Hardware → models
Intel Arc B580 12 GB
Start with Qwen3 14B at Q4_K_M for general chat. For a coding-focused option, try Qwen2.5-Coder 14B.
Use an Intel-compatible backend and driver. llama.cpp lists B580 in its SYCL documentation; check the exact runtime combination and GPU detection.
Models to start with
Planning estimates12 GB dedicated VRAM · 4,000 tokens · 32 GB system RAM · full GPU residency
| Model / starting file | Why consider it | GPU memory estimate | Memory left¹ |
|---|---|---|---|
| Qwen3 14BQ4_K_M | General chat with a larger modelUses more memory than 8B. A larger model is not proof of better answers for your task. | 10.2 GB | 1.4 GB |
| Qwen2.5-Coder 14BQ4_K_M | Coding with a larger modelMore memory than the 7B version. Compare them on your own codebase. | 10.2 GB | 1.4 GB |
| Qwen3 8BQ4_K_M | General chat and mixed tasksA middle-sized option with more memory headroom than the 14B models. | 6.2 GB | 5.4 GB |
| Llama 3.2 3B InstructQ4_K_M | Small text assistantA smaller weight footprint. Check the exact Llama checkpoint and licence. | 3.7 GB | 7.9 GB |
From the published model guides here; the Finder covers the full catalogue. ¹ Memory left after the Finder’s reserve. Other applications can use this space. Larger models and higher precision do not guarantee better answers.
Use an Intel-compatible backend and driver. llama.cpp lists B580 in its SYCL documentation; check the exact runtime combination and GPU detection. CPU offloading changes the memory allocation and may slow generation.
Setup and common questions
Which Arc card does this cover?
Intel’s desktop Arc B580 has 12 GB GDDR6 and uses Xe2. This page does not cover the B570 or an integrated Arc GPU. Check the exact product and dedicated memory when selecting hardware.
Can I use a CUDA download?
CUDA is NVIDIA’s backend. Choose a build supporting Intel instead. llama.cpp’s SYCL documentation lists Arc B580 among verified devices, but the correct driver and oneAPI/runtime combination still matter. Follow the current backend instructions for your operating system.
Why does the model load but respond slowly?
Check the log for GPU detection and offloaded layers before changing the model. CPU fallback can load the same file with very different performance. This page establishes planning memory fit only; no reviewed B580 speed measurements are available here.
How are these recommendations chosen?
We use the main Finder’s balanced task selection and memory calculation, limited to models with published guides here. Task scores and speed proxies are planning inputs, not measured quality rankings.
GPU memory includes weights, context cache and a 0.9 GB runtime allowance, plus a 3% reserve (0.15–1.5 GB). Cache format and runtime buffers can change the fit. See the calculation assumptions.
Sources and related guides
Specification sources checked 3 October 2026.
These comparisons are planning estimates. No reviewed or independently reproduced measurements are available yet. About test evidence.