Publicly listed GPU pricing monitoring results across platforms on October 17 (auto-refreshed hourly; full dataset available atReal-time price comparison homepage; prices are subject to the order page.
| GPU model | Today’s lowest monitored price | 24 hours | Platform | vs. yesterday | |
|---|---|---|---|---|---|
| RTX 3090 | ¥0.65/hour | ¥16/day | Salad (USA) | No change | Rent Now |
| V100 16GB | ¥0.79/hour | ¥19/day | Verda (Finland) | No change | Rent Now |
| RTX 4090 | ¥0.87/hour | ¥21/day | Vast.ai (USA) | No change | Rent Now |
| RTX 4090D 24GB | ¥1.14/hour | ¥27/day | Suanjiayun | No change | Rent Now |
| T4 16GB | ¥1.35/hour | ¥32/day | MatPool | No change | Rent Now |
| RTX 5090 | ¥1.53/hour | ¥37/day | Vast.ai (USA) | No change | Rent Now |
| A6000 48GB | ¥2.20/hour | ¥53/day | Gpushare | No change | Rent Now |
| A40 48GB | ¥2.52/hour | ¥60/day | RunPod (USA) | No change | Rent Now |
| L4 24GB | ¥3.17/hour | ¥76/day | RunPod (USA) | No change | Rent Now |
| H20 96GB | ¥3.40/hour | ¥82/day | PPIO Paiou Cloud | No change | Rent Now |
| MI300X 192GB | ¥3.60/hour | ¥86/day | RunPod (USA) | No change | Rent Now |
| RTX 6000 Ada 48GB | ¥4.74/hour | ¥114/day | Verda (Finland) | No change | Rent Now |
| L40 48GB | ¥4.97/hour | ¥119/day | RunPod (USA) | No change | Rent Now |
| A800 80GB | ¥5.15/hour | ¥124/day | MornAI | No change | Rent Now |
| L40S 48GB | ¥5.69/hour | ¥137/day | RunPod (USA) | No change | Rent Now |
| A100 80GB | ¥7.26/hour | ¥174/day | Verda (Finland) | No change | Rent Now |
| RTX PRO 6000 96GB | ¥8.37/hour | ¥201/day | Verda (Finland) | No change | Rent Now |
| H100 80GB | ¥11.88/hour | ¥285/day | PPIO Paiou Cloud | No change | Rent Now |
| H200 141GB | ¥19.77/hour | ¥474/day | Verda (Finland) | No change | Rent Now |
| B200 180GB | ¥28.48/hour | ¥684/day | Verda (Finland) | No change | Rent Now |
Today’s Insights
· Lowest price network-wide today:RTX 3090 As low as ¥0.65/hour on Salad (USA) (≈¥16/day), spot pricing, interruptible—suitable for short tasks or checkpoint-capable workloads.
· When consumer-grade GPUs (RTX 3090/4090) show minimal price variation across platforms, prioritize inventory availability and ecosystem maturity: popular GPUs often sell out during peak hours, and platforms with rich prebuilt images and tutorials save significant environment setup time.
· For long-running jobs (>1 week), don’t fixate on hourly rates: weekly or monthly plans typically cut total cost by 20–40%; always calculate the full cost before ordering.
· Spot/preemptible instances can be reclaimed at any time—avoid them for training jobs that cannot checkpoint. High-memory GPUs (A100/H100/H800) remain broadly cheaper on overseas P2P markets than domestic quotes; users accepting foreign currency payments should compare.