On January 14, 2021, the Beijing Laboratory Service Guarantee Center released the tender announcement of the "Super-large-scale Artificial Intelligence Model Training Platform Project", with a budget of 346.37 million yuan.
Project Background: The equipment related to the ultra-large-scale artificial intelligence model training platform project purchased by the Beijing Laboratory Service Assurance Center is deployed in the Beijing Zhiyuan Artificial Intelligence Research Institute (hereinafter referred to as the Zhiyuan Research Institute), which is a new research and development institution in the field of artificial intelligence built in Beijing. Zhiyuan Research Institute has invested in the construction of a 300P computing power platform including 60 NVIDIA DGX A100 computing node servers, high-speed network switches, storage devices and management and scheduling software systems, etc., and relied on the platform to train the world's leading ultra-large-scale artificial intelligence model.
In order to ensure the compatibility and interoperability of the platform, the core equipment purchased this time should meet the following parameter requirements:
(a) Align with existing platforms
Os-X: Ubuntu 20.04.3 LTS, Ubuntu kernel 5.4.0-52-generic
(2) Not less than the following parameters
1. Graphics driver: NVIDIA Driver Version: 470.57.02
2. Graphics card memory: 40GB
3. Graphics VBIOS version: 92.00.36.00.04
4. Development tool CUDA package version: 11.4
5. Software Docker engine version: 20.10.7
6. Monitor GPU tool DCGM version: 2.2.9
Procurement List:
Technical parameters requirements:
