# Can the Radeon RX 7900 XTX run h2o-lightning-31b?

> Yes. The Radeon RX 7900 XTX runs h2o-lightning-31b fully in VRAM at Q4_K_M (about 21 GB of 24 GB), at roughly 28 tokens/s.

Radeon RX 7900 XTX: 24 GB (24 GB usable), 960 GB/s. Estimates at an 8K context.

## Every quantisation

| Quant | Quality | ≈ GB | Fit | Est. tok/s |
|---|---|---|---|---|
| Q2_K | Heavy loss | 12.58 | fits | 51 |
| Q3_K_M | Noticeable loss | 17.47 | fits | 35 |
| Q4_K_M (pick) | Small loss | 21.06 | fits | 28 |
| Q5_K_M | Near-lossless | 24.27 | offload | 20 |
| Q6_K | Near-lossless | 27.72 | offload | 6.8 |
| Q8_0 | Lossless | 35.02 | offload | 2.8 |
| F16 | Lossless | 63.69 | no | — |

## Run it

Serve an OpenAI-compatible API (vllm):

```bash
vllm serve h2oai/h2o-lightning-31b
```

Docker Compose (AMD ROCm) (vllm):

```yaml
services:
  vllm:
    image: rocm/vllm:latest
    command: vllm serve h2oai/h2o-lightning-31b --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
    environment:
      - HF_TOKEN=${HF_TOKEN}
```

Serve an OpenAI-compatible API (sglang):

```bash
sglang serve h2oai/h2o-lightning-31b --host 0.0.0.0 --port 30000
```

Ask about an image (transformers):

```python
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="h2oai/h2o-lightning-31b", dtype="auto", device_map="auto")
messages = [
    {
        "role": "user",
        "content": [
            {"type": "image", "url": "https://huggingface.co/datasets/huggingface/documentation-images/resolve/main/bee.jpg"},
            {"type": "text", "text": "Describe this image."},
        ],
    }
]
out = pipe(text=messages, max_new_tokens=256)
print(out[0]["generated_text"][-1]["content"])
```


## Vision models that fit comfortably

- [Qwen/Qwen3.8-27B](https://selfhost.directory/models/qwen/qwen3.8-27b/on/rx-7900-xtx): Q5_K_M ≈21.82 GB, est. 27 tok/s
- [Qwen/Qwen3.6-35B-A3B](https://selfhost.directory/models/qwen/qwen3.6-35b-a3b/on/rx-7900-xtx): Q4_K_M ≈22.44 GB, est. 290 tok/s
- [Qwen/Qwen3.6-27B](https://selfhost.directory/models/qwen/qwen3.6-27b/on/rx-7900-xtx): Q5_K_M ≈21.58 GB, est. 27 tok/s
- [google/gemma-4-26B-A4B-it](https://selfhost.directory/models/google/gemma-4-26b-a4b-it/on/rx-7900-xtx): Q5_K_M ≈20.32 GB, est. 186 tok/s
- [meta-models/Muse-Glimmer-30B](https://selfhost.directory/models/meta-models/muse-glimmer-30b/on/rx-7900-xtx): Q5_K_M ≈21.62 GB, est. 25 tok/s
- [Qwen/Qwen3.5-35B-A3B](https://selfhost.directory/models/qwen/qwen3.5-35b-a3b/on/rx-7900-xtx): Q4_K_M ≈22.65 GB, est. 290 tok/s

## Other GPUs that run it

- [Tesla P40](https://selfhost.directory/models/h2oai/h2o-lightning-31b/on/tesla-p40): Q4_K_M, est. 10 tok/s
- [Instinct MI50 32GB](https://selfhost.directory/models/h2oai/h2o-lightning-31b/on/instinct-mi50): Q6_K, est. 22 tok/s
- [GeForce RTX 5090](https://selfhost.directory/models/h2oai/h2o-lightning-31b/on/rtx-5090): Q6_K, est. 38 tok/s
- [Apple M4 Pro (48 GB)](https://selfhost.directory/models/h2oai/h2o-lightning-31b/on/apple-m4-pro-48gb): Q6_K, est. 5.8 tok/s
- [RTX A6000](https://selfhost.directory/models/h2oai/h2o-lightning-31b/on/rtx-a6000): Q8_0, est. 13 tok/s
- [Apple M1 Max (64 GB)](https://selfhost.directory/models/h2oai/h2o-lightning-31b/on/apple-m1-max-64gb): Q8_0, est. 6.6 tok/s
- [NVIDIA A100 80GB](https://selfhost.directory/models/h2oai/h2o-lightning-31b/on/a100-80gb): F16, est. 18 tok/s
- [Apple M4 Max 40-core (128 GB)](https://selfhost.directory/models/h2oai/h2o-lightning-31b/on/apple-m4-max-128gb-40c): F16, est. 4.8 tok/s

---
Model: https://selfhost.directory/models/h2oai/h2o-lightning-31b. GPU: https://selfhost.directory/models/gpus/rx-7900-xtx. Source: https://selfhost.directory/models/h2oai/h2o-lightning-31b/on/rx-7900-xtx