Best GPU for running LLMs locally under $300
For local models, VRAM decides what runs and memory bandwidth decides how fast. These are the 4 consumer GPUs in our table with a US launch MSRP of $300 or less, scored on both with the same formulas as the rest of the site. Prices are launch MSRPs, not today's street prices, which can be higher or lower.
The short answer
- Run the biggest models: the Arc B580 has 12 GB, and runs 150 of 298 open models we track fully in VRAM at 4-bit or better (8K context). The largest popular one is KORMo Web 11B (11B).
Every card under $300
Sorted by VRAM, then bandwidth. Model columns show the best quant that fits fully in VRAM at 8K context and an estimated speed ceiling, or the llama.cpp offload plan when the model only runs with part of it in system RAM (speed range from DDR4-3200 to DDR5-6000). How we calculate · offload method.
| GPU | Launch MSRP | VRAM | Bandwidth | Models at 4-bit+ | Qwen3 8B | gpt-oss 20B | Qwen3 32B | gpt-oss 120B |
|---|---|---|---|---|---|---|---|---|
| Arc B580 | $249 | 12 GB | 456 GB/s | 150 | Q8_0 · ~52 tok/s | Offload Q4_K_M, 32 GB RAM · ~72–98 tok/s | Offload Q4_K_M, 32 GB RAM · ~2.2–4.2 tok/s | Offload Q4_K_M, 64 GB RAM · ~13–25 tok/s |
| RTX 5060 | $299 | 8 GB | 448 GB/s | 135 | Q4_K_M · ~89 tok/s | Offload Q4_K_M, 32 GB RAM · ~29–52 tok/s | Offload Q4_K_M, 32 GB RAM · ~1.6–3.3 tok/s | Offload Q4_K_M, 64 GB RAM · ~12–24 tok/s |
| RX 7600 | $269 | 8 GB | 288 GB/s | 135 | Q4_K_M · ~57 tok/s | Offload Q4_K_M, 32 GB RAM · ~27–45 tok/s | Offload Q4_K_M, 32 GB RAM · ~1.6–3.2 tok/s | Offload Q4_K_M, 64 GB RAM · ~12–23 tok/s |
| RTX 4060 | $299 | 8 GB | 272 GB/s | 135 | Q4_K_M · ~54 tok/s | Offload Q4_K_M, 32 GB RAM · ~27–44 tok/s | Offload Q4_K_M, 32 GB RAM · ~1.6–3.2 tok/s | Offload Q4_K_M, 64 GB RAM · ~12–23 tok/s |
How to choose
Start from the model you want. A model needs roughly its 4-bit file size plus KV cache and about 1 GB of runtime overhead. Pick the cheapest card where it fits with room for your context length, then prefer higher bandwidth for faster replies. If a model is just out of reach, mixture-of-experts models such as gpt-oss and Qwen3 A3B can still run with most expert weights in system RAM at a usable speed; dense 70B models usually cannot.
By memory size: Best models for 8 GBBest models for 12 GBBest models for 16 GBBest models for 24 GBBest models for 32 GB
Other budgets: Under $500Under $800Under $1,000Under $1,500Under $2,000 · Mac vs PC · Can I run…?
Prices: NVIDIA, AMD and Intel US launch MSRPs (Founders/reference editions). Specs: vendor spec pages, linked on each GPU page. Speed is a bandwidth ceiling, not a benchmark; real speed is typically 50–80% of it. Data refreshes hourly. Report an error.