The key change is a new hardware platform, including six servers from YADRO. It includes specialized servers for deep learning, inference, and big data. Special attention is paid to the new G4208P G3 GPU server, designed specifically for artificial intelligence tasks.
The technical specifications of the new server are impressive: it is equipped with two fourth and fifth generation Intel processors and supports scalability up to 8 GPUs with NVLINK Bridge technology. This makes it easy to increase computing power and adapt the system to various tasks.
System performance has increased thanks to the refined Kubernetes build. Optimization has improved the efficiency of GPU resource utilization by 30%. This allows companies to process more data without increasing capacity.
The PAK-AI complex is designed for rapid deployment of AI infrastructure within the corporate network. The solution includes:
- Pre-configured software
- Tools for working with AI
- Data management system
- Application marketplace with its own SDK
- Support for various GPU and vGPU MIG formats
An important feature of the updated version is the introduction of the Yandex Cloud AI Studio platform for creating AI applications. The system supports working with various types of accelerators, including PCIe and SXM.
Security and compliance with regulatory requirements remain a priority. PAK-AI provides:
- High level of data protection
- Compliance with information security regulations
- Transparent data management
- Economic efficiency through built-in pricing
The flexibility of the solution allows companies to customize the system to their needs. PAK-AI can be scaled by increasing the number of server racks or replacing components with more powerful ones. The built-in marketplace allows you to manage existing applications and publish your own developments.
Industry representatives note the importance of the update in the face of growing demand for domestic AI solutions. PAK-AI is built on a Russian technology stack and meets modern business requirements for speed of implementation and efficiency of AI technologies.
The economic efficiency of the solution is manifested in reducing the cost of implementing AI technologies and accelerating the launch of business solutions through ready-made LLM models and pre-configured tools. Companies are able to optimize infrastructure costs without losing performance.