Yes. The Arc A380 runs lumi-4b fully in VRAM at Q6_K (about 5.7 GB of 6.0 GB), at roughly 31 tokens/s.
- best quant
- Q6_K
- needs
- 5.7 GB
- usable
- 6.0 GB
- est. speed
- ~31 tok/s
Estimates for an 8K context; offload assumes 32 GB of system RAM.Try other contexts on the lumi-4b page →
Every quantisation of lumi-4b on the Arc A380
| Quant | Memory | Fit | Speed |
|---|---|---|---|
| Q2_K | 3.1 GB | fits | ~78 tok/s |
| Q3_K_M | 3.7 GB | fits | ~53 tok/s |
| Q4_K_M | 4.2 GB | fits | ~42 tok/s |
| Q5_K_M | 4.6 GB | fits | ~36 tok/s |
| Q6_K pick | 5.7 GB | fits | ~31 tok/s |
| Q8_0 | 5.9 GB | tight | ~24 tok/s |
| FP16 / BF16 | 9.6 GB | offload | ~7.2 tok/s |
Speed is a bandwidth-based estimate for generating text (186 GB/s); prompt processing and long contexts are slower.
Measured speed on the Arc A380
No community reports for this pair yet.
Report your own speed →Run lumi-4b on the Arc A380
Using Q6_K, the best fit for Arc A380.
llama-server -hf Mateus-Nery/lumi-4b-GGUF:Q6_KNeeds 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 Mateus-Nery/lumi-4b-GGUF:Q6_K --host 0.0.0.0 --port 8080 -ngl 99Server and web UI on port 8080.
Chat models that fit the Arc A380 comfortably
All that fit →| Model | Params | Best quant | Memory | Speed |
|---|---|---|---|---|
| 4B | Q6_K | ≈4.9 GBFits in memory | ~31 tok/s | |
| 3.8B | Q8_0 | ≈5.5 GBFits in memory | ~25 tok/s | |
| 4B | Q6_K | ≈4.9 GBFits in memory | ~31 tok/s | |
| 3.2B | Q8_0 | ≈4.7 GBFits in memory | ~30 tok/s | |
| 4.1B | Q6_K | ≈4.9 GBFits in memory | ~30 tok/s | |
| 4.1B | Q6_K | ≈4.9 GBFits in memory | ~30 tok/s |
Other GPUs that run lumi-4b
Frequently asked questions
- Can the Arc A380 run lumi-4b?
- Yes. The Arc A380 runs lumi-4b fully in VRAM at Q6_K (about 5.7 GB of 6.0 GB), at roughly 31 tokens/s.
- How fast does lumi-4b run on the Arc A380?
- An estimate from memory bandwidth gives about 31 tokens/s at Q6_K.
- Which quantisation of lumi-4b should I use on the Arc A380?
- Q6_K: 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.