For local LLMs, buy VRAM first, bandwidth second. A card that holds your model entirely beats a faster card that has to offload.
VRAM and bus width from NVIDIA spec pages; bandwidth from Wikipedia series tables (not listed on NVIDIA pages); MSRPs from NVIDIA press releases. Current street prices: TBD.
| Budget band (launch MSRP) | VRAM options | Fits at Q4 (approx.) |
|---|---|---|
| Under $450 | RTX 5060 Ti 16GB | Up to ~14B, gpt-oss-20b |
| $750–1,000 | RTX 5070 Ti / 5080 (16 GB) | Same capacity, more bandwidth |
| Used 24 GB | RTX 3090 / 4090 | 27–32B dense, 30B MoE |
| $1,999 | RTX 5090 (32 GB) | 32B at higher quality, more context |
Fits are our estimates; ranking is by spec, not by our own benchmarks (TBD).
Check a specific model against your hardware with Can My PC Run It?, or get a parts shortlist from the AI PC Builder.