Cheapest Cloud GPU Right Now: Live Rates and Cost Checks

Right now the cheapest cloud GPU in our verified rental ledger is $0.10/hr for a RTX A4000 at TensorDock. That is the lowest listed GPU compute rate, not a promise about your final bill. It can come from interruptible or community capacity, and storage, networking or minimum terms may sit outside the hourly rate.

Cheapest depends on what the workload needs. Start with the smallest VRAM tier that fits, decide whether the job can restart, then compare the matching firm or interruptible offers in the live price grid. Use the checks below before choosing a provider.

The six levers that make GPU rental cheap

1. Use community and spot tiers. The single biggest saving is dropping off dedicated datacenter hardware. A community tier pools GPUs from independent hosts and prices the same card well below the secure equivalent. A spot or interruptible instance goes further, renting spare capacity at a discount in exchange for the provider being able to reclaim it. Both are fine when your job can checkpoint and resume: batch inference, rendering, fault-tolerant training. Neither is right for a live endpoint that cannot go down. Read the secure vs community trade-off before you switch.

2. Match billing granularity to runtime. Per-second billing matters for short or bursty work because unused fractions of an hour do not get rounded into a larger charge. RunPod documents per-second Pod billing; Lambda documents one-minute increments for On-Demand Cloud. Serverless GPU pricing is a separate product model and should be compared on worker runtime and idle behavior, not mixed into an instance table.

3. Right-size your VRAM with quantization. You do not need an 80GB card to run a model that fits in 24GB. Quantization stores a model's weights at lower precision (8-bit or 4-bit instead of 16-bit) and shrinks its VRAM footprint, often enough to move it from an expensive datacenter GPU down to a consumer card. An $1.99/hr H100 is the obvious pick for a big model, but a quantized 7B or 13B model runs happily on a $0.35/hr RTX 4090 at a fraction of the rate. The cheapest GPU is the smallest one your model actually fits on.

4. Check region availability before committing. Capacity, latency and transfer rules can differ by region. Our current offer ledger does not normalize region-level inventory or network transfer, so it does not calculate a region-adjusted winner. Confirm the deployable region and transfer policy at the provider source before moving a dataset or production endpoint.

5. Check measured price history. A current quote is easier to judge when you can see how it changed. The price history keeps the append-only daily series, the GPU Rental Price Index measures the like-for-like market level, and price movers shows weekly firm-rate changes. Use the measured series instead of assuming prices always move in one direction.

6. Check credits without confusing them with a free tier. Startup, research and trial credits can reduce a first bill, but they are temporary, conditional and sometimes application-only. The credits directory separates those programs from durable pricing. Compare the normal rate first so the provider still makes sense after a credit expires.

The traps that quietly raise your bill

The advertised GPU rate is only the compute line item. Four things can raise total cost:

• Egress fees. Moving data out of a network can be billed per gigabyte. The sourced fees matrix records free ingress and egress for RunPod and Lambda at its verification date, while other providers use different policies. Check that fees matrix and its linked provider source before deciding.
• Storage. Persistent, volume and network storage can be billed separately from compute and may continue charging while an instance is stopped. Capacity and retention time determine the amount.
• Minimum commitments. Reserved and cluster terms can lower the displayed per-hour rate while committing you to a longer billing window. If utilization is low, a minimum commitment can raise effective cost.
• Stale offers. A price without a source and verification date is not decision-grade. Every offer in our live grid links to its source, carries a fetch date and ages out after the documented freshness window.

Together these inputs determine effective cost per hour. The current grid sorts by the verified GPU compute rate only; it does not invent missing storage, network or minimum-charge values and does not claim to be a full-cost calculator.

Where to go next

Use the live price table to find the lowest verified compute rate for the GPU and reliability tier you need. Then inspect the provider profile and fees matrix for charges outside compute. If you are weighing a longer commitment or hardware purchase, the rent-vs-buy calculator shows its formula on the page. For server access, clusters and GPU-as-a-service terminology, use the GPU server guide.