T/AI 118.2-2022 Artificial intelligence computing center Part 2: Test methods
T/AI 118.2-2022 Artificial intelligence computing center Part 2: Test methods
Basic Information
Scope
Scope: "Technical Specifications for Artificial Intelligence Computing Centers Part 2: Testing Methods". This part specifies the testing methods for the performance and reliability characteristics of artificial intelligence computing centers, and is applicable to the testing and evaluation of the performance and reliability characteristics of artificial intelligence computing centers. It can also provide experimental basis for system optimization and bottleneck detection. Main technical content: The technical specifications for artificial intelligence computing centers aim to solve the following industrial problems. This part focuses on the fourth problem and the test-related parts of the first three problems. a) The system architectures and compositions of various artificial intelligence computing centers may vary significantly, and there are no baseline specifications for the computing characteristics and basic configurations of specific computing centers, which brings certain obstacles to the deployment, debugging, and porting of applications; b) The performance of artificial intelligence computing centers is one of the core conditions for their applications, and performance optimization should be based on testing. However, there are currently no comprehensive performance testing methods specifically for artificial intelligence computing centers; c) The reliability of artificial intelligence computing centers is an important concern for users. Currently, there are no technical requirements specifically for the reliability of artificial intelligence computing centers. The reliability requirements and testing methods in data center and software system standards also need to be adjusted and compromised to a certain extent; d) There are no unified standardized guidelines for the certification procedures of artificial intelligence computing center construction. In the certification process, inconsistencies in personnel, tools, use cases, and processes may lead to dev