Yes. The Arc B580 runs lumi-4b fully in VRAM at FP16 / BF16 (about 9.6 GB of 12 GB), at roughly 32 tokens/s.
- best quant
- FP16 / BF16
- needs
- 9.6 GB
- usable
- 12 GB
- est. speed
- ~32 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 B580
| Quant | Memory | Fit | Speed |
|---|---|---|---|
| Q2_K | 3.1 GB | fits | ~192 tok/s |
| Q3_K_M | 3.7 GB | fits | ~129 tok/s |
| Q4_K_M | 4.2 GB | fits | ~104 tok/s |
| Q5_K_M | 4.6 GB | fits | ~89 tok/s |
| Q6_K | 5.7 GB | fits | ~77 tok/s |
| Q8_0 | 5.9 GB | fits | ~59 tok/s |
| FP16 / BF16 pick | 9.6 GB | fits | ~32 tok/s |
Speed is a bandwidth-based estimate for generating text (456 GB/s); prompt processing and long contexts are slower.
Measured speed on the Arc B580
No community reports for this pair yet.
Report your own speed →Run lumi-4b on the Arc B580
Using FP16 / BF16, the best fit for Arc B580.
llama-server -hf Mateus-Nery/lumi-4b-GGUF:BF16Needs 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:BF16 --host 0.0.0.0 --port 8080 -ngl 99Server and web UI on port 8080.
Chat models that fit the Arc B580 comfortably
All that fit →| Model | Params | Best quant | Memory | Speed |
|---|---|---|---|---|
| 4B | Q8_0 | ≈5.8 GBFits in memory | ~59 tok/s | |
| 3.2B | Q8_0 | ≈4.7 GBFits in memory | ~73 tok/s | |
| 3.8B | Q8_0 | ≈5.5 GBFits in memory | ~62 tok/s | |
| 4B | Q8_0 | ≈5.8 GBFits in memory | ~59 tok/s | |
| 4.1B | Q8_0 | ≈5.9 GBFits in memory | ~57 tok/s | |
| 3.1B | Q8_0 | ≈4.3 GBFits in memory | ~76 tok/s |
Other GPUs that run lumi-4b
Frequently asked questions
- Can the Arc B580 run lumi-4b?
- Yes. The Arc B580 runs lumi-4b fully in VRAM at FP16 / BF16 (about 9.6 GB of 12 GB), at roughly 32 tokens/s.
- How fast does lumi-4b run on the Arc B580?
- An estimate from memory bandwidth gives about 32 tokens/s at FP16 / BF16.
- Which quantisation of lumi-4b should I use on the Arc B580?
- FP16 / BF16: 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.