RTX 4090: what AI models can it run?
24 GB GDDR6X, 384-bit bus. Spec source
Sample price: $1800 (USD sample), not real offers.
Fits at Q4_K_M (16)
| Model | Q4 VRAM | Q8 VRAM |
|---|---|---|
| Llama 3.2 3B | 3.2 GB | ✅ 4.7 GB |
| Llama 3.1 8B | 6.4 GB | ✅ 10.4 GB |
| Qwen3 4B | 3.8 GB | ✅ 5.7 GB |
| Qwen3 8B | 6.5 GB | ✅ 10.6 GB |
| Qwen3 14B | 10.9 GB | ✅ 18.3 GB |
| Qwen3 32B | 22.8 GB | ❌ 39.3 GB |
| Qwen3 30B-A3B (MoE) | 21.5 GB | ❌ 36.8 GB |
| Qwen2.5-Coder 32B | 22.9 GB | ❌ 39.3 GB |
| Gemma 3 4B | 3.8 GB | ✅ 5.5 GB |
| Gemma 3 12B | 9 GB | ✅ 14.8 GB |
| Gemma 3 27B | 19.2 GB | ❌ 32.6 GB |
| Mistral 7B | 5.8 GB | ✅ 9.5 GB |
| Mistral Small 3.1 24B | 16.7 GB | ❌ 28.6 GB |
| Phi-4 14B | 11 GB | ✅ 18.2 GB |
| DeepSeek-R1-Distill-Qwen 14B | 10.9 GB | ✅ 18.3 GB |
| gpt-oss 20B | 16.2 GB | ✅ 16.2 GB |