Cheapest GPUs to rent — budget picks that still deliver

Budget GPUs on GPUniq rent from about $0.05/hr. The sweet spot is the RTX 3060 12GB: enough VRAM for Stable Diffusion, 7B-parameter LLMs in 4-bit, and most coursework, at rock-bottom prices. Pay a little more for an RTX 3090 (24GB) when model size — not speed — is what blocks you.

Prices below are live minimum / median USD per hour on the GPUniq marketplace, as of 2026-07-21 00:37 UTC. Refreshed every 5 minutes.

#GPUVRAMMin $/hrMedian $/hrAvailableWhy this card
1RTX 306012GB GDDR6$0.05$0.0938912GB for pocket change — Stable Diffusion, small LLMs, learning CUDA.
2RTX 40608GB GDDR6$0.08$0.1035Ada efficiency and DLSS-era drivers at entry-level prices; 8GB VRAM.
3RTX 3060 Ti8GB GDDR6$0.07$0.1364Noticeably faster than the 3060 when 8GB VRAM is enough.
4GTX 1660 Super6GB GDDR6$0.08$0.0815The absolute floor for CUDA experiments and classic ML.
5RTX 309024GB GDDR6X$0.15$0.221132The "budget 24GB" — when VRAM, not budget, is the real constraint.

Source: GPUniq live marketplace analytics (2026-07-21 00:37 UTC). Each linked page shows price history for 24h/7d/30d/365d.

Frequently asked questions

What is the cheapest GPU I can rent for Stable Diffusion?

An RTX 3060 12GB comfortably runs SDXL and Flux-schnell inference. On GPUniq it rents from roughly $0.05–0.10/hr — see the live minimum price in the table above; it refreshes every 5 minutes.

Can I run an LLM on a budget GPU?

Yes — 7B models in 4-bit quantization fit in 6–8GB and run fine on an RTX 3060 or even a GTX 1660 Super. For 13B in 4-bit you want 12GB+ (RTX 3060 12GB), and for 24GB the cheapest ticket is an RTX 3090.

Why are these GPUs so cheap to rent?

GPUniq is a marketplace: thousands of consumer cards compete on price, and per-second billing means you pay only for actual compute. Older and mid-range models are abundant, which keeps their market price near the electricity floor.

More GPU guides