# Can the Arc Pro B60 run gemma-4-26B-A4B-it?

> Yes. The Arc Pro B60 runs gemma-4-26B-A4B-it fully in VRAM at Q5_K_M (about 20 GB of 24 GB), at roughly 88 tokens/s.

Arc Pro B60: 24 GB (24 GB usable), 456 GB/s. Estimates at an 8K context.

## Every quantisation

| Quant | Quality | ≈ GB | Fit | Est. tok/s |
|---|---|---|---|---|
| Q2_K | Heavy loss | 10.67 | fits | 191 |
| Q3_K_M | Noticeable loss | 14.71 | fits | 128 |
| Q4_K_M | Small loss | 17.67 | fits | 103 |
| Q5_K_M (pick) | Near-lossless | 20.32 | fits | 88 |
| Q6_K | Near-lossless | 23.16 | tight | 76 |
| Q8_0 | Lossless | 28.65 | offload | 28 |
| F16 | Lossless | 51.77 | no | — |

## Run it

From the Ollama library (ollama):

```bash
ollama run gemma4:26b
```

UD-Q5_K_M straight from Hugging Face (ollama):

```bash
ollama run hf.co/unsloth/gemma-4-26B-A4B-it-GGUF:UD-Q5_K_M
```

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

```bash
llama-server -hf unsloth/gemma-4-26B-A4B-it-GGUF:UD-Q5_K_M
```

Docker with Vulkan (llama.cpp):

```bash
docker run -d --name llama --device /dev/dri -p 8080:8080 \
  -v ~/.cache/llama.cpp:/root/.cache/llama.cpp \
  ghcr.io/ggml-org/llama.cpp:server-vulkan \
  -hf unsloth/gemma-4-26B-A4B-it-GGUF:UD-Q5_K_M --host 0.0.0.0 --port 8080 -ngl 99
```

Download with the lms CLI (lmstudio):

```bash
lms get unsloth/gemma-4-26B-A4B-it-GGUF@UD-Q5_K_M
```

Ask about an image (transformers):

```python
from transformers import pipeline

pipe = pipeline("image-text-to-text", model="google/gemma-4-26B-A4B-it", 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/arc-pro-b60): Q5_K_M ≈21.82 GB, est. 13 tok/s
- [Qwen/Qwen3.6-27B](https://selfhost.directory/models/qwen/qwen3.6-27b/on/arc-pro-b60): Q5_K_M ≈21.58 GB, est. 13 tok/s
- [google/gemma-3-27b-it](https://selfhost.directory/models/google/gemma-3-27b-it/on/arc-pro-b60): Q4_K_M ≈20.56 GB, est. 15 tok/s
- [Qwen/Qwen3.5-27B](https://selfhost.directory/models/qwen/qwen3.5-27b/on/arc-pro-b60): Q5_K_M ≈21.67 GB, est. 13 tok/s
- [meta-models/Muse-Glimmer-30B](https://selfhost.directory/models/meta-models/muse-glimmer-30b/on/arc-pro-b60): Q5_K_M ≈21.62 GB, est. 12 tok/s
- [Qwen/Qwen3.6-35B-A3B](https://selfhost.directory/models/qwen/qwen3.6-35b-a3b/on/arc-pro-b60): Q4_K_M ≈22.44 GB, est. 138 tok/s

## Other GPUs that run it

- [Tesla T4](https://selfhost.directory/models/google/gemma-4-26b-a4b-it/on/tesla-t4): Q3_K_M, est. 90 tok/s
- [Arc A770 16GB](https://selfhost.directory/models/google/gemma-4-26b-a4b-it/on/arc-a770): Q3_K_M, est. 158 tok/s
- [Radeon RX 7800 XT](https://selfhost.directory/models/google/gemma-4-26b-a4b-it/on/rx-7800-xt): Q3_K_M, est. 176 tok/s
- [Tesla P40](https://selfhost.directory/models/google/gemma-4-26b-a4b-it/on/tesla-p40): Q5_K_M, est. 67 tok/s
- [Radeon RX 7900 XTX](https://selfhost.directory/models/google/gemma-4-26b-a4b-it/on/rx-7900-xtx): Q5_K_M, est. 186 tok/s
- [Instinct MI50 32GB](https://selfhost.directory/models/google/gemma-4-26b-a4b-it/on/instinct-mi50): Q8_0, est. 133 tok/s
- [GeForce RTX 5090](https://selfhost.directory/models/google/gemma-4-26b-a4b-it/on/rtx-5090): Q8_0, est. 232 tok/s
- [Apple M4 Pro (48 GB)](https://selfhost.directory/models/google/gemma-4-26b-a4b-it/on/apple-m4-pro-48gb): Q8_0, est. 35 tok/s

---
Model: https://selfhost.directory/models/google/gemma-4-26b-a4b-it. GPU: https://selfhost.directory/models/gpus/arc-pro-b60. Source: https://selfhost.directory/models/google/gemma-4-26b-a4b-it/on/arc-pro-b60