in system RAM. Expect about 3.5–4.8 tok/s.

gemma 4 31B it details · GeForce RTX 3070 page · How this estimate is made

Run commands

llama-server -hf unsloth/gemma-4-31B-it-GGUF:Q4_K_M -c 4096 -ngl 18

ollama run gemma

lms get unsloth/gemma-4-31B-it-GGUF
# Then load quant Q4_K_M

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%.
• Q4_K_M size is the real GGUF file size from Hugging Face.
• 0.5 GB is reserved because this GPU also drives a display, then 5% of the rest is left free.
• Speed is quoted at 4K context, not at the maximum context that still fits.
• About 18 of 60 layers stay on the GPU. 11.8 GB is read from system RAM.

Models that fit this machine more comfortably

• Can I run Qwen3.8 27B on GeForce RTX 3070? — Heavily degraded, UD-IQ1_M, 35–48 tok/s (estimate)
• gemma 3 27b it — Heavily degraded, UD-IQ1_S, 35–48 tok/s (estimate)
• medgemma 27b text it — Heavily degraded, UD-IQ1_S, 33–45 tok/s (estimate)

Updated 2026-10-24. Profile on this page is the device default plus 32 GB of DDR5-5600 system RAM.