Decision tools
GPU Rankings: Performance, Price, and Value
GPU Index and value rankings to compare cloud GPU options for training, inference, and budget workloads.
Get recommendation → Open finder →
Rankings FAQ
How is Value Score calculated?
Value Score is computed from weighted performance signals and price per hour. Weights vary by use case (training, inference, budget).
Why can rankings change over time?
GPU cloud pricing and inventory change frequently. We update snapshots periodically, so rank order can move as prices shift.
Should I use Rankings or Finder first?
Use Rankings for a fast shortlist. Then use Finder to filter by region, VRAM, billing model, and provider constraints.
Use-case behavior: Training prioritizes H100/A100 class; Inference prioritizes 4090/3090/A10G value cards; Budget focuses lowest $/GPU/hr with VRAM ≥ 8GB.
| Rank | GPU | Provider | City | VRAM | $/GPU/hr | Monthly est. | Value score | Actions |
|---|
Method: $/GPU/hr = instance hourly price ÷ detected GPU count in plan name (when applicable). Monthly estimate uses 730h. Value score = (weighted perf score) / ($/GPU/hr). Data source: Finder pricing snapshots. Last updated: 2026-03-19. Methodology.
Value Score = (Perf Score × Use-case Weight + VRAM Weight) / Price per hour. Training emphasizes compute, inference balances compute and VRAM, budget prioritizes lower cost.
Start with our current promotion-priority providers: MaxCloudON, 99Stack, Cherry Servers, CloudClusters, DatabaseMart, GPUMart, Liquid Web, RunPod, Vultr, ServerMania.