Local AI rig: 24 GB GPU build
24 GB is the sweet spot for local AI: it runs 27B to 32B models at 4-bit with room for a real conversation. See exactly what fits on our 24 GB tier page.
We don't show prices here because we only print prices we checked the same day, and Amazon prices change often. Click through to see today's price.
Parts list
| Part | Pick | Why | Link |
|---|---|---|---|
| GPU | GeForce RTX 3090 (24 GB) | 24 GB GDDR6X on a 384-bit bus, 350 W graphics card power (NVIDIA). Discontinued, so listings may be used or renewed. | Check price on Amazon |
| GPU (newer option) | GeForce RTX 4090 (24 GB) | 24 GB GDDR6X, 384-bit, 450 W. NVIDIA lists an 850 W power supply (NVIDIA). | Check price on Amazon |
| System RAM | 64 GB DDR5 (2×32 GB) | Lets you offload MoE experts or extra layers to the CPU for models a bit bigger than 24 GB. See MoE expert offload. | Check price on Amazon |
| Storage | 2 TB NVMe SSD | 32B models at Q4 are large downloads, and you'll keep several. | Check price on Amazon |
| Power supply | 850 W or larger ATX PSU | Matches NVIDIA's listed 850 W for the RTX 4090. | Check price on Amazon |
| CPU + motherboard | Any current 8-core desktop CPU with a matching board | Helps when part of a model runs on the CPU. | Check price on Amazon |
Amazon links are affiliate links: we may earn a commission at no cost to you. Disclosure.
Next steps
- Compare 24 GB cards in our 24 GB buying guide and see RTX 3090 vs RTX 4090.
- Check your exact model in the VRAM calculator.
- Other builds: 16 GB budget rig · 32 GB rig.
- Only need it a few hours a month? Renting a GPU may be cheaper.