GPU comparison
Mac Studio M3 Ultra (96GB) vs Intel Arc B580 (12GB) for local LLMs
How the two stack up for running open-source LLMs locally — memory, bandwidth, price, and how many of the 48 tracked models each one can run.
| Mac Studio M3 Ultra (96GB) | Intel Arc B580 (12GB) | |
|---|---|---|
| Memory | 96.0 GB | 12.0 GB |
| Bandwidth | 819 GB/s | 456 GB/s |
| Price (approx) | $5,299 | $289 |
| LLMs it runs | 41 of 48 | 19 of 48 |
| Best model it runs | Gemma 4 31B · 15–22 tok/s | Devstral Small 2 24B · 32–48 tok/s |
The Mac Studio M3 Ultra (96GB) runs 22 more of the tracked models (41 vs 19), thanks to its 96.0 GB of memory. The Mac Studio M3 Ultra (96GB) has more memory bandwidth (819 GB/s), so it generates tokens faster at the same model and quant.