Youyun Intelligent Computing

What is Youyun Intelligent Computing?

Compshare, part of UCloud, is a platform for leasing GPU computing power and for accessing APIs related to large-scale models.

It is designed for AI training, model inference, scientific computing, development and testing, as well as production deployment, and offers GPU and model services on demand.

Main functions

• Create GPU instances of different models as needed.
• It supports both containers and virtual machines.
• Hourly, daily, monthly, and preemptive billing are available.
• Charging for GPU computing stops after shutdown.
• The cardless mode can be switched to in order to maintain the operating environment.
• Pre-installed with CUDA, PyTorch, and TensorFlow.
• More than 300 community application images are available.
• Mount the public model library for quick startup.
• Connect via Jupyter, SSH, and VS Code.
• Call the APIs for large language, image, and video models.
• Provides team budget and invoice management.
• Supports API, CLI, and Agent Skills.

Which users are it suitable for

What are the different forms of GPU instances?

Common GPU card types

The platform offers GPUs for consumer use as well as for data centers; the availability, memory versions, and stock levels vary depending on the region.

• The RTX 5090 is suitable for new-generation development and testing.
• The RTX 4090 is suitable for training and high-performance inference.
• The RTX 4090 48G is suitable for those with higher memory requirements.
• The RTX 3090 is suitable for experiments with small and medium-sized models.
• The RTX 3080Ti is suitable for tasks where budget is a concern.
• The A100 and A800 are designed for data center workloads.
• H20 is suitable for specific large-model inference scenarios.
• V100S, P40, etc. can be used for compatibility tasks.

How to choose a GPU

Tutorial on creating GPU instances

1. Register and complete the required real-name verification.
2. Go to the GPU instance creation page.
3. Select the region and availability zone.
4. Check the current card type, inventory, and price.
5. Determine the number of GPUs, as well as the CPU and memory.
6. Choose between container or virtual machine format.
7. Choose a basic, community, or custom image.
8. Configure the system drive and persistent storage.
9. Choose between pay-as-you-go, daily, or monthly billing.
10. Create an instance after verifying the estimated costs.
11. Wait for the instance to become available.
12. Connect via Jupyter, SSH, or VS Code.
13. Run small tasks to verify the driver and environment.
14. Set up cost reminders and task saving strategies.

How to choose an image?

Base images are suitable for custom environments, while community images are appropriate for rapid deployment of specific frameworks or applications.

Methods for connecting to instances

• JupyterLab is suitable for notebook experiments.
• SSH is suitable for terminal management and script tasks.
• VS Code is suitable for remote code development.
• Clients such as FinalShell are suitable for graphical connections.
• Windows virtual machines can use Remote Desktop.
• FileZilla or XFTP can be used for file transfer.

Before making the first connection, verify the instance address, port, username, and the login credentials generated by the platform.

File upload and download

1. Choose the transmission method based on the file size.
2. Small files can be uploaded through the Jupyter interface.
3. Project code can be transmitted via Git or a client.
4. Large datasets should be stored in persistent storage first.
5. Verify the number and size of files after transmission.
6. Independent backups are kept for critical data.
7. Confirm that the results have been exported before deleting the instance.

How to understand that there is no charge for shutting down the device?

No charges are applied for GPU computing resources after the instance is shut down, but costs may still apply for resources such as disks, cloud storage, and images.

When it is necessary to preserve the environment but no GPU is to be used, the total cost of the card-free mode can be compared with that of shutting down the system and using persistent storage.

• Save the results promptly after the task is completed.
• Confirm that the process has been completely stopped.
• Check whether the instance is shut down.
• Check the storage resources that are still incurring charges.
• Delete unused images that are no longer valuable after a long period of time.
• Regularly check the bills and the list of resources.

Real-time reference price for GPUs

The following are the lowest prices listed on the official website as of August 31, 2026; location, configuration, inventory levels, and any promotions can affect the final price.

Before creation, the final amount shown in the console should be taken as the reference; the system drive, data drive, images, and network may incur additional charges.

Storage price

Storage continues to retain data even after the GPU stops working, which can thus become a significant cost for long-term projects.

Large model API

The model API enables developers to invoke models through standard requests, without the need to download, deploy, or maintain the underlying inference environment themselves.

• Supports calling large language models.
• It supports text-to-image and image-to-image generation.
• It supports video generation from text as well as video generation from images.
• It can be connected to Dify and RAGFlow.
• It can be integrated with automation tools such as n8n.
• It can be used with development tools such as Claude Code.
• The specific models, magnifications, and prices will be updated.

Pricing for the Coding Plan package

The Coding Plan provides a quota for model calls on a periodic basis, with different models having varying call multiplicities.

