# GeForce RTX 3090 (used) for local AI

URL: https://insidethemachine.ai/machines/rtx-3090
Type: NVIDIA graphics card. Chip: RTX 3090 (used). Memory: vram, options 24 GB. Bandwidth: 936 GB/s. Spec source: https://en.wikipedia.org/wiki/List_of_Nvidia_graphics_processing_units (2026-10-15).
Price: from about $1,430 (2026-10-02, approximate). Status: used.

## Which models fit

| memory GB | usable GB | models that fit | best 3 |
|---|---|---|---|
| 24 | 23 | 56 | Qwen3.8 27B (22.7 GB), Qwen3.5 27B (21.5 GB), Qwen3.6 27B (21.5 GB) |

## Published speeds (not our tests)

- gpt-oss-20b: 147.53 tok/s decode, 4400.3 tok/s prompt (llama.cpp (Linux, flash attention), MXFP4, , community-measured) https://www.hardware-corner.net/guides/gpt-oss-20b-gpu-benchamrks/ (2025-10-26)

Notes: eBay median of 35 clean used sales (range $950–$1,799). The classic budget 24GB card; NVLink-capable.

## Questions

**What is the best local LLM for the GeForce RTX 3090 (used)?**
With 24 GB, Qwen3.8 27B (needs 22.7 GB, AA 33.7). 56 of 124 models fit.

**How fast is the GeForce RTX 3090 (used) for local AI?**
Published: gpt-oss-20b at 147.53 tokens/sec (llama.cpp (Linux, flash attention), ). These are other people's measurements, not ours.

**How much does the GeForce RTX 3090 (used) cost?**
From about $1,430 (as of 2026-10-02). 2026 prices are inflated by the memory-chip shortage; check before buying.