Перейти к содержанию

YADRO updates AI server: now supports up to eight powerful GPUs

The new configuration is designed for training, fine-tuning, and running neural networks

Russian technology company YADRO has unveiled an updated version of its G4208P G3 AI server. It can now accommodate up to eight GPUs, each with a power of up to 600 W, and computational density has doubled. The server is designed for resource-intensive tasks related to training, fine-tuning, and running modern AI models, according to the company.

Image source: Grok

The updated version features enhanced power and cooling systems. The full configuration can operate at an inlet air temperature of up to +30°C. This allows AI equipment to be placed in existing data center infrastructure, reducing the need for additional site preparation.

The G4208P G3 supports accelerators from various classes and manufacturers. The specific configuration can be chosen based on the accelerator model, software stack, workload characteristics, and project budget.

YADRO also conducts applied research and tests servers under AI workloads. The company publishes measurement results along with descriptions of configurations, methodologies, and test conditions. The company stated that it focused on real-world operating conditions in data centers, the ability to choose a configuration for a specific workload, and further scaling of infrastructure as tasks grow.

Analogues of AI servers with GPUs

The Russian market for AI servers offers a range of solutions supporting powerful graphics accelerators. For example, iRU produces Rock G4212G08 and Rock G4204G10 servers, designed to accommodate up to eight and ten GPUs respectively, each with a power of up to 600 W. They are intended for machine learning and other tasks with high graphics accelerator loads.

Kraftway also offers its solutions. The Trusted TS3030 server is designed to accommodate up to six full-size GPUs with a TDP of up to 350 W and is aimed at artificial intelligence tasks, big data, and high-performance computing. Aquarius also develops server platforms with GPU support and, together with partners, works on integrated solutions for AI tasks. In 2026, individual AQserv servers based on Xeon 6 were also included in the register of Russian industrial products.

Read more on the topic: