Yes. The Arc A310 runs d1-2B fully in VRAM at Q8_0 (about 2.7 GB of 4.0 GB), at roughly 33 tokens/s.
- best quant
- Q8_0
- needs
- 2.7 GB
- usable
- 4.0 GB
- est. speed
- ~33 tok/s
Estimates for an 8K context; offload assumes 32 GB of system RAM.Try other contexts on the d1-2B page →
Every quantisation of d1-2B on the Arc A310
| Quant | Memory | Fit | Speed |
|---|---|---|---|
| Q2_K | 1.5 GB | fits | ~108 tok/s |
| Q3_K_M | 1.8 GB | fits | ~73 tok/s |
| Q4_K_M | 1.9 GB | fits | ~59 tok/s |
| Q5_K_M | 2.1 GB | fits | ~50 tok/s |
| Q6_K | 2.4 GB | fits | ~43 tok/s |
| Q8_0 pick | 2.7 GB | fits | ~33 tok/s |
| FP16 / BF16 | 4.3 GB | offload | ~17 tok/s |
Speed is a bandwidth-based estimate for generating text (124 GB/s); prompt processing and long contexts are slower.
Measured speed on the Arc A310
No community reports for this pair yet.
Report your own speed →Run d1-2B on the Arc A310
Using Q8_0, the best fit for Arc A310.
ollama run d1:2bThe library build picks its own default quantisation. Ollama’s ROCm/CUDA builds don’t cover this GPU. Recent Ollama releases use Vulkan when its Vulkan backend and driver are installed (older ones needed OLLAMA_VULKAN=1); llama.cpp or LM Studio with Vulkan are the safer bet.
ollama run hf.co/LiquidAI/d1-2B-GGUF:Q8_0Ollama’s ROCm/CUDA builds don’t cover this GPU. Recent Ollama releases use Vulkan when its Vulkan backend and driver are installed (older ones needed OLLAMA_VULKAN=1); llama.cpp or LM Studio with Vulkan are the safer bet.
llama-server -hf LiquidAI/d1-2B-GGUF:Q8_0Needs a Vulkan build of llama.cpp. Downloads the GGUF on first run.
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 LiquidAI/d1-2B-GGUF:Q8_0 --host 0.0.0.0 --port 8080 -ngl 99Server and web UI on port 8080.
lms get LiquidAI/d1-2B-GGUF@Q8_0Then load it from the LM Studio app or with `lms load`.
from transformers import pipeline
pipe = pipeline("image-text-to-text", model="LiquidAI/d1-2B", 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 the Arc A310 comfortably
All that fit →| Model | Params | Best quant | Memory | Speed |
|---|---|---|---|---|
| 1.9B | Q8_0 | ≈2.7 GBFits in memory | ~33 tok/s | |
| 2.3B | Q8_0 | ≈2.8 GBFits in memory | ~28 tok/s | |
| 2.1B | Q8_0 | ≈3.2 GBFits in memory | ~30 tok/s | |
| 1.9B | Q8_0 | ≈3.0 GBFits in memory | ~33 tok/s | |
| 2.2B | Q8_0 | ≈2.6 GBFits in memory | ~29 tok/s | |
| 1.6B | Q8_0 | ≈1.9 GBFits in memory | ~41 tok/s |
Other GPUs that run d1-2B
Frequently asked questions
- Can the Arc A310 run d1-2B?
- Yes. The Arc A310 runs d1-2B fully in VRAM at Q8_0 (about 2.7 GB of 4.0 GB), at roughly 33 tokens/s.
- How fast does d1-2B run on the Arc A310?
- An estimate from memory bandwidth gives about 33 tokens/s at Q8_0.
- Which quantisation of d1-2B should I use on the Arc A310?
- Q8_0: the largest one that fits. Expected quality: lossless compared with the full-precision weights.
Memory and speed are estimates from the model's size and the card's memory bandwidth; real numbers depend on the runtime, drivers and context length. Community figures are medians of user reports.