Only with CPU offload
Gemma 4 12B at Q4_K_M needs 8.2 GB but only 7.0 GB is addressable, so about 17% of the layers would stream from system RAM at roughly 60 GB/s. Expect around 20 tokens per second — usable for batch work, painful for chat.
The "unified" Gemma 4: text, image and audio in one 12B that fits a 12 GB card at Q4. 140+ languages.
Hardware requirements What hardware do I need to run Gemma 4 12B? →At Q4_K_M with an 8K context: a 11 GB GPU (RTX 2080 Ti) or a 16 GB Mac (M4). Minimum, recommended, high-quality and long-context tiers, with all 114 devices checked at every quantisation.
Fits instead: Gemma 4 E4B (5.2 GB) · Gemma 4 E2B (3.5 GB)
The VRAM budget
Quantisation ladder
| Quant | Weights | Total @ 8K | Max context | Tok/s | Quality | Fit |
|---|---|---|---|---|---|---|
| F16 | 22.4 GB | 23.8 GB | — | ~2.1 | Reference | 16.8 GB over |
| Q8_0 | 11.9 GB | 13.3 GB | — | ~5.2 | −0.1% ppl | 6.3 GB over |
| Q6_K | 9.2 GB | 10.6 GB | — | ~8.6 | −0.4% ppl | 3.6 GB over |
| Q5_K_M | 7.9 GB | 9.3 GB | — | ~12 | −0.8% ppl | 2.3 GB over |
| Q4_K_M | 6.7 GB | 8.2 GB | — | ~20 | −1.9% ppl | 1.2 GB over |
| Q3_K_M | 5.5 GB | 6.9 GB | 10K | 57 | −5.4% ppl | Fits |
| Q2_K | 4.7 GB | 6.1 GB | 22K | 66 | −15% ppl | Long context |
Quality is the published perplexity delta against f16 weights. Max context assumes an f16 KV cache; q8_0 roughly doubles it. This model interleaves sliding-window layers (1024 tokens, 8 of 48 layers global), which is why its cache barely grows with context.
How to run it
$ llama-server \
-hf google/gemma-4-12B-it:Q4_K_M \
-c 8192 -ngl 39
The engine underneath most of the others. Every knob is exposed. More on llama.cpp.
Related models on an Arc A750
| Model | Total | Tok/s est. | |
|---|---|---|---|
| Gemma 4 E4B 8.0B · same family |
Q4_K_M · 5.2 GB | 69 | Runs great |
| Gemma 4 26B-A4B 26.5B MoE · same family |
Q4_K_M · 16.0 GB | ~10 | CPU offload |
| Gemma 4 E2B 5.1B · same family |
Q4_K_M · 3.5 GB | 108 | Runs great |
| Gemma 4 31B 31.3B · same family |
Q4_K_M · 20.2 GB | ~2.6 | CPU offload |
| Gemma 3 12B 12.2B · similar size |
Q4_K_M · 8.3 GB | ~19 | CPU offload |
| Mistral NeMo 12B 12.2B · similar size |
Q4_K_M · 8.7 GB | ~16 | CPU offload |
| Ministral 3 14B 13.9B · similar size |
Q4_K_M · 9.7 GB | ~11 | CPU offload |
Gemma 4 12B on other hardware
| Device | Total | Tok/s est. | |
|---|---|---|---|
| GeForce RTX 4060 Laptop 7.0 GB usable · 256 GB/s |
8.2 GB | ~15 | CPU offload |
| GeForce RTX 2080 Ti 9.7 GB usable · 616 GB/s |
8.2 GB | 55 | Runs great |
| GeForce RTX 3060 12 GB 10.6 GB usable · 360 GB/s |
8.2 GB | 32 | Runs great |
| M4 · 16 GB 10.7 GB usable · 120 GB/s |
8.2 GB | 9.9 | Runs great |
| GeForce RTX 4090 22.4 GB usable · 1008 GB/s |
8.2 GB | 90 | Runs great |
| Radeon RX 7900 XTX 22.4 GB usable · 960 GB/s |
8.2 GB | 86 | Runs great |
| M3 Max · 64 GB 48.0 GB usable · 400 GB/s |
8.2 GB | 33 | Runs great |