selfhost.directory

The index of self-hostable projects — ranked, categorized and served with live updates.
Own your data and stop giving it to big tech.

ibm-granite logo

Can the Arc A310 run granite-4.1-3b-llmon-preview?

granite-4.1-3b-llmon-preview·3.7B·Arc A310·4 GB VRAM

Saved in this browser, used on every model page.
Yes

Yes. The Arc A310 runs granite-4.1-3b-llmon-preview fully in VRAM at Q5_K_M (about 3.7 GB of 4.0 GB), at roughly 26 tokens/s.

best quant
Q5_K_M
needs
3.7 GB
usable
4.0 GB
est. speed
~26 tok/s

Estimates for an 8K context; offload assumes 32 GB of system RAM.Try other contexts on the granite-4.1-3b-llmon-preview page →

Every quantisation of granite-4.1-3b-llmon-preview on the Arc A310

QuantQualityMemoryFitSpeed
Q2_K Heavy loss2.6 GBfits~57 tok/s
Q3_K_M Noticeable loss2.9 GBfits~38 tok/s
Q4_K_M Small loss3.3 GBfits~31 tok/s
Q5_K_M pickNear-lossless3.7 GBfits~26 tok/s
Q6_K Near-lossless4.1 GBoffload~22 tok/s
Q8_0 Lossless4.9 GBoffload~15 tok/s
FP16 / BF16 Lossless8.3 GBoffload~6 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 granite-4.1-3b-llmon-preview on the Arc A310

Using Q5_K_M, the best fit for Arc A310.

llama.cpp · OpenAI-compatible server with a web UIdocs
llama-server -hf liodon-ai/granite-4.1-3b-llmon-preview-GGUF:Q5_K_M

Needs a Vulkan build of llama.cpp. Downloads the GGUF on first run.

llama.cpp · Docker with Vulkandocs
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 liodon-ai/granite-4.1-3b-llmon-preview-GGUF:Q5_K_M --host 0.0.0.0 --port 8080 -ngl 99

Server and web UI on port 8080.

Transformers · Chat with the pipeline APIdocs
from transformers import pipeline

pipe = pipeline("text-generation", model="ibm-granite/granite-4.1-3b-llmon-preview", dtype="auto", device_map="auto")
messages = [{"role": "user", "content": "Explain self-hosting in one paragraph."}]
out = pipe(messages, max_new_tokens=512)
print(out[0]["generated_text"][-1]["content"])

Chat models that fit the Arc A310 comfortably

All that fit →
ModelParamsBest quantMemorySpeedLicence
3.2BQ5_K_M≈3.6 GBFits in memory~30 tok/sConditions
ibm-granite logogranite-4.2-3bibm-granite
3.7BQ5_K_M≈3.7 GBFits in memory~26 tok/sCommercial OK
3.1BQ6_K≈3.5 GBFits in memory~27 tok/sNon-commercial
ibm-granite logogranite-4.1-3bibm-granite
3.4BQ5_K_M≈3.5 GBFits in memory~28 tok/sCommercial OK
HuggingFaceTB logoSmolLM3-3BHuggingFaceTB
3.1BQ6_K≈3.5 GBFits in memory~27 tok/sCommercial OK
2.6BQ6_K≈3.4 GBFits in memory~32 tok/sConditions

Other GPUs that run granite-4.1-3b-llmon-preview

Frequently asked questions

Can the Arc A310 run granite-4.1-3b-llmon-preview?
Yes. The Arc A310 runs granite-4.1-3b-llmon-preview fully in VRAM at Q5_K_M (about 3.7 GB of 4.0 GB), at roughly 26 tokens/s.
How fast does granite-4.1-3b-llmon-preview run on the Arc A310?
An estimate from memory bandwidth gives about 26 tokens/s at Q5_K_M.
Which quantisation of granite-4.1-3b-llmon-preview should I use on the Arc A310?
Q5_K_M: the largest one that fits. Expected quality: near-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.