Skip to content

Llama Guard 4 12B on RTX 5090: VRAM, speed and fit

Llama Guard 4 12B at 8-bit runs on the RTX 5090 with 88K of context at about 93 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on RTX 5090 · 8-bit (Q8_0) · Ollama / LM Studio · Chat

GPU memory
29.4 GB
of 32.0 GB
Context
88K
auto
Speed
93 tok/s
40 when full
First reply
0.2 s
2K prompt

Where the memory goes

Model
11.9 GB
Context
16.5 GB
Overhead
1.0 GB
OS
0.0 GB
Free
2.6 GB
Memory
29.4 GB of 32.0 GB
Context
88K tokens · auto
Speed
93 tok/s · 40 when full
First reply
0.2 s for a 2K prompt
Running cost
$9.30/mo at home · $58/mo on a cloud 2× RTX 2000 Ada

Validation: Validated

What Llama Guard 4 12B on RTX 5090 was checked against
CheckTargetResultStatus
Fits in GPU memory32.0 GB usable29.4 GB neededPass
Headroom5% free2.6 GB free (8%)Pass
Context for chat16K or more88KPass
Writing speed15 tok/s or more93 tok/sPass
First replyunder 3 s, 2K prompt0.2 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
Llama Guard 4 12B
Parameters
12B
Precision
8-bit (Q8_0)
Context
88K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
RTX 5090
GPU can use
32.0 GB of 32.0 GB

Running cost

At home
$9.30/mo
425 W at the wall
Renting instead
$58/mo
2× RTX 2000 Ada at $0.48/h

4 h a day, power at $0.18 per kWh, cloud at RunPod on-demand prices as of Nov 22, 2026.

How to run it

llama.cpp
llama-server -hf DevQuasar/meta-llama.Llama-Guard-4-12B-GGUF
vLLM
vllm serve meta-llama/Llama-Guard-4-12B

Llama Guard 4 12B on similar machines

All machines: Llama Guard 4 12B VRAM requirements. Everything that runs here: what can the RTX 5090 run?

Other models that run on RTX 5090

Questions

Can Llama Guard 4 12B run on RTX 5090?

Llama Guard 4 12B at 8-bit runs on the RTX 5090 with 88K of context at about 93 tokens a second.

Report VR-1YPPABY, calculated Nov 22, 2026 from published specs, not benchmarked.