Netease Intelligent News November 23 news, November 18, at the SC19 Global Supercomputing Conference held in Denver, USA, Inspur officially released a new AI open acceleration system MX1, which can support multiple OAM (OCP Accelerator) in one AI server. Module) Specification of the AI chip.

It is understood that