Best GPU for local LLMs under $400
Every graphics card we track that launched at $400 or less, ranked by how many local AI models it runs well, then by speed. Our top pick is the Intel Arc A770 16GB.
Every card under $400, ranked
Speed columns are estimated tokens per second at the best quant that fits, 8K context.
| GPU | VRAM | Launch price | Runs well | Largest model | Llama 3.1 8B | Qwen3 14B | Qwen3 32B | |
|---|---|---|---|---|---|---|---|---|
 Intel Arc A770 16GB | 16 GB | $349 | 28/69 | gpt-oss 20B | ~39 tok/s | ~28 tok/s | tight | Prices |
 Radeon RX 9060 XT 16GB | 16 GB | $349 | 28/69 | gpt-oss 20B | ~23 tok/s | ~16 tok/s | tight | Prices |
| Radeon RX 7600 XT | 16 GB | $329 | 28/69 | gpt-oss 20B | ~20 tok/s | ~14 tok/s | tight | Prices |
| Intel Arc B580 | 12 GB | $249 | 27/69 | Qwen3 14B | ~32 tok/s | ~30 tok/s | too big | Prices |
 RTX 3060 12GB | 12 GB | $329 | 27/69 | Qwen3 14B | ~25 tok/s | ~24 tok/s | too big | Prices |
| RTX 5060 | 8 GB | $299 | 17/69 | Qwen3 8B | ~47 tok/s | tight | too big | Prices |
| RTX 5060 Ti 8GB | 8 GB | $379 | 17/69 | Qwen3 8B | ~47 tok/s | tight | too big | Prices |
| RTX 3060 Ti | 8 GB | $399 | 17/69 | Qwen3 8B | ~47 tok/s | tight | too big | Prices |
| RTX 4060 Ti 8GB | 8 GB | $399 | 17/69 | Qwen3 8B | ~30 tok/s | tight | too big | Prices |
| RTX 4060 | 8 GB | $299 | 17/69 | Qwen3 8B | ~29 tok/s | tight | too big | Prices |
Prices are US launch prices (MSRP) for comparison; street prices change daily. Links go to Amazon, where we earn a commission at no cost to you.
Frequently asked questions
What is the best GPU for local LLMs under $400?
The Intel Arc A770 16GB (16 GB, $349 at launch). It runs 28 of the 69 models we track at Q4_K_M or better, up to gpt-oss 20B.
Does VRAM or speed matter more?
VRAM first: it decides which models fit at all. Memory bandwidth then sets how many tokens per second you get. A slower card with more VRAM usually beats a faster card that cannot fit the model.
Is a used RTX 3090 worth it?
For local AI it often is: 24 GB of VRAM and high bandwidth for well under its launch price on the used market. Check the seller, the card's condition and that your power supply can handle it.
Run it in the cloud
Not ready to buy? Try it on a rented GPU first.
Rent the card you are considering by the hour and see how your models actually run.