The going rate for AI hardware.
Tracking 3,485 active listings across 22 GPUs, last updated 2026-09-09. Medians and ranges are asking prices, not sale prices. Auctions, scams, and statistical outliers are filtered per our methodology.
Consumer 17 GPUs
Most wanted 5 GPUs
| GPU | VRAM | Going rate (used) |
|---|---|---|
| NVIDIA GeForce RTX 3090 | 24 GB | $1,500 |
| NVIDIA GeForce RTX 5090 | 32 GB | $5,500 |
| NVIDIA GeForce RTX 4090 | 24 GB | $3,053 |
| NVIDIA GeForce RTX 3090 Ti | 24 GB | $1,500 |
| AMD Radeon RX 7900 XTX | 24 GB | $1,000 |
More consumer GPUs 12 GPUs
| GPU | VRAM | Going rate (used) |
|---|---|---|
| NVIDIA GeForce RTX 5060 Ti 16GB | 16 GB | $700 |
| NVIDIA GeForce RTX 5070 Ti | 16 GB | $1,300 |
| NVIDIA GeForce RTX 5080 | 16 GB | $1,625 |
| Intel Arc B580 | 12 GB | $310 |
| NVIDIA GeForce RTX 4070 Ti SUPER | 16 GB | $847 |
| NVIDIA GeForce RTX 4080 SUPER | 16 GB | $1,300 |
| NVIDIA GeForce RTX 4060 Ti 16GB | 16 GB | – |
| Intel Arc A770 16GB | 16 GB | $346 |
| NVIDIA GeForce RTX 3080 12GB | 12 GB | $485 |
| NVIDIA GeForce RTX 4080 | 16 GB | $1,200 |
| NVIDIA GeForce RTX 3060 12GB | 12 GB | $338 |
| NVIDIA GeForce RTX 3080 10GB | 10 GB | $440 |
Workstation 4 GPUs
| GPU | VRAM | Going rate (used) |
|---|---|---|
| NVIDIA RTX A4000 | 16 GB | $1,000 |
| NVIDIA RTX A5000 | 24 GB | $2,499 |
| NVIDIA RTX A6000 | 48 GB | $5,042 |
| NVIDIA TITAN RTX | 24 GB | $1,050 |
Datacenter 1 GPU
| GPU | VRAM | Going rate (used) |
|---|---|---|
| NVIDIA Tesla P40 | 24 GB | $325 |
What can it run? 11 models
Best used hardware per model, precomputed from VRAM fit and current asking prices.
- Qwen3.8 27B 27.0B · Dense The new flagship local model with native image input Our pick 1× NVIDIA GeForce RTX 3090 · $1,500 · ~28.4 tok/s est. Renting wins at 40 h/mo
- Qwen3.6 27B 27.0B · Dense The previous Qwen flagship and still a top-tier local model Our pick 1× NVIDIA GeForce RTX 3090 · $1,500 · ~26 tok/s est. Renting wins at 40 h/mo
- Qwen3.6 35B-A3B 35.0B · MoE Fast all-round assistant Our pick 1× NVIDIA RTX A6000 · $4,270 · ~201 tok/s est. Renting wins at 40 h/mo
- Gemma 4 31B 30.7B · Dense Among the strongest consumer-grade local models Our pick 1× NVIDIA GeForce RTX 3090 · $1,500 · ~25.6 tok/s est. Renting wins at 40 h/mo
- Gemma 4 26B-A4B 25.2B · MoE Fast everyday assistant Our pick 1× NVIDIA GeForce RTX 3090 · $1,500 · ~174 tok/s est. Renting wins at 40 h/mo
- GLM-4.7 Flash 30B-A3B 30.0B · MoE Quick chat and agent tool use Our pick 1× NVIDIA GeForce RTX 3090 · $1,500 · ~256 tok/s est. Renting wins at 40 h/mo
- Gemma 4 12B 12.0B · Dense Light assistant for mid-range GPUs Our pick 1× NVIDIA GeForce RTX 3060 12GB · $305 · ~23.5 tok/s est. Renting wins at 40 h/mo
- Qwen3.5 9B 9.0B · Dense Starter model for modest hardware Our pick 1× NVIDIA GeForce RTX 3060 12GB · $305 · ~29.2 tok/s est. Renting wins at 40 h/mo
- Devstral Small 2 24B 24.0B · Dense Coding assistant and code agents Our pick 1× NVIDIA GeForce RTX 3090 · $1,500 · ~32.7 tok/s est. Renting wins at 40 h/mo
- Qwen3-Coder-Next 80B-A3B 80.0B · MoE Heavyweight coding assistant Our pick 4× NVIDIA GeForce RTX 3090 · $6,050 · ~258 tok/s est.
- Nemotron 3 Super 120B-A12B 121.0B · MoE Deep reasoning and long-form answers Our pick 4× NVIDIA GeForce RTX 3090 · $6,050 · ~54.3 tok/s est.