Skip to content

Gemma 4 12B on RTX 3080 10 GB: VRAM, speed and fit

Gemma 4 12B at 5-bit runs on the RTX 3080 10 GB with 24K of context at about 57 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on RTX 3080 10 GB · 5-bit (Q5_K_M) · Ollama / LM Studio · Chat

GPU memory
9.5 GB
of 10.0 GB
Context
24K
auto
Speed
57 tok/s
Very fast
First reply
0.8 s
2K prompt

Where the memory goes

Model
7.9 GB
Context
512 MB
Overhead
1.1 GB
OS
0.0 GB
Free
523 MB
Memory
9.5 GB of 10.0 GB
Context
24K tokens · auto
Speed
57 tok/s · very fast
First reply
0.8 s for a 2K prompt
Running cost
$5.95/mo at home · $29/mo on a cloud RTX 2000 Ada

Validation: Validated

What Gemma 4 12B on RTX 3080 10 GB was checked against
CheckTargetResultStatus
Fits in GPU memory10.0 GB usable9.5 GB neededPass
Headroom5% free523 MB free (5%)Pass
Context for chat16K or more24KPass
Writing speed15 tok/s or more57 tok/sPass
First replyunder 3 s, 2K prompt0.8 sPass

Targets: a reply you can read along with, a first reply the use case tolerates, and 5% of memory left free. Calculated from published specs, not benchmarked.

The setup

Model
Gemma 4 12B
Parameters
12B
Precision
5-bit (Q5_K_M)
Context
24K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
RTX 3080 10 GB
GPU can use
10.0 GB of 10.0 GB

Worth trying

  • 8-bit cache: Room for 80K context

Running cost

At home
$5.95/mo
272 W at the wall
Renting instead
$29/mo
RTX 2000 Ada at $0.24/h

4 h a day, power at $0.18 per kWh, cloud at RunPod on-demand prices as of Nov 21, 2026.

How to run it

Ollama
ollama run gemma4:12b
llama.cpp
llama-server -hf unsloth/gemma-4-12b-it-GGUF
vLLM
vllm serve google/gemma-4-12B-it
MLX (Mac)
mlx_lm.chat --model mlx-community/gemma-4-12B-it-4bit

Gemma 4 12B on similar machines

All machines: Gemma 4 12B VRAM requirements. Everything that runs here: what can the RTX 3080 10 GB run?

Other models that run on RTX 3080 10 GB

Questions

Can Gemma 4 12B run on RTX 3080 10 GB?

Gemma 4 12B at 5-bit runs on the RTX 3080 10 GB with 24K of context at about 57 tokens a second.

Report VR-15HBYTT, calculated Nov 22, 2026 from published specs, not benchmarked.