Monitored public pricing across platforms on Oct 9 (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 | ¥0.72/hour | ¥17/day | MornAI | No change | Rent Now |
| RTX 4090 | ¥0.87/hour | ¥21/day | Vast.ai (Overseas) | 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 (Overseas) | No change | Rent Now |
| A6000 48GB | ¥2.85/hour | ¥68/day | MatPool | No change | Rent Now |
| A800 80GB | ¥5.15/hour | ¥124/day | MornAI | No change | Rent Now |
| A100 80GB | ¥6.98/hour | ¥168/day | Suanjiayun | No change | Rent Now |
| H100 80GB | ¥14.23/hour | ¥342/day | Vast.ai (Overseas) | No change | Rent Now |
Today’s Insights
· RTX 3090 at ¥17/day (MornAI) remains the best-value option, ideal for fine-tuning and inference of 7B/13B models.
· RTX 4090 at ¥21/day (Vast.ai overseas spot) costs nearly two-thirds less than A6000—optimal for lightweight large-model training.
· A100 80GB at ¥168/day (Suanjiayun) is the current lowest; outperforms MornAI’s A800 (¥124/day but with constrained memory bandwidth).
· H100 at ¥342/day (Vast.ai overseas spot) is the only available H100 channel — recommended only for FP8 training users with multi-thousand-GPU-scale requirements.
· T4 at ¥32/day (MatPool) is notably pricier than RTX 3090 and no longer offers competitive value—new tasks are not recommended.