
Suanjiayun (算家云)
A domestic GPU compute rental platform billed by the second

Tool details
Suanjiayun (算家云) is a GPU compute rental platform operated by Guizhou Suanjia Computing Services Co., Ltd. What it sells is container instances and whole-machine compute billed by the second, not a generative AI application. The home page lays unit prices from the RTX 3060 up to the A100 straight out in the open, starting at 0.58 yuan per card per hour.
Core capabilities
- Per-second billing: the meter starts when the instance boots and the compute charge stops when it shuts down, timed to the second with a 0.01 yuan minimum, so experimental workloads waste almost nothing
- 8 card types in stock: RTX 3060 0.58, 4090D 1.14, 3090 1.20, 4090 1.24, 5090 2.68, L40S 5.08 and A100 SXM4 9.80 yuan per card-hour
- Image and model marketplace: 201 images and 199 data entries, with DeepSeek, ComfyUI and Qwen each launching a container in one click after localization testing
- Long-term rental discount ladder: daily billing at 9 tenths of the list rate, weekly at 85 hundredths, monthly at 8 tenths, 7.8 tenths for 3–5 months, and 7 tenths at most on annual terms
- Bare-metal and supercomputing adjacents: 8-card bare metal spans 12 price bands from the H200 down to the 3060, alongside supercomputing and private-deployment appliances
- Operations toolchain: Jupyter, SSH/SFTP, port mapping, K8s nodes, instance cloning and resizing are all built in
Where it fits Individual developers and graduate students running Stable Diffusion, ComfyUI and LoRA fine-tuning; university research groups that use the research images for computation; small and mid-sized businesses that need compute capacity temporarily; and public-sector or enterprise units with domestic-stack and private-deployment requirements.
Distinctive strengths Set against AutoDL and Hengyuan Cloud, its prices are lower and its ecosystem is weaker. The real differentiation is in credentials and product breadth: a value-added telecom B2 licence, a Guizhou ICP filing, a data-centre background under Gui'an state-owned capital, and a domestic-chip zone plus appliances. It does not develop foundation models of its own; on the model side it provides prebuilt images only, and hosted APIs sit on a separate sub-site.
Editorial verdict The low prices can be verified; engineering usability has to be discounted. In this test only the home page responded — sign-in, the console and the pricing sub-pages all timed out, and the ordering flow was not verified end to end. It suits individuals and research groups that put cost first and can cover the gaps themselves; enterprises that need SLAs and compliance documents should run a small trial before committing.
Pricing
Subscription¥0.58 / GPU / hour (RTX 3060 12G, on-demand billing)New users who register, follow the official account and complete identity verification receive up to 5 Suanjia coins (about 5 yuan) in trial credit, and must top up once that is used; there is no permanently free tierThe lowest tier on the site; the home page shows no struck-through list price for it; suited to entry-level image generation and small-model experiments
Struck-through list price ¥1.98; the Samsung-memory version of the same generation, and the mainstream value tier
Struck-through list price ¥1.68; 936 GB/s of memory bandwidth, suited to inference and small-to-medium fine-tuning
Struck-through list price ¥2.08; the console banner labels it a limited-time promotional price rather than a fixed catalogue rate
Struck-through list price ¥3.28; 32G of memory and 1792 GB/s of bandwidth
Struck-through list price ¥7.98; 48G of large memory, suited to multi-GPU inference and mid-sized training
Struck-through list price ¥12.00; 624 TFLOPS of compute and 1935 GB/s of bandwidth, the main training tier
Additional notes
First, all figures above are the on-demand rates in yuan per card-hour measured on the home page on 2026-10-30; the page carries both a limited-time promotional label and a struck-through list price, and whether these persist is up to the operator's announcements — the same-generation RTX 4090 48G version is ¥2.28 per card-hour and is not listed separately. Second, the long-term discount ladder runs daily 9 tenths, weekly 85 hundredths, monthly 8 tenths, 7.8 tenths for 3–5 months, 7.5 tenths for 6–11 months and 7 tenths annually, and a separate discount announcement dated 2026-06-26 quotes on-demand rates that do not match the current home-page prices. Third, storage is charged on the data disk at 0.01 yuan per day per GB, deducted at the 24-hour mark each day, and the project network drive includes 50GB free; public bandwidth specifications, traffic fees and SLAs are not published. Fourth, 8-card bare-metal machines (12 price bands from the H200 down to the 3060, ¥630–75000 per machine per month) are range prices that require a quote. Fifth, top-up balances are in principle neither withdrawable nor refundable; where part of a refund is allowed, Alipay deducts a fee of about 5.6% and WeChat about 6.0%, the refund terms conflict in three places, and 14 terms paths including /pricing returned 404 in this test.
FAQ
User reviews
User Reviews
No reviews yet
No one has reviewed this tool yet