Qwen3.5 9B (Q8_K_XL) — 9.4 GBon NVIDIA RTX 3080

Q8_K_XL NVIDIA RTX 3080

Overview

Qwen3.5 9B is a 9.65B parameter dense language model by Qwen, with code, multilingual, thinking, tool-calls, vision capabilities. It supports a context window of up to 262,144 tokens.

Qwen3.5 9B is the flagship small model in Alibaba's Qwen 3.5 family, built on the Gated Delta Networks hybrid architecture with 9.65 billion parameters, outperforming gpt-oss-120B on GPQA Diamond with 81.7 versus 80.1 at thirteen times fewer parameters. It is natively multimodal, processing text, images, and video, with built-in thinking capabilities for chain-of-thought reasoning. The model supports a 262K context window and covers over 201 languages. Released under the Apache 2.0 license, it runs in roughly 5 GB of VRAM at Q4, making it a top choice for self-hosted deployment on consumer hardware.

At Q8_K_XL quantization (high quality tier), the model weighs 12.08 GB. This exceeds the 10 GB of VRAM on NVIDIA RTX 3080. Inference is still possible via CPU offload or memory-mapped loading from disk, but expect significantly reduced performance.

Hardware Requirements

Model size	12.08 GB
VRAM available	10 GB
VRAM used	9.4 GB
Min RAM required	4.2 GB
GPU layers	21 / 32
Context size	1,865
Backend	cuda13
Flash attention	Yes

Performance Notes

Reduced performance

11 of 32 layers run on CPU. Your system needs at least 4.2 GB RAM. If system RAM is insufficient, data will be read from disk on demand, with significant performance impact.

Shorter conversations

Context size has been reduced due to memory constraints. This may affect the model's ability to handle long conversations.

Deploy

Server Local

Prerequisites

Ensure your GPU nodes are prepared with the NVIDIA container toolkit:

ansible-playbook prositronic.infra.nvidia_container_toolkit

Command

helmfile --state-values-file <(curl -s https://www.prositronic.eu/values/qwen3-5-9b/q8_k_xl/nvidia-rtx3080.yaml) apply

Generated values.yaml

/values/qwen3-5-9b/q8_k_xl/nvidia-rtx3080.yaml

Loading values…

Install llama.cpp

Install llama.cpp from the official build instructions for your platform and backend.

Download Model

curl -L -o qwen3-5-9b.gguf "https://huggingface.co/unsloth/Qwen3.5-9B-GGUF/resolve/main/Qwen3.5-9B-UD-Q8_K_XL.gguf"

Start Server

llama-server \
  -m qwen3-5-9b.gguf \
  --n-gpu-layers 21 \
  --ctx-size 1865 \
  --flash-attn

Verify

curl http://localhost:8080/health

Explore

• Configure a deployment
• Browse models
• Browse hardware

Other quantizations

• Qwen3.5 9B (Q8_0)
• Qwen3.5 9B (Q5_K_M)
• Qwen3.5 9B (Q4_K_M)

Other hardware

• Qwen3.5 9B on CPU Only
• Qwen3.5 9B on NVIDIA A100 80GB
• Qwen3.5 9B on NVIDIA H100

Frequently Asked Questions

How much VRAM does Qwen3.5 9B (Q8_K_XL) need?

The Q8_K_XL quantization of Qwen3.5 9B requires 12.08 GB. 21 of 32 layers fit in the 10 GB of VRAM on NVIDIA RTX 3080; remaining layers are offloaded to CPU.

Can I run Qwen3.5 9B on NVIDIA RTX 3080?

Yes, with reduced performance. NVIDIA RTX 3080 can run Qwen3.5 9B (Q8_K_XL), but only 21 of 32 layers fit in VRAM. The rest are offloaded to CPU.

What is quantization?

Quantization reduces a model's numerical precision from its original floating-point format to a more compact representation. This shrinks the file size and VRAM footprint, making it possible to run large models on consumer hardware. The trade-off is a small reduction in output quality. Q8_K_XL compresses Qwen3.5 9B from its original size down to 12.08 GB.

What quantization should I choose for Qwen3.5 9B?

Q8_K_XL is a high-quality quantization. Higher-quality quants (Q8, Q6) preserve more model accuracy but need more VRAM. Lower quants (Q4, Q3, Q2) reduce VRAM usage at the cost of some quality. Choose based on your available hardware and quality requirements.

What is flash attention and why is it enabled?

Flash attention is a memory-efficient algorithm that speeds up the attention mechanism in transformer models. It reduces VRAM usage during inference by avoiding the materialisation of the full attention matrix. For Qwen3.5 9B on NVIDIA RTX 3080, flash attention is enabled to maximise context length and throughput within the available 10 GB of VRAM.

Why are some layers offloaded to CPU?

NVIDIA RTX 3080 has 10 GB of VRAM, but Qwen3.5 9B (Q8_K_XL) requires approximately 12.08 GB. Only 21 of 32 layers fit in VRAM; the remaining layers run on CPU, which is slower but still functional.

How do I run Qwen3.5 9B (Q8_K_XL) with Ollama?

Run ollama run qwen3.5:9b-q8_k_xl to start Qwen3.5 9B (Q8_K_XL). Ollama handles downloading the model weights automatically on first run.

Last updated: March 13, 2026