Suanjia Cloud

What is Suanjia Cloud?

Suanjia Cloud is a GPU computing power leasing and AI training/inference platform operated by Guizhou Suanjia Computing Service Co., Ltd.

Users can create cloud GPU instances as needed for model training, inference, fine-tuning, AI-based image generation, and video creation.

Main functions

• Rent multiple GPU models as needed.
• Create and manage project instances.
• Use base images and application images.
• Start the AI model from the mirror community.
• Save data using the project’s cloud storage.
• Create and reuse personal project mirrors.
• The JupyterLab development environment is used.
• Via SSH or remote desktop connection.
• Open the ports required by the application.
• Long-term leasing and corporate solutions are available.
• Complementary large-model aggregation API.
• Provide integrated local solutions for governments and enterprises.

GPU resources and versions

The platform offers a Professional version called Pro, a Youth version called Air, as well as a section dedicated to domestic products; these different versions are designed for stable production, short-term testing, and scenarios involving domestic technology solutions.

Current reference price for GPUs

The home page of the official website shows the starting price per hour for some GPUs; the actual amount varies depending on the version, video memory capacity, node type, availability, and any ongoing promotions.

The prices shown on the official website during verification are the starting prices only; they do not represent the fixed transaction prices for all regions and configuration options.

How to choose a GPU

When choosing a GPU, it is not enough to consider only the model number; it is also necessary to evaluate the video memory, the size of the models, the precision, the batch size, and the training method.

Models and mirror communities

The official website states that the mirror community offers more than 200 models, covering areas such as language, images, videos, 3D, audio, digital avatars, and programming.

• The base image provides the framework and development environment.
• Model images can reduce the number of manual installation steps.
• Community mirrors require checking for version and maintenance status.
• Confirm the model license and intended use before starting.
• Pre-trained weights may need to be downloaded separately.
• The availability of an image does not mean that the model can be used for commercial purposes.
• Production projects should rely on fixed versions.

Project instance

A project instance is the environment in which computing tasks are actually executed; when creating one, it is necessary to select a GPU, an image, storage options, and billing settings.

1. Confirm the framework version required for the project.
2. Estimate the VRAM required for the model and data.
3. Select the appropriate GPU and resource version.
4. Set the system drive and project storage.
5. Choose a base image or a community image.
6. Create an instance after checking the order price.
7. After startup, verify the status of the GPU and drivers.
8. Save information about the reproducible environment.

Project cloud storage and data persistence

The project cloud storage is used to save code, datasets, weights, and results, thereby preventing the loss of important files after the instance is terminated.

• Distinguish between the system drive, project drive, and cloud drive.
• Confirm the saving rules after shutdown and release.
• Large datasets are uploaded in batches according to directories.
• Training checkpoints are written to persistent storage on a regular basis.
• Important results should be downloaded to your local device separately.
• Sensitive data is masked before being uploaded.
• Files that are not used for a long time should be deleted promptly.

Project image

Project images allow the saved, configured environment to be retained, facilitating task replication or the restoration of dependencies in subsequent instances.

1. Clear the cache and unnecessary temporary files.
2. Record the framework, driver, and dependency versions.
3. Remove keys and personal sensitive configurations.
4. Stop the training task that is still in the process of writing.
5. Create an image and provide a clear name for it.
6. Verify whether the image can be started from a new instance.
7. Retain version notes for critical images.

Remote connection method

The help center provides operation guides for JupyterLab, VS Code, PyCharm, SSH, VNC, and file transfer.

Ports are open.

When deploying a web interface, inference API, or visualization tools, it is usually necessary to map the internal ports of the instance to external ports.

1. Verify the port that the application is actually listening on.
2. Ensure that the service listens on the correct network address.
3. Only the ports required for business operations are opened.
4. Set a strong password or access token.
5. Do not expose the database and management ports.
6. Test the access from an external network.
7. Close the port mapping after the task is completed.

Calcium Cloud Usage Tutorial

1. Register an account and complete the necessary real-name verification.
2. Check the current prices for GPUs and storage before making a top-up.
3. Go to the center of the container and select the instance type.
4. Select the GPU and video memory based on the model size.
5. Select a base image or an application image.
6. Configure storage space and project cloud disk.
7. Create an instance after verifying the billing method.
8. Connect via JupyterLab or SSH.
9. Upload code, models, and datasets.
10. First, run small-scale tasks to verify the environment.
11. Start formal training or reasoning tasks.
12. Continuously monitor video memory, logs, and costs.
13. Save weights, results, and environment images.
14. After shutting down, check whether there are still storage fees.
15. Release resources according to the rules when they are no longer in use.

Prices and billing methods

Suanjia Cloud charges primarily based on GPU usage time, storage, and additional resources; it also offers long-term leasing options and customized solutions.

The price of the GPU and the cost of storage may be calculated separately, and the fees associated with shutting down, stopping, or releasing resources are also determined in accordance with the rules of the current order.

Cost estimation method

The actual cost should include the expenses related to GPU operation, storage, data transmission, long-term storage, and manual maintenance, and not just the price displayed on the homepage.

1. Record the estimated number of hours each task will take to complete.
2. Calculate the number of GPUs and the hourly price.
3. Estimate data, weights, and cache capacity.
4. Add retry attempts for failed joins and parameter tuning time.
5. Evaluate the storage retention fee after shutdown.
6. Compare the utilization rates of on-demand and long-term rentals.
7. Set reminders for account balance and fees.

Bridge calculation API

The Suanqiao API is an aggregation service for large models offered by Suanjia Cloud; it provides a unified key, an OpenAI-compatible format, and billing based on tokens.

It is a different product from GPU instance leasing: the former is charged based on the number of times the model is used, while the latter provides access to computing resources that can be controlled by the user.

Data security

Cloud GPUs can be used to process code, datasets, and model weights; security measures should be implemented based on the sensitivity level of the project before their use.

• Do not write keys to public images.
• Use environment variables to manage access credentials.
• Restrict external ports and management interfaces.
• Team members use separate accounts and permissions.
• Sensitive data should be masked or encrypted before being uploaded.
• Regularly back up training checkpoints and results.
• Clean up temporary credentials before releasing the instance.
• Companies review privacy and service agreements when making purchases.

Refunds and unpaid balances

The official help center provides information on refunds, overdue payments, and order management rules; refunds may not be available for the computing power that has already been used or for the benefits that were given as gifts.

• Read the current refund terms before making a top-up.
• Confirm the configuration and amount before creating an instance.
• If an error is detected, the resources should be stopped as soon as possible.
• Save orders, bills, and failure records.
• Outstanding payments may affect instance and data access.
• Enterprise users should apply for invoice details in advance.

Open-source status

The Suanjia cloud platform is a commercial cloud service; to date, no official open-source repository for this platform has been found.

Mirror communities can deploy open-source models and algorithms, but the license of those models is a separate matter from whether the platform software is open-source.

Which users are it suitable for

• Students and developers who are learning deep learning.
• Universities and research teams that train models.
• Creators who work in the field of AI-powered image and video generation.
• Algorithm engineers who need model fine-tuning.
• Startups that deploy inference services.
• Organizations that need domestic computing power for adaptation.
• We are targeting government and enterprise clients who prefer on-premises deployment.

Product advantages

• Various GPU and resource versions are available.
• Supports on-demand, long-term rental, and customized delivery.
• Mirror communities lower the barriers to deploying models.
• The documentation covers common development tools.
• Supports the reuse of cloud storage and project mirrors.
• It offers both API and on-premises all-in-one solutions.

Usage restrictions

• Low-priced resources may be affected by inventory and promotions.
• The starting price on the official website is not a fixed price for all configurations.
• Even after shutting down, attention should still be paid to storage costs.
• Community mirrors need to conduct their own security audits.
• Production services should have their stability and network tested.
• The fact that a cloud platform is not open source does not mean that its models are also not open source.
• Mining is prohibited, as are any uses that violate the platform’s rules.

Frequently Asked Questions

What services does Suanjia Cloud mainly provide?

Suanjia Cloud offers cloud-based GPU instances, model images, project storage solutions, remote development capabilities, long-term leasing options, Suanqiao API, as well as enterprise AI integrated systems.

Can Suanjia Cloud be used for free?

The official website offers a free way to get started as well as promotions for new users; however, the actual free quota for GPUs, storage, and additional resources is determined by the details on the current account page.

How much does Juja Cloud GPU cost per hour?

During verification, the official website showed that the cost for the RTX 4090D is 1.14 yuan per hour, while the RTX 4090 costs 1.24 yuan per hour; the prices for other models are determined according to the rates displayed on the page at that time.

Is there still a charge after shutting down Suanjia Cloud?

The costs associated with GPU usage and storage can be separate; when the device is turned off, no charges are incurred for usage resources, but storage that remains in use may still be subject to fees – it is necessary to check the order details for relevant rules.

Is Suanjia Cloud suitable for training large models?

The platform offers a variety of GPU options as well as large amounts of memory, which can be used for training, fine-tuning, and inference; the feasibility of using these resources depends on the size of the model, its precision, and the parallelization approach employed.

What is the difference between bridge APIs and GPU leasing?

The Bridge API invokes the hosted model on a token basis, while GPU leasing provides computing instances that allow users to install their own environment and deploy models; the billing methods for these two options are different.

Is Suanjia Cloud an open-source platform?

No, Juajia Cloud is a commercial cloud service; open-source models can be run on it, but this does not mean that the cloud platform itself has open-source code.