# Can the Radeon RX 9070 XT run lumi-4b?

> Yes. The Radeon RX 9070 XT runs lumi-4b fully in VRAM at FP16 / BF16 (about 9.6 GB of 16 GB), at roughly 44 tokens/s.

Radeon RX 9070 XT: 16 GB (16 GB usable), 640 GB/s. Estimates at an 8K context.

## Every quantisation

| Quant | Quality | ≈ GB | Fit | Est. tok/s |
|---|---|---|---|---|
| Q2_K | Heavy loss | 3.09 | fits | 269 |
| Q3_K_M | Noticeable loss | 3.71 | fits | 181 |
| Q4_K_M | Small loss | 4.17 | fits | 146 |
| Q5_K_M | Near-lossless | 4.58 | fits | 125 |
| Q6_K | Near-lossless | 5.65 | fits | 108 |
| Q8_0 | Lossless | 5.94 | fits | 83 |
| F16 (pick) | Lossless | 9.59 | fits | 44 |

## Run it

OpenAI-compatible server with a web UI (llama.cpp):

```bash
llama-server -hf Mateus-Nery/lumi-4b-GGUF:BF16
```

Docker with ROCm (AMD) (llama.cpp):

```bash
docker run -d --name llama --device /dev/kfd --device /dev/dri -p 8080:8080 \
  -v ~/.cache/llama.cpp:/root/.cache/llama.cpp \
  ghcr.io/ggml-org/llama.cpp:server-rocm \
  -hf Mateus-Nery/lumi-4b-GGUF:BF16 --host 0.0.0.0 --port 8080 -ngl 99
```

Serve an OpenAI-compatible API (vllm):

```bash
vllm serve ceia-br/lumi-4b
```

Smaller FP8 build (vllm):

```bash
vllm serve liodon-ai/lumi-4b-FP8
```

Docker Compose (AMD ROCm) (vllm):

```yaml
services:
  vllm:
    image: rocm/vllm:latest
    command: vllm serve ceia-br/lumi-4b --host 0.0.0.0 --port 8000
    ports:
      - "8000:8000"
    ipc: host
    devices:
      - /dev/kfd
      - /dev/dri
    group_add:
      - video
    security_opt:
      - seccomp=unconfined
    volumes:
      - ~/.cache/huggingface:/root/.cache/huggingface
```

Serve an OpenAI-compatible API (sglang):

```bash
sglang serve ceia-br/lumi-4b --host 0.0.0.0 --port 30000
```


## Chat models that fit comfortably

- [Qwen/Qwen3-4B](https://selfhost.directory/models/qwen/qwen3-4b/on/rx-9070-xt): Q8_0 ≈5.82 GB, est. 82 tok/s
- [meta-llama/Llama-3.2-3B-Instruct](https://selfhost.directory/models/meta-llama/llama-3.2-3b-instruct/on/rx-9070-xt): Q8_0 ≈4.73 GB, est. 103 tok/s
- [microsoft/Phi-4-mini-instruct](https://selfhost.directory/models/microsoft/phi-4-mini-instruct/on/rx-9070-xt): Q8_0 ≈5.49 GB, est. 86 tok/s
- [Qwen/Qwen3-4B-Instruct-2507](https://selfhost.directory/models/qwen/qwen3-4b-instruct-2507/on/rx-9070-xt): Q8_0 ≈5.82 GB, est. 82 tok/s
- [XHToken/Spark-X2.6-4B](https://selfhost.directory/models/xhtoken/spark-x2.6-4b/on/rx-9070-xt): Q8_0 ≈5.9 GB, est. 81 tok/s
- [Qwen/Qwen2.5-3B-Instruct](https://selfhost.directory/models/qwen/qwen2.5-3b-instruct/on/rx-9070-xt): Q8_0 ≈4.32 GB, est. 107 tok/s

## Other GPUs that run it

- [GeForce RTX 4060](https://selfhost.directory/models/ceia-br/lumi-4b/on/rtx-4060): Q8_0, est. 35 tok/s
- [Apple M4 (16 GB)](https://selfhost.directory/models/ceia-br/lumi-4b/on/apple-m4-16gb): F16, est. 8.3 tok/s
- [Radeon RX 6700 XT](https://selfhost.directory/models/ceia-br/lumi-4b/on/rx-6700-xt): F16, est. 27 tok/s
- [GeForce RTX 3060 12GB](https://selfhost.directory/models/ceia-br/lumi-4b/on/rtx-3060): F16, est. 25 tok/s
- [Arc B580](https://selfhost.directory/models/ceia-br/lumi-4b/on/arc-b580): F16, est. 32 tok/s
- [Tesla T4](https://selfhost.directory/models/ceia-br/lumi-4b/on/tesla-t4): F16, est. 22 tok/s
- [Arc A770 16GB](https://selfhost.directory/models/ceia-br/lumi-4b/on/arc-a770): F16, est. 39 tok/s
- [Radeon RX 7800 XT](https://selfhost.directory/models/ceia-br/lumi-4b/on/rx-7800-xt): F16, est. 43 tok/s

---
Model: https://selfhost.directory/models/ceia-br/lumi-4b. GPU: https://selfhost.directory/models/gpus/rx-9070-xt. Source: https://selfhost.directory/models/ceia-br/lumi-4b/on/rx-9070-xt