PerGB

Cheapest GPUs to run Qwen3 32B locally

Qwen3 32B needs about 23 GB of VRAM at Q4 with an 8K-token context (37 GB at Q8). The cheapest way to hold it right now: Tesla P40 for $270.00, about 8.7–13 tokens/s.

The cheapest setups right now

Live on eBay
SetupVRAMSpeed (est.)Price nowPowerSoftware
Tesla P4024 GB0.5 GB spare8.7–13 tokens/s$270.00250 WCUDA 8.0 to 12.9 (not CUDA 13)
2 × Tesla P100 PCIe 16 GB32 GB7.5 GB spare18–28 tokens/s$345.00the 2 cheapest listings500 WCUDA 8.0 to 12.9 (not CUDA 13)
3 × Tesla P100 PCIe 12 GB36 GB10.5 GB spare14–21 tokens/s$386.99the 3 cheapest listings750 WCUDA 8.0 to 12.9 (not CUDA 13)
2 × Radeon VII32 GB7.5 GB spare18–31 tokens/s$503.00the 2 cheapest listings600 WNot on ROCm's support list (older ROCm or Vulkan)
2 × Instinct MI50 16 GB32 GB7.5 GB spare18–31 tokens/s$538.62the 2 cheapest listings600 WNot on ROCm's support list (older ROCm or Vulkan)
Instinct MI50 32 GB32 GB8.5 GB spare18–31 tokens/s$633.68300 WNot on ROCm's support list (older ROCm or Vulkan)
2 × Tesla V100 PCIe 16 GB32 GB7.5 GB spare23–34 tokens/s$659.00the 2 cheapest listings500 WCUDA 9.0 to 12.9 (not CUDA 13)
2 × Radeon Pro VII32 GB7.5 GB spare18–31 tokens/s$725.00the 2 cheapest listings500 WNot on ROCm's support list (older ROCm or Vulkan)
4 × GeForce RTX 2070 Super32 GB5.5 GB spare11–17 tokens/s$739.97the 4 cheapest listings860 WCUDA 10.0 and newer
4 × Quadro RTX 400032 GB5.5 GB spare10–16 tokens/s$744.96the 4 cheapest listings640 WCUDA 10.0 and newer
4 × GeForce RTX 207032 GB5.5 GB spare11–17 tokens/s$749.47the 4 cheapest listings700 WCUDA 10.0 and newer
2 × Radeon RX 680032 GB7.5 GB spare9.0–15 tokens/s$780.00the 2 cheapest listings500 WNot on ROCm's support list (older ROCm or Vulkan)
4 × GeForce RTX 2060 Super32 GB5.5 GB spare11–17 tokens/s$784.51the 4 cheapest listings700 WCUDA 10.0 and newer
4 × GeForce RTX 208032 GB5.5 GB spare11–17 tokens/s$795.00the 4 cheapest listings860 WCUDA 10.0 and newer
2 × Radeon RX 7600 XT32 GB7.5 GB spare5.1–8.7 tokens/s$799.98the 2 cheapest listings380 WNot on ROCm's support list (older ROCm or Vulkan)

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.