This paper addresses the design of GRID systems to solve resource-intensive problems, particularly focusing on the structural-parametric synthesis of neural network models. The authors develop new mathematical models for assessing the performance and reliability of GRID systems and propose a multi-criteria optimization model for designing these systems. They also introduce a decision support system (DSS) that uses a multi-criteria genetic algorithm to optimize the configuration of centralized GRID systems. The study employs evolutionary optimization and queuing theory to enhance the efficiency of distributed solutions. The proposed approach is validated through computational experiments, demonstrating its effectiveness in selecting an efficient structure for a centralized GRID system. Key contributions include: 1. New mathematical models for evaluating the performance and reliability of GRID systems. 2. A multi-criteria optimization model for designing GRID systems to solve high-resource computing problems. 3. A DSS for designing GRID systems using a multi-criteria genetic algorithm. The research aims to improve the efficiency and reliability of GRID systems in solving complex scientific and technical problems, particularly those involving the structural-parametric synthesis of neural network models.This paper addresses the design of GRID systems to solve resource-intensive problems, particularly focusing on the structural-parametric synthesis of neural network models. The authors develop new mathematical models for assessing the performance and reliability of GRID systems and propose a multi-criteria optimization model for designing these systems. They also introduce a decision support system (DSS) that uses a multi-criteria genetic algorithm to optimize the configuration of centralized GRID systems. The study employs evolutionary optimization and queuing theory to enhance the efficiency of distributed solutions. The proposed approach is validated through computational experiments, demonstrating its effectiveness in selecting an efficient structure for a centralized GRID system. Key contributions include: 1. New mathematical models for evaluating the performance and reliability of GRID systems. 2. A multi-criteria optimization model for designing GRID systems to solve high-resource computing problems. 3. A DSS for designing GRID systems using a multi-criteria genetic algorithm. The research aims to improve the efficiency and reliability of GRID systems in solving complex scientific and technical problems, particularly those involving the structural-parametric synthesis of neural network models.