简体中文
EnglishCloud GPU Price Comparison: Current Per-Hour Rates by GPU Model and Provider
Updated
The cheapest secure on-demand cloud GPU rates currently start at $0.44/hr for an RTX A6000 (LeaderGPU), $0.53/hr for an RTX 5090 (Vast.ai), and $2.27/hr for an H100 SXM5 (Vast.ai), according to GPUPerHour's live snapshot from 31 providers (last stock observation: 2026-10-01 13:40 UTC). These figures apply only to secure on-demand instances — spot, reserved, and peer-to-peer marketplace listings are tracked separately and excluded from the comparison because their pricing isn't directly comparable. Availability can change before you book.
Current Cheapest Price by GPU Model
All prices are per GPU-hour, normalized from multi-GPU instance pricing (instance price ÷ GPU count). "Providers with stock" reflects how many providers had available inventory at the time of the snapshot.
| GPU | Cheapest current price | Provider | 30-day catalog avg. | Providers with stock |
|---|---|---|---|---|
| H100 SXM5 | $2.27/hr | Vast.ai | $2.18/hr | 3 |
| H100 PCIe | $2.59/hr | QuantaCloud | $2.77/hr | 5 |
| H100 NVL | $3.11/hr | Massed Compute | $1.91/hr | 3 |
| H200 SXM | $3.50/hr | Ori | $2.70/hr | 2 |
| H200 NVL | $3.43/hr | QuantaCloud | $2.88/hr | 4 |
| B300 SXM6 | $7.89/hr | RunPod | $5.97/hr | 1 |
| A100 SXM4 80GB | $0.67/hr | Vast.ai | $1.06/hr | 7 |
| A100 PCIe 80GB | $0.68/hr | LeaderGPU | $1.69/hr | 4 |
| L40S | $0.97/hr | Massed Compute | $0.90/hr | 4 |
| RTX PRO 6000 Blackwell MIG 24GB | $0.59/hr | RunPod | $1.79/hr | 5 |
| RTX 5090 | $0.53/hr | Vast.ai | $0.46/hr | 2 |
| RTX 4090 | $0.74/hr | RunPod | $0.33/hr | 3 |
| RTX A6000 | $0.44/hr | LeaderGPU | $0.57/hr | 7 |
| MI300X | $3.39/hr | Hot Aisle | $2.75/hr | 1 |
Why Only Secure On-Demand Listings Are Comparable
GPUPerHour's methodology normalizes every listing to a price per GPU per hour and tags it by pricing type and security tier. The current-price column uses only secure on-demand listings. Spot, reserved, and peer-to-peer marketplace listings are tracked but excluded because:
- Spot pricing fluctuates with market demand and can be interrupted — it isn't a stable per-hour rate.
- Reserved pricing typically involves commitments or prepayment, so the headline rate doesn't reflect true on-demand cost.
- Peer-to-peer marketplace listings may vary in security tier and reliability, making them a different product category.
If you're comparing raw hourly cost for a workload that needs guaranteed, secure capacity, the on-demand column is the right benchmark. If you can tolerate interruption, spot listings may be cheaper — but that's a different trade-off.
How Per-GPU-Hour Pricing Is Normalized
For multi-GPU instances, the listed instance price is divided by the GPU count to produce a per-GPU-hour figure. This lets you compare, for example, an 8×H100 node against a single H100 rental on the same basis. When you're evaluating a provider, check whether the quoted price is per instance or per GPU — the normalized figure is what makes cross-provider comparison meaningful.
Historical Averages: Context, Not a Price Guarantee
The 30-day catalog average column is the median of daily catalog averages over the last 30 days, including all pricing types and availability states (spot, reserved, unavailable listings). It does not measure historical live prices or savings. Use it as context for whether a current rate is high or low relative to recent catalog activity — not as a guaranteed price or a savings estimate.
For example, the H100 NVL's current cheapest rate ($3.11/hr) is above its 30-day catalog average ($1.91/hr), while the RTX 4090's current rate ($0.74/hr) is above its average ($0.33/hr) — both suggest current on-demand pricing is running higher than the broader catalog median, but that median includes non-comparable listing types.
Practical Steps to Compare and Rent
- Pick your GPU model based on workload — H100/H200/B300 for large-scale training and inference, A100 for general deep learning, L40S for graphics and inference, RTX 4090/5090/A6000 for smaller-scale or cost-sensitive work, MI300X for AMD-based workloads.
- Check the cheapest current secure on-demand rate for that model in the table above.
- Confirm availability — the "providers with stock" column shows how many providers had inventory; this can change before booking.
- Verify the pricing type — make sure the listing you're booking is secure on-demand, not spot or reserved, if you need guaranteed capacity.
- Check the per-GPU normalization — confirm whether the provider quotes per instance or per GPU.
- Book and deploy — GPUPerHour links to provider deploy pages for each listing.
Key Caveats
- Availability changes quickly. The snapshot refreshes at most every 5 minutes, and current offers must be observed within 15 minutes of the summary's last refresh to count.
- Historical averages include unavailable and spot offers — they are not a measure of what you could have paid.
- Only 31 providers are tracked, and only those with live stock tracking and public pricing/inventory endpoints. Providers without live tracking aren't represented.
- Daily history and CSV downloads are available via GPUPerHour's Cloud GPU Price Index if you want to track trends over time.