优云智算

Compshare is part of UCloud cloud computing (stock code 688158) and is a professional GPU computing power rental service platform that provides one-stop AI cloud services, covering the entire process from development and testing to production-grade deployment.

Purpose and Positioning The platform is aimed at AI training, model inference, and high-performance computing scenarios, providing GPU infrastructure for model, algorithm, and application developers. With second-level startup and fluid billing, it allows budgets to be converted more directly into computing power. It also provides model API services for application, Agent, and Coding developers, offering one-stop access to mainstream global models, covering image, video, language, and speech models.

Main Capabilities

1. GPU Resources and Billing: Rich GPU resources, no queuing required, ready to use immediately, flexible billing, and dedicated IP. It supports on-demand use and precise billing, with seamless scheduling in Agents via CLI, enabling minute-level delivery of training, inference, and applications. It supports second-level precise billing, no charge when powered off (data retained), and provides a no-GPU mode (0.15 yuan/hour). Pay-as-you-go can save 50% of costs, with no hidden fees.
2. Infrastructure Scale: It has two self-built data centers, with a scale of over ten thousand cards, elastic, flexible, and scalable, supporting minute-level expansion.
3. Environments and Images: GPU instances come preinstalled with mainstream environment frameworks, AI applications, and workflow images, ready to use immediately and starting in minutes. It provides Ubuntu and Windows systems, delivering complete operating system permissions and security isolation boundaries; it provides a Docker container environment preinstalled with mainstream AI frameworks and tools. It has 500+ community images, covering scenarios such as LLM, image, and video, and supports one-click deployment. Image types include official images, community images, and private images.
4. Storage Solutions: It provides a 30TB pre-downloaded public model library with over ten million IOPS capability, and users can directly call the public model directory. Free 100GB cloud disk, supporting online expansion, up to 8TB per disk, and up to 24 disks can be attached to a single instance. Data and computing power are decoupled, supporting shared mounting by multiple instances and long-term retention, facilitating team collaboration and cross-task reuse.
5. Network Capabilities: GPU instances come with free high-speed bandwidth, dedicated public IP, and accelerated overseas model downloads. A single availability zone shares 2Gb of external network bandwidth, with BGP three-line network (China Mobile, China Telecom, China Unicom). A single instance is bound to a fixed, dedicated IP, supporting direct SSH connection, custom TCP/UDP ports, ICP filing, and more. It provides download acceleration capabilities for global AI models and open-source websites, enabling one-click download of overseas resources without additional tools.

Usage Users can access GPU instances through SSH, Web Terminal, Jupyterlab, VSCode, and other methods. The platform supports seamless scheduling of GPU cloud services in Agents via CLI. Optional GPU instance models include 5090, 4090, 4090 48G, A800, H20, and others.

Applicable Scenarios Suitable for AI training, model inference, high-performance computing, one-stop AI cloud services from development and testing to production-grade deployment, and model access needs for application, Agent, and Coding developers.

Limitations Detailed information such as pricing, discounts, SLA, specific number of availability zones, and refund policy is not currently provided.