Benchmarks · GPU & AI
GPU cloud price-performance: per-GPU-hour prices and cost per million tokens
Price-performance combines two sources: published list prices (with access date and billing basis) and measured throughput. This page normalises list prices by GPU memory and bandwidth, which are the resources that bound LLM inference. Prices in different currencies are never mixed in one chart.
Updated:
Direct answer
Compare GPU clouds by price per GPU-hour divided by memory (GB) and by bandwidth (TB/s), then confirm with cost per million output tokens from a measured run at your target concurrency. List prices vary by commitment, region and configuration; the tables show the basis for each figure.
What is measured
- Published list price per GPU-hour with billing basis
- Price per GB of GPU memory per hour
- Price per TB/s of memory bandwidth per hour
- Cost per million output tokens (verified runs only)
Published per-GPU-hour prices
| Provider | GPU | Per GPU-hour | Per GB-hour | Per TB/s-hour | Basis |
|---|---|---|---|---|---|
| PortvMind | NVIDIA H100 | TRY 195.835 | TRY 2.4479 | TRY 58.458 | gpu.h100s.xlarge (1× H100) hourly list price, tr-ist-01, excluding tax. Multi-GPU profiles price from ₺141.12 per GPU-hour (3× and 4× profiles). Source · 6 November 2026 |
| DigitalOcean | NVIDIA H100 | $4.410 | $0.0551 | $1.316 | On-demand, 1× H100 GPU Droplet. Source · 6 November 2026 |
| DigitalOcean | NVIDIA H200 | $4.470 | $0.0317 | $0.931 | On-demand, 1× H200 GPU Droplet. Source · 6 November 2026 |
| DigitalOcean | NVIDIA L40S | $1.570 | $0.0327 | $1.817 | On-demand, 1× L40S GPU Droplet. Source · 6 November 2026 |
| Vultr | NVIDIA H100 | $2.300 | $0.0287 | $0.687 | On-demand "starting at" per-GPU rate, 8-GPU HGX configuration. Source · 6 November 2026 |
| Vultr | NVIDIA L40S | $1.671 | $0.0348 | $1.934 | On-demand per-GPU rate, 1× L40S. Source · 6 November 2026 |
Offers without a public list price (on request or commitment-based): PortvMind NVIDIA H200, PortvMind NVIDIA L40S, AWS NVIDIA H100, AWS NVIDIA H200, AWS NVIDIA L40S, Azure NVIDIA H100, Azure NVIDIA H200, Google Cloud NVIDIA H100, Google Cloud NVIDIA H200.
Charts by currency
Currencies are never mixed in one chart, and no exchange-rate conversion is applied.
Published list prices
Data table
| Configuration | Price (TRY/GPU-h) |
|---|---|
| PortvMind H100 | 195.835 |
Published list prices
Data table
| Configuration | Price (USD/GPU-h) |
|---|---|
| DigitalOcean L40S | 1.57 |
| Vultr L40S | 1.671 |
| Vultr H100 | 2.3 |
| DigitalOcean H100 | 4.41 |
| DigitalOcean H200 | 4.47 |
Verified results
Related research
GPU & AI infrastructure
Best GPU for LLM inference: matching H100, H200 and L40S to model size
Which GPU to use for LLM inference by model size, precision, context length and concurrency — a sizing guide for Qwen, Llama, DeepSeek and MiniMax.
Pricing
GPU cloud cost comparison: H100, H200 and L40S prices per GPU-hour
Published H100, H200 and L40S cloud prices per GPU-hour with sources, normalised per GB of memory and per TB/s of bandwidth, and what they leave out.