When you shop for cloud GPUs, the headline number is rarely the whole story. Here is the checklist I use to avoid being misled.
1. Start with the provider's own pricing page
Aggregators and blog posts go stale fast. Read the price on the provider's official pricing page, and keep the link.
2. Compare per-hour price by GPU model and VRAM
"One GPU" can mean very different hardware. Compare like for like: the same model (for example an A100 vs. an H100) and the same VRAM size (40 GB vs. 80 GB), then look at price per hour.
3. Watch for storage and egress fees
Persistent disks, snapshots and outbound data transfer can add a surprising amount to the bill, especially for large datasets and model checkpoints. Check whether they are included.
4. Separate spot from on-demand pricing
Spot or preemptible instances can be much cheaper, but they can be interrupted. Record which type each price refers to, and never mix them in one comparison.
5. Date your comparison
GPU prices change often. Write down the date you checked each price so readers (and future you) know how fresh it is.
I keep a dated, source-checked comparison at AskGPU.
Top comments (0)