Yes. The Arc A580 runs WeVisCut fully in VRAM at Q4_K_M (about 7.0 GB of 8.0 GB), at roughly 49 tokens/s.
- best quant
- Q4_K_M
- needs
- 7.0 GB
- usable
- 8.0 GB
- est. speed
- ~49 tok/s
Estimates for an 8K context; offload assumes 32 GB of system RAM.Try other contexts on the WeVisCut page →
Every quantisation of WeVisCut on the Arc A580
| Quant | Memory | Fit | Speed |
|---|---|---|---|
| Q2_K | 4.5 GB | fits | ~91 tok/s |
| Q3_K_M | 6.0 GB | fits | ~61 tok/s |
| Q4_K_M pick | 7.0 GB | fits | ~49 tok/s |
| Q5_K_M | 7.8 GB | tight | ~42 tok/s |
| Q6_K | 8.7 GB | offload | ~23 tok/s |
| Q8_0 | 11 GB | offload | ~9.5 tok/s |
| FP16 / BF16 | 19 GB | offload | ~2.8 tok/s |
Speed is a bandwidth-based estimate for generating text (512 GB/s); prompt processing and long contexts are slower.
Measured speed on the Arc A580
No community reports for this pair yet.
Report your own speed →Run WeVisCut on the Arc A580
Using Q4_K_M, the best fit for Arc A580.
llama-server -hf prithivMLmods/tencent-WeVisCut-GGUF:Q4_K_MNeeds 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 prithivMLmods/tencent-WeVisCut-GGUF:Q4_K_M --host 0.0.0.0 --port 8080 -ngl 99Server and web UI on port 8080.
Omni models that fit the Arc A580 comfortably
All that fit →| Model | Params | Best quant | Memory | Speed |
|---|---|---|---|---|
| 9.4B | Q5_K_M | ≈7.3 GBFits in memory | ~42 tok/s | |
| 8B | Q5_K_M | ≈6.5 GBFits in memory | ~50 tok/s | |
| 10.7B | Q6_K | ≈7.0 GBFits in memory | ~32 tok/s | |
| 9.1B | Q4_K_M | ≈7.0 GBFits in memory | ~51 tok/s | |
| 8.3B | Q6_K | ≈7.6 GBFits in memory | ~41 tok/s | |
| 8.7B | Q6_K | ≈7.0 GBFits in memory | ~39 tok/s |
Other GPUs that run WeVisCut
Frequently asked questions
- Can the Arc A580 run WeVisCut?
- Yes. The Arc A580 runs WeVisCut fully in VRAM at Q4_K_M (about 7.0 GB of 8.0 GB), at roughly 49 tokens/s.
- How fast does WeVisCut run on the Arc A580?
- An estimate from memory bandwidth gives about 49 tokens/s at Q4_K_M.
- Which quantisation of WeVisCut should I use on the Arc A580?
- Q4_K_M: the largest one that fits. Expected quality: small loss 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.