This week we monitored publicly listed prices for mainstream GPU models across 26 global platforms. The currently lowest observed prices per GPU model are shown below. Spot/preemptible and off-peak pricing carries interruption risk—clearly marked; please select according to your needs. Full real-time data available atReal-time price comparison homepage。
| GPU model | This week’s lower monitored prices | Platform | |
|---|---|---|---|
| RTX 3090 | ¥0.63/hourSpot | Vast.ai (USA) | Rent Now |
| RTX 4090 | ¥0.87/hourSpot | Vast.ai (USA) | Rent Now |
| RTX 5090 | ¥1.53/hourSpot | Vast.ai (USA) | Rent Now |
| A100 80GB | $5.47/hourPay-as-you-go | Paperspace (United States) | Rent Now |
| A800 80GB | ¥5.15/hourPay-as-you-go | MornAI | Rent Now |
| H100 80GB | ¥8.28/hrPay-as-you-go | Paperspace (United States) | Rent Now |
| L40S 48GB | ¥5.69/hourSpot | RunPod (USA) | Rent Now |
| T4 16GB | ¥1.35/hourPay-as-you-go | MatPool | Rent Now |
| A6000 48GB | ¥2.38/hourSpot | RunPod (USA) | Rent Now |
| H200 141GB | ¥21.62/hourSpot | Verda (Finland) | Rent Now |
| B200 180GB | ¥30.69/hourSpot | Verda (Finland) | Rent Now |
| L4 24GB | ¥3.17/hourSpot | RunPod (USA) | Rent Now |
| L40 48GB | ¥4.97/hourSpot | RunPod (USA) | Rent Now |
| A40 48GB | ¥2.52/hourSpot | RunPod (USA) | Rent Now |
| H20 96GB | ¥3.40/hourSpot | PPIO Paiou Cloud | Rent Now |
| RTX 4090D 24GB | ¥1.14/hourPay-as-you-go | Suanjiayun | Rent Now |
| RTX PRO 6000 96GB | ¥9.15/hourSpot | Verda (Finland) | Rent Now |
| RTX 6000 Ada 48GB | ¥5.18/hourSpot | Verda (Finland) | Rent Now |
| V100 16GB | ¥0.84/hourSpot | Verda (Finland) | Rent Now |
| MI300X 192GB | ¥3.60/hourSpot | RunPod (USA) | Rent Now |
| RTX 6000 24GB | ¥10.80/hourPay-as-you-go | Akamai / Linode (United States) | Rent Now |
| RTX 4000 Ada 20GB | $3.74/hourPay-as-you-go | Akamai / Linode (United States) | Rent Now |
This Week’s Insights
· Consumer-grade GPUs (3090/4090) face fiercest competition—P2P markets (Vast.ai) and emerging domestic platforms (Suanjiayun, Paratera) alternate as lowest-price leaders, but low prices usually come with trade-offs: ‘must be claimed quickly’ or ‘preemptible’.
· For stable pay-as-you-go GPU usage, MatPool’s RTX 4090 (¥1.54/hour) and AutoDL’s RTX 4090 (¥1.88/hour) are reliable choices balancing cost and ecosystem support.
· For long-running tasks (>1 week), avoid hourly pricing: MornAI and Suanjiayun weekly/monthly plans typically cost 20–40% less than hourly equivalents.
· For high-VRAM workloads (A100/H100/H800), domestic platforms generally quote higher rates than overseas P2P markets. If your workload accepts foreign currency payment and tolerates interruptions, Vast.ai remains the lowest-cost source.