Qooqa Tech has upgraded its computing infrastructure to support its AI recruiter, aiming to handle increased request volumes and deploy larger language models. The company opted for a dedicated server solution provided by ITGlobal.com, which integrates three Nvidia H200 accelerators into its existing environment.
The new setup features two Nvidia H200 units connected via Nvidia NVLink Bridge to process a primary language model with approximately 122 billion parameters. A third accelerator is dedicated to a secondary language model and embedding tasks, allowing the system to manage multiple components on a single server.
ITGlobal.com integrated the dedicated hardware with Qooqa Tech's existing virtual infrastructure using L3 connectivity based on VMware NSX. This configuration enables high-speed data exchange and routing between the cloud and dedicated segments, meeting the client's requirement for local hosting in Russia.
Current performance metrics indicate the system processes over 600,000 requests per day. The infrastructure achieves a generation speed of 5,000 tokens per second with a median response time of approximately one second, supporting the company's transition to more complex AI models.
