Skip to content

Apertus 8B Instruct 2509 on 2× RTX 3090: VRAM, speed and fit

Apertus 8B Instruct 2509 at 8-bit runs on the 2× RTX 3090 with 64K of context at about 76 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on 2× RTX 3090 · 8-bit (Q8_0) · Ollama / LM Studio · Chat

GPU memory
17.4 GB
of 48.0 GB
Context
64K
auto
Speed
76 tok/s
38 when full
First reply
0.3 s
2K prompt

Where the memory goes

Model
8.0 GB
Context
8.0 GB
Overhead
1.5 GB
OS
0.0 GB
Free
30.6 GB
Memory
17.4 GB of 48.0 GB
Context
64K tokens · auto
Speed
76 tok/s · 38 when full
First reply
0.3 s for a 2K prompt
Running cost
$11/mo at home · $30/mo on a cloud RTX A4500

Validation: Validated

What Apertus 8B Instruct 2509 on 2× RTX 3090 was checked against
CheckTargetResultStatus
Fits in GPU memory48.0 GB usable17.4 GB neededPass
Headroom5% free30.6 GB free (64%)Pass
Context for chat16K or more64KPass
Writing speed15 tok/s or more76 tok/sPass
First replyunder 3 s, 2K prompt0.3 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
Apertus 8B Instruct 2509
Parameters
8.1B
Precision
8-bit (Q8_0)
Context
64K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
2× RTX 3090
GPU can use
48.0 GB of 48.0 GB

Running cost

At home
$11/mo
500 W at the wall
Renting instead
$30/mo
RTX A4500 at $0.25/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

llama.cpp
llama-server -hf unsloth/Apertus-8B-Instruct-2509-GGUF
vLLM
vllm serve swiss-ai/Apertus-8B-Instruct-2509
MLX (Mac)
mlx_lm.chat --model mlx-community/Apertus-8B-Instruct-2509-4bit

Apertus 8B Instruct 2509 on similar machines

All machines: Apertus 8B Instruct 2509 VRAM requirements. Everything that runs here: what can the 2× RTX 3090 run?

Other models that run on 2× RTX 3090

Questions

Can Apertus 8B Instruct 2509 run on 2× RTX 3090?

Apertus 8B Instruct 2509 at 8-bit runs on the 2× RTX 3090 with 64K of context at about 76 tokens a second.

Report VR-1281HGB, calculated Nov 22, 2026 from published specs, not benchmarked.