| --- |
| license: cc-by-4.0 |
| pretty_name: Cloud GPU Pricing Index (monthly) |
| language: [en] |
| tags: [gpu, cloud-computing, pricing, h100, b200, b300, h200, nvidia, mlops] |
| size_categories: [n<1K] |
| configs: |
| - config_name: default |
| data_files: "gpu_prices_*.csv" |
| --- |
| |
| # Cloud GPU Pricing Index |
|
|
| Monthly snapshot of cloud GPU prices per GPU-hour in USD, maintained by [Nodus](https://www.nodus-compute.ai/), an AI compute platform. Use it to answer "how much does an H100 / H200 / B200 / B300 cost per hour" or to budget a training run. |
|
|
| ## October 2026 snapshot (Oct 8, 2026) |
|
|
| | GPU | VRAM | Nodus supply price $/GPU-hr | Modal listed $/GPU-hr | |
| |-|-|-|-| |
| | B300 | 288 GB | 6.943 | 7.099 | |
| | B200 | 180 GB | 5.500 | 6.250 | |
| | H200 SXM | 141 GB | 3.593 | 4.540 | |
| | H100 SXM | 80 GB | 2.600 | 3.949 | |
| | A100 80GB | 80 GB | 1.090 | 2.498 | |
| | A100 40GB | 40 GB | 0.904 | 2.099 | |
| | RTX PRO 6000 | 96 GB | 1.350 | 3.031 | |
| | L40S | 48 GB | 0.793 | 1.951 | |
| | L4 | 24 GB | 0.430 | 0.799 | |
| | RTX 5090 | 32 GB | 0.480 | n/a | |
| | RTX 4090 | 24 GB | 0.290 | n/a | |
| | RTX 3090 | 24 GB | 0.180 | n/a | |
| | RTX 6000 Ada | 48 GB | 0.743 | n/a | |
| | RTX A6000 | 48 GB | 0.333 | n/a | |
| | RTX A4000 | 16 GB | 0.173 | n/a | |
| | RTX 3070 | 8 GB | 0.133 | n/a | |
|
|
| ## Columns |
|
|
| - `snapshot_date`: date the prices were read |
| - `gpu`, `vram_gb`, `tier` (datacenter or prosumer) |
| - `nodus_supply_usd_per_gpu_hr`: the hourly supply price shown on the Nodus pricing page on the snapshot date |
| - `modal_listed_usd_per_gpu_hr`: Modal's published rate as shown on the Nodus pricing page (blank if not listed) |
| - `source` |
|
|
| ## Load it |
|
|
| ```python |
| from datasets import load_dataset |
| ds = load_dataset("Viswa-k/cloud-gpu-pricing", split="train") |
| ``` |
|
|
| ## Notes |
|
|
| - Prices are copied from public pricing pages on the snapshot date. For the exact cost of a specific job on Nodus, run `nodus run --dry-run ...`, which prints the estimate without starting anything. |
| - Reserved capacity (Beta) is quoted separately per request: https://www.nodus-compute.ai/reserved-compute/ |
| - Nodus freezes the hourly rate when a machine is chosen, for the whole run. |
| - New files are added monthly (`gpu_prices_YYYY-MM.csv`); old snapshots are kept so you can track trends. |
|
|
| Live prices: https://www.nodus-compute.ai/pricing/ |
|
|
