Skip to content

Qwen3.5 27B on RTX 3090: VRAM, speed and fit

Qwen3.5 27B at 5-bit runs on the RTX 3090 with 48K of context at about 32 tokens a second.

Text generation · Chat · Ollama / LM Studio

Runs comfortably · on RTX 3090 · 5-bit (Q5_K_M) · Ollama / LM Studio · Chat

GPU memory
22.6 GB
of 24.0 GB
Context
48K
auto
Speed
32 tok/s
Fast
First reply
1.6 s
2K prompt

Where the memory goes

Model
18.4 GB
Context
3.1 GB
Overhead
1.1 GB
OS
0.0 GB
Free
1.4 GB
Memory
22.6 GB of 24.0 GB
Context
48K tokens · auto
Speed
32 tok/s · fast
First reply
1.6 s for a 2K prompt
Running cost
$6.35/mo at home · $33/mo on a cloud RTX A5000

Validation: Validated

What Qwen3.5 27B on RTX 3090 was checked against
CheckTargetResultStatus
Fits in GPU memory24.0 GB usable22.6 GB neededPass
Headroom5% free1.4 GB free (6%)Pass
Context for chat16K or more48KPass
Writing speed15 tok/s or more32 tok/sPass
First replyunder 3 s, 2K prompt1.6 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
Qwen3.5 27B
Parameters
27.8B
Precision
5-bit (Q5_K_M)
Context
48K, auto
Chats at once
1
Use
Chat
Runs in
Ollama / LM Studio
Hardware
RTX 3090
GPU can use
24.0 GB of 24.0 GB

Worth trying

  • 8-bit cache: Room for 88K context

Running cost

At home
$6.35/mo
290 W at the wall
Renting instead
$33/mo
RTX A5000 at $0.27/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 unsloth/Qwen3.5-27B-GGUF
vLLM
vllm serve Qwen/Qwen3.5-27B
MLX (Mac)
mlx_lm.chat --model mlx-community/Qwen3.5-27B-mxfp8

Qwen3.5 27B on similar machines

All machines: Qwen3.5 27B VRAM requirements. Everything that runs here: what can the RTX 3090 run?

Other models that run on RTX 3090

Questions

Can Qwen3.5 27B run on RTX 3090?

Qwen3.5 27B at 5-bit runs on the RTX 3090 with 48K of context at about 32 tokens a second.

Report VR-0T5FETZ, calculated Nov 22, 2026 from published specs, not benchmarked.