The numbers in the table are estimates provided by the official website; different tools may send multiple requests for a single task, so the actual number of tasks that can be completed is not equal to the number of requests made.

How to choose a Coding Plan

Limits on model packages

• Different models have different deduction multiples.
• A single task may result in multiple API calls.
• Different packages have concurrency limits.
• The service will be discontinued once the periodic quota is exhausted.
• Switching to pay-as-you-go may require a new key.
• The list of supported models will be updated by the official team.
• Check the current package’s FAQ before purchasing.

Team and teaching management

The team features allow for inviting members, allocating budgets, recovering unused balances, as well as viewing order and transaction records.

1. The root account creates a team and obtains the Team ID.
2. Invite registered platform members.
3. Wait for team members to accept the team invitation.
4. Assign an independent budget to each member.
5. Members are asked to switch to the team perspective.
6. View member orders and usage details.
7. Adjust the amount according to the progress of the course.
8. Recover unused budget to the main account.
9. Export the necessary bills and operation records.

Cost control for training tasks

• First, validate the code using a small amount of data.
• Monitor GPU utilization and VRAM usage.
• Regularly save checkpoints and logs.
• Interruptible tasks take into account preemptive instances.
• Data preprocessing should not utilize expensive GPUs.
• Shut down immediately after the experiment is completed.
• Clean up unnecessary disks and private images.
• Set up a separate budget for each project.

Data and security considerations

Cloud instances handle code, models, datasets, and keys; businesses and research projects should first determine their data protection requirements.

• Do not store API keys in public repositories.
• Use environment variables or key management methods.
• Restrict SSH ports and login credentials.
• Sensitive data is masked before being uploaded.
• Important results are saved in persistent storage.
• For community mirrors, the source and content are checked first.
• Clean up confidential files before deleting the instance.
• Regulated data must first undergo a compliance assessment.

API, CLI, and open-source status

YouYun Intelligent Computing provides instance management APIs, UCloud SDK calling methods, CompShare CLI, as well as skills designed for agents.

These development tools and examples can be used publicly, but the Youyun AI Cloud platform itself is not an open-source project.

API integration steps

1. Create an API key in the console.
2. Store the public key and private key only in a secure environment.
3. Choose the official SDK or request the API directly.
4. Set the correct region and availability zone.
5. First, call the price and inventory query interface.
6. Submit a request to create a minimal instance.
7. Record the identifier of the returned instance.
8. Polling the instance status and handling failures.
9. Stop or delete the resources once the task is completed.
10. Fee protection is set for all creation operations.

Product advantages

• There is a wide variety of GPU models and deployment methods available.
• Supports both on-demand and long-term billing options.
• A pre-installed environment reduces configuration costs.
• The public model library reduces duplicate downloads.
• Provides model APIs and Coding Plans.
• The team budget is suitable for teaching and collaboration.
• APIs, CLI, and Skills facilitate automation.

Usage restrictions

• GPU inventory and prices change dynamically.
• Charging may still occur even after the device is turned off.
• The quality of community mirrors needs to be verified by yourself.
• Preemptive instances may be interrupted.
• Model packages have rate and concurrency limits.
• An Agent task may be invoked multiple times.
• Production deployment requires additional stability testing.

Frequently Asked Questions

What services does Youyun Intelligent Computing mainly offer?

It offers on-demand GPU cloud instances, large-model APIs, pre-installed images, a public model library, team budget management, as well as APIs, CLI, and Agent Skills.

How much does YounCloud AI’s GPU cost per hour?

As of August 31, 2026, the official website indicated that the price of the 4090 was as low as 2.15 yuan per unit, the 5090 was as low as 3.32 yuan per unit, and the A800 was as low as 6.23 yuan per hour; the final prices shall be subject to those displayed on the console.

Is there still a charge after Youyun Intelligent Computing is shut down?

After shutdown, charges for GPU computing are usually stopped, but resources such as cloud disks, persistent storage, and private images may still incur fees.

What connection methods does Youyun Intelligent Computing support?

Containers can be accessed via Jupyter, SSH, FinalShell, and VS Code; Windows virtual machines also support remote desktop access. The specific methods of access depend on the type of instance.

Is the number of times the Coding Plan is called equal to the number of tasks?

It’s not the same; an Agent or a programming task may send multiple requests to the models, and different models have different scaling factors, so the actual number of tasks varies.

Does Youyun Intelligent Computing provide APIs and CLI?

Developers can use the CompShare API, UCloud SDK, CompShare CLI, and Agent Skill to manage instances, storage, and team resources.

Is Youyun Zhi Suo an open-source platform?

No, the cloud platform itself is a commercial service; the availability of the official CLI, Skills, and development examples does not mean that the backend source code of the platform is made public.