GPU Compute Marketplace
Hourly billing — you only pay while it runs. Training, inference, video generation and image generation on one machine.
GPUs & Pricing
Filter by compute, VRAM and price. Every GPU supports hourly billing — the meter stops when you power off.
| GPU | VRAM | FP16 TFLOPS | Hourly | Monthly | Best for | Action |
|---|---|---|---|---|---|---|
NVIDIA H800 High-End · 大规模训练首选 | 80 GB | 756 | ¥9.98 /hr | ¥5,999 /mo | 训练大模型微调 | Rent now |
NVIDIA RTX PRO 6000 High-End · 视频、图像生成的高性价比选择 | 96 GB | 503.8 | ¥8.36 /hr | ¥4,999 /mo | 视频生成图像生成推理 | Rent now |
NVIDIA A800 80GB High-End | 80 GB | 312 | ¥5.59 /hr | ¥3,499 /mo | 训练推理 | Rent now |
NVIDIA RTX 5090 Mainstream | 32 GB | 210 | ¥2.78 /hr | ¥1,699 /mo | 推理个人开发 | Rent now |
NVIDIA RTX 4090 Mainstream | 24 GB | 165.2 | ¥1.88 /hr | ¥1,099 /mo | 推理微调出图 | Rent now |
NVIDIA H20 High-End | 96 GB | 148 | ¥9.98 /hr | ¥5,999 /mo | 推理长上下文 | Rent now |
NVIDIA V100 Mainstream | 32 GB | 125 | ¥1.88 /hr | ¥1,099 /mo | 训练老项目兼容 | Rent now |
NVIDIA RTX A4000 Entry | 16 GB | 76.7 | ¥0.92 /hr | ¥559 /mo | 轻量推理开发调试 | Rent now |
NVIDIA RTX 3090 Mainstream | 24 GB | 71 | ¥1.32 /hr | ¥799 /mo | 推理出图 | Rent now |
NVIDIA RTX 3080Ti Entry | 12 GB | 70 | ¥0.98 /hr | ¥599 /mo | 轻量推理 | Rent now |
NVIDIA RTX 2080Ti Entry | 11 GB | 53.8 | ¥0.88 /hr | ¥529 /mo | 开发调试学生 | Rent now |
Estimate your cost
Drag the sliders to see what your workload costs. If usage exceeds the monthly plan, we calculate the cheaper option automatically.
Estimates only — your invoice is the source of truth. Billing stops when powered off.
Will my model fit in VRAM?
Pick model size and precision to see the VRAM needed, plus the cheapest GPU that can run it.
Estimated from weights + optimizer + activation headroom. Actual depends on the implementation.
Compute leaderboard
Ranked by FP16 throughput to help you balance performance against cost.
Values are vendor-published FP16 tensor throughput; real performance also depends on memory bandwidth and model architecture.
How it works
Pick a GPU
Choose by VRAM and compute, order hourly or monthly.
Boot and go
Deep-learning stack preinstalled, mainstream frameworks ready out of the box.
Run your job
Connect via SSH or Jupyter — train, infer, generate.
Stop and stop paying
Power off to stop billing. Your data stays, restart anytime.
Do the math before you buy
Not sure about specs? Tell us your model and budget and we'll size it for you.