Best GPU under $1,000 for local LLMs
The used GeForce RTX 3060 12 GB ($305). It is the cheapest per tok/s of the 11 priced GPUs at or under $1,000: about 48 tok/s estimate on NVIDIA Nemotron 3.5 Bolt E4B, or $6.35 per tok/s. For the most memory, 16 GB, the Radeon RX 9070 starts at $570.
Every card we have a dated price for at or under the cap, on one speed yardstick that fits all of them and one that needs their full memory.
Best value under $1,000: the NVIDIA GeForce RTX 3060 12 GB at $305 (used), ~48 tok/s estimate on NVIDIA Nemotron 3.5 Bolt E4B, the largest current model all 11 cards can hold: $6.35 per tok/s.
Most memory: 16 GB. The cheapest 16 GB card is the AMD Radeon RX 9070 at $570, which runs Mistral-Small-3.2-24B-Instruct-2506 (about 15.83 GB) at ~34 tok/s estimate.
Fastest under $1,000: the NVIDIA GeForce RTX 5070 (672 GB/s, 12 GB, $800 new). What $1,000 doesn't buy: more than 16 GB — the cheapest bigger card is the AMD Radeon RX 7900 XTX (24 GB) at $1,350, needed for models like K2-Horizon-MoVA-36B-A4B.
Last updated Oct 16, 2026. Speeds are bandwidth-based estimates, not benchmarks. Prices are dated snapshots and move daily; "No price yet" means we have no verified price.
Every priced GPU at or under $1,000
| GPU | VRAM | Bandwidth | Low price | NVIDIA Nemotron 3.5 Bolt E4B estimate | $ per tok/s | Mistral-Small-3.2-24B-Instruct-2506 estimate | Largest current model that fits |
|---|---|---|---|---|---|---|---|
| GeForce RTX 3060 12 GB | 12 GB | 360 GB/s | low $305 | ~48 tok/s | $6.35 | doesn't fit | phi-4 |
| GeForce RTX 5060 | 8 GB | 448 GB/s | low $410 | ~59 tok/s | $6.95 | doesn't fit | NVIDIA Nemotron 3.5 Bolt E4B |
| Radeon RX 9070 | 16 GB | 640 GB/s | low $570 | ~80 tok/s | $7.13 | ~34 tok/s | Mistral-Small-3.2-24B-Instruct-2506 |
| Arc B580 | 12 GB | 456 GB/s | low $330 | ~44 tok/s | $7.50 | doesn't fit | phi-4 |
| GeForce RTX 5060 Ti 8 GB | 8 GB | 448 GB/s | low $450 | ~59 tok/s | $7.63 | doesn't fit | NVIDIA Nemotron 3.5 Bolt E4B |
| Radeon RX 7800 XT | 16 GB | 624 GB/s | low $599 | ~78 tok/s | $7.68 | ~34 tok/s | Mistral-Small-3.2-24B-Instruct-2506 |
| GeForce RTX 5070 | 12 GB | 672 GB/s | low $800 | ~86 tok/s | $9.30 | doesn't fit | phi-4 |
| GeForce RTX 4070 SUPER | 12 GB | 504 GB/s | low $620 | ~66 tok/s | $9.39 | doesn't fit | phi-4 |
| Radeon RX 9070 XT | 16 GB | 640 GB/s | low $810 | ~80 tok/s | $10 | ~34 tok/s | Mistral-Small-3.2-24B-Instruct-2506 |
| GeForce RTX 5060 Ti 16 GB | 16 GB | 448 GB/s | low $780 | ~59 tok/s | $13 | ~27 tok/s | Mistral-Small-3.2-24B-Instruct-2506 |
| Radeon RX 9060 XT 16 GB | 16 GB | 320 GB/s | low $620 | ~40 tok/s | $16 | ~17 tok/s | Mistral-Small-3.2-24B-Instruct-2506 |
Sorted by $ per tok/s on NVIDIA Nemotron 3.5 Bolt E4B (6.18 GB at Q4_K_M), the largest current model that fits every card here. Mistral-Small-3.2-24B-Instruct-2506 (14.33 GB) needs about 15.83 GB, so only the 16 GB cards run it. "Used" prices are marketplace and refurbished listings.
Who this is for
You have a hard budget and want one card for chat or coding models up to the 16 GB class. If the models you care about need more (see 16 vs 24 vs 32 GB), stretching for the Radeon RX 7900 XTX will serve you better than the fastest card here. To check a specific model and card, use the tokens per second calculator; for a whole machine, see our priced builds under $2,000.
Questions people ask
What is the best GPU under $1,000 for local LLMs?
The used GeForce RTX 3060 12 GB ($305). It is the cheapest per tok/s of the 11 priced GPUs at or under $1,000: about 48 tok/s estimate on NVIDIA Nemotron 3.5 Bolt E4B, or $6.35 per tok/s. For the most memory, 16 GB, the Radeon RX 9070 starts at $570.
What is the most VRAM you can get for $1,000?
16 GB, among the cards we have a dated price for. That fits Mistral-Small-3.2-24B-Instruct-2506 (14.33 GB at Q4_K_M, about 15.83 GB needed), an estimated 34 tok/s on the AMD Radeon RX 9070. The next size up is the AMD Radeon RX 7900 XTX with 24 GB at $1,350, which fits K2-Horizon-MoVA-36B-A4B. estimate
Is an 8 GB card worth it for local AI?
Only for small models. The largest current model an 8 GB card holds is NVIDIA Nemotron 3.5 Bolt E4B (6.18 GB). The 8 GB cards here cost from $410, against $570 for the cheapest 16 GB card.
Related
- GPUWhat the NVIDIA GeForce RTX 3060 12 GB can run
- GPUWhat the AMD Radeon RX 9070 can run
- GPUWhat the NVIDIA GeForce RTX 5070 can run
- GPUWhat the AMD Radeon RX 7900 XTX can run
- ModelWhat GPU you need for NVIDIA Nemotron 3.5 Bolt E4B
- ModelWhat GPU you need for Mistral-Small-3.2-24B-Instruct-2506
- Can it runCan the Radeon RX 9070 run Mistral-Small-3.2-24B-Instruct-2506?
- Can it runCan the Radeon RX 9070 run gpt-oss-20b?
- Can it runCan the Radeon RX 7900 XTX run Qwen3.6-35B-A3B?
- Can it runCan the Radeon RX 7900 XTX run Meta Muse Glimmer 30B?
- Build16 GB starter: Radeon RX 9070$2,101.57, 16 GB
- Build16 GB NVIDIA: RTX 5060 Ti 16 GB$1,951.57, 16 GB
- Build24 GB: one used RTX 3090$2,650.58, 24 GB
- Build32 GB: two Radeon RX 9070$3,589.93, 32 GB
- GuideAMD vs NVIDIA GPUs for local AI in 2026
- GuideBest GPU for MoE (mixture-of-experts) LLMs
- GuideBest GPU for local LLMs (2026)