GPU comparison
Mac Studio M3 Ultra (96GB) vs NVIDIA RTX 3090 (24GB) for local LLMs
How the two stack up for running open-source LLMs locally — memory, bandwidth, price, and how many of the 34 tracked models each one can run.
| Mac Studio M3 Ultra (96GB) | NVIDIA RTX 3090 (24GB) | |
|---|---|---|
| Memory | 96.0 GB | 24.0 GB |
| Bandwidth | 819 GB/s | 936 GB/s |
| Price (approx) | $5,299 | $1,099 |
| LLMs it runs | 30 of 34 | 25 of 34 |
| Best model it runs | Gemma 4 31B · 15–22 tok/s | Gemma 4 31B · 39–58 tok/s |
The Mac Studio M3 Ultra (96GB) runs 5 more of the tracked models (30 vs 25), thanks to its 96.0 GB of memory. The NVIDIA RTX 3090 (24GB) has more memory bandwidth (936 GB/s), so it generates tokens faster at the same model and quant.