Hardware comparison
NVIDIA RTX 3090 (24GB) vs NVIDIA RTX 4080 (16GB) for local LLMs
How the two stack up for running open-source LLMs locally — memory, bandwidth, price, and how many of the 69 tracked models each one can run.
Scroll sideways to compare both machines. Keyboard: focus the table, then use the arrow keys. Metric names stay visible.
| Hardware | NVIDIA RTX 3090 (24GB) | NVIDIA RTX 4080 (16GB) |
|---|---|---|
| Memory | 24.0 GB (better for this comparison) | 16.0 GB |
| Bandwidth | 936 GB/s (better for this comparison) | 717 GB/s |
| Price (approx) | $1,099 (better for this comparison)GPU only · host PC required, sold separately | $1,549GPU only · host PC required, sold separately |
| LLMs it runs | 51 of 69 (better for this comparison) | 48 of 69 |
| Best model it runs | Gemma 4 31BQ3_K_M · estimated 39–58 tok/s | Gemma 4 31BQ2_K · estimated 37–55 tok/s |
Both prices cover GPU cards only. A host PC is required and its cost is excluded. Prices are approximate catalog estimates; check current configurations and prices before buying.
The recommended quants differ. These estimated speeds are not a like-for-like comparison. Estimates use 4,096 context tokens, llama.cpp and one stream.
The NVIDIA RTX 3090 (24GB) runs 3 more of the tracked models (51 vs 48), thanks to its 24.0 GB of memory. The NVIDIA RTX 3090 (24GB) has more memory bandwidth (936 GB/s). Higher bandwidth can help generation speed, but model, quant, architecture and runtime also affect the estimates.