Cheapest GPUs to run Qwen3 235B-A22B locally
Qwen3 235B-A22B needs about 145 GB of VRAM at Q4 with an 8K-token context (252 GB at Q8). A mixture of experts: 22B active. The cheapest way to hold it right now: 3 × A16 for $6,789, about 18–27 tokens/s.
The cheapest setups right now
Live on eBay| Setup | VRAM | Speed (est.) | Price now | Power | Software |
|---|---|---|---|---|---|
| 3 × A16 | 192 GB44.5 GB spare | 18–27 tokens/s | $6,789the 3 cheapest listings | 750 W | CUDA 11.1 and newer |
| 4 × Quadro RTX 8000 | 192 GB43.5 GB spare | 15–23 tokens/s | $8,075the 4 cheapest listings | 1,040 W | CUDA 10.0 and newer |
| 4 × Arc Pro B60 48 GB | 192 GB43.5 GB spare | 7.2–12 tokens/s | $8,214the 4 cheapest listings | 800 W | No CUDA (oneAPI / Vulkan) |
| 3 × Instinct MI210 | 192 GB44.5 GB spare | 26–44 tokens/s | $11,398the 3 cheapest listings | 900 W | ROCm (on AMD's support list) |
| 4 × A40 | 192 GB43.5 GB spare | 16–23 tokens/s | $13,944the 4 cheapest listings | 1,200 W | CUDA 11.1 and newer |
| 4 × Radeon Pro W7900 | 192 GB43.5 GB spare | 14–23 tokens/s | $16,799the 4 cheapest listings | 1,180 W | ROCm (on AMD's support list) |
| 4 × RTX A6000 | 192 GB43.5 GB spare | 17–26 tokens/s | $16,848the 4 cheapest listings | 1,200 W | CUDA 11.1 and newer |
| 4 × GeForce RTX 4090 D 48 GB | 192 GB43.5 GB spare | 23–34 tokens/s | $20,389the 4 cheapest listings | 1,700 Wover a 15 A circuit | CUDA 11.8 and newer |
| 4 × A100 PCIe 40 GB | 160 GB11.5 GB spare | 35–52 tokens/s | $23,238the 4 cheapest listings | 1,200 W | CUDA 11.0 and newer |
| 2 × A100 PCIe 80 GB | 160 GB13.5 GB spare | 44–65 tokens/s | $23,499the 2 cheapest listings | 600 W | CUDA 11.0 and newer |
| 4 × GeForce RTX 4090 48 GB | 192 GB43.5 GB spare | 23–34 tokens/s | $23,868the 4 cheapest listings | 1,800 Wover a 15 A circuit | CUDA 11.8 and newer |
| 4 × L40 | 192 GB43.5 GB spare | 19–29 tokens/s | $26,597the 4 cheapest listings | 1,200 W | CUDA 11.8 and newer |
| 4 × RTX PRO 5000 Blackwell 48 GB | 192 GB43.5 GB spare | 30–45 tokens/s | $27,688the 4 cheapest listings | 1,200 W | CUDA 12.8 and newer |
| 3 × RTX PRO 5000 Blackwell 72 GB | 216 GB68.5 GB spare | 30–45 tokens/s | $27,744the 3 cheapest listings | 900 W | CUDA 12.8 and newer |
| 2 × RTX PRO 6000 Blackwell | 192 GB45.5 GB spare | 40–60 tokens/s | $28,180the 2 cheapest listings | 1,200 W | CUDA 12.8 and newer |
Per question: scaled from the model's reported speed on an RTX PRO 6000 by each card's compute and memory bandwidth, within about half either way. Licenses, scores and sources: decision models.
A set of cards is priced at that many of the cheapest live listings, shipping included. Speed: an estimate from memory bandwidth for one user chatting, within a third either way.
How it works
The memory a model needs is its weights at the quantisation you pick, plus the attention cache for the context and a little for working buffers. Q4 is what most people run at home, close to full quality at a third of the size. Each card keeps about 1 GB for itself, so two 24 GB cards hold 46 GB.
Writing each token reads every active weight once, so the speed is about the memory bandwidth divided by the model's size, times what the software gets out of the card (less off CUDA). Cards sharing a model layer by layer take turns, so a set runs about as fast as one card would. Treat it as a rough figure for one person chatting; a long prompt takes longer to read in.
Power is the cards' rated power. Over about 1,440 W with the rest of the PC, a set needs more than one 15 A circuit (or a 240 V one), and older cards often need a power cable adapter, which each card's page mentions.