How Much VRAM Do I Need for LLM Training and Inference?
A practical engineer's guide to VRAM for LLM training, fine-tuning, and inference. Learn how to estimate VRAM for 7B and 70B...
2026-10-25 Read guideVendors we track
10 GPU models tracked
Vast.ai GPU pricing
Vast.ai is a marketplace: independent hosts list their own machines and you bid for them. That makes it the cheapest option for interruptible work and for older GPUs, at the cost of more variable reliability and a search-and-bid workflow instead of a fixed catalogue.
Cheapest single GPU right now
$0.1081
Per GPU per hour, on-demand, across all tracked models.
- GPU listings
- 10
- Spot capacity
- Yes
- NVLink
- No
- Price data last updated
- 2026-10-26
Pros
- Usually the lowest hourly prices in the market
- Very wide hardware selection including older and consumer GPUs
- Interruptible (spot-like) bidding and on-demand rental side by side
- Per-second billing
Cons
- Host quality, disk speed and network vary a lot
- You must vet hosts; there is no SLA comparable to a single-vendor cloud
- Images and setup are your responsibility
30-day price history
Each point is a daily snapshot we collected ourselves. Solid line = on-demand, dashed = spot.
On-demand
Spot / interruptible
Source: our own daily collection snapshots
Egress & storage
Storage and bandwidth are priced by each host and shown on the offer card. Check the host's rate before committing to a large transfer.
Live GPU price comparison
Filter by provider, GPU or region — filters run locally in your browser.
| Provider | GPU model | VRAM | GPUs | vCPU | Region | On-demand /hr | Spot /hr | Actions |
|---|---|---|---|---|---|---|---|---|
| Vast.ai | RTX 4090 | 24 GB | 1 | — | Distributed (global hosts) | $0.1081 Best price | $0.0533 | Rent this GPU |
| Vast.ai | V100 | 16 GB | 1 | — | Distributed (global hosts) | $0.1622 Best price | $0.1333 | Rent this GPU |
| Vast.ai | RTX 5090 | 32 GB | 1 | — | Distributed (global hosts) | $0.2129 Best price | $0.1000 | Rent this GPU |
| Vast.ai | RTX 3090 | 24 GB | 1 | — | Distributed (global hosts) | $0.2422 Best price | $0.1867 | Rent this GPU |
| Vast.ai | A100 PCIe | 40 GB | 1 | — | Distributed (global hosts) | $0.2681 Best price | $0.2000 | Rent this GPU |
| Vast.ai | L40S | 48 GB | 1 | — | Distributed (global hosts) | $0.4689 Best price | $0.4670 | Rent this GPU |
| Vast.ai | A100 SXM | 80 GB | 1 | — | Distributed (global hosts) | $1.0422 Best price | — | Rent this GPU |
| Vast.ai | H100 SXM | 94 GB | 1 | — | Distributed (global hosts) | $2.4741 Best price | $2.2000 | Rent this GPU |
| Vast.ai | B200 | 180 GB | 1 | — | Distributed (global hosts) | $6.2521 Best price | $6.2500 | Rent this GPU |
| Vast.ai | H200 SXM | 141 GB | 1 | — | Distributed (global hosts) | $14.8339 Best price | $5.3333 | Rent this GPU |
No offers match these filters. Relax the price limit or clear the search.
Frequently asked questions
Everything about renting cloud GPUs, affiliate tracking and how we collect pricing data.
How is this provider's GPU pricing structured?
Almost all cloud GPU providers bill by the second or the hour for the compute, then separately for persistent storage and outbound data transfer. The figures on this page are the compute rate for one GPU, normalised so you can compare different instance shapes without doing the arithmetic yourself.
Can I get interruptible or spot capacity here?
That depends on the provider and is shown in the availability column. Marketplace-style platforms usually offer a bid-based interruptible tier, while dedicated clouds tend to sell only on-demand capacity. If spot is available it is almost always cheaper, provided your workload can checkpoint and restart.
Is there a minimum contract or commitment?
For on-demand and spot capacity there generally is not: you pay per hour or per second and can stop at any time. Reserved contracts exist at some providers and can cut the effective rate, but they are not required for the prices shown on this page.
How are storage and egress charged?
Persistent volumes are billed per GB per month, and outbound transfer is charged per GB beyond whatever allowance the provider includes. Those costs are additive to the GPU hour and can dominate the bill on data-heavy pipelines, so budget for them explicitly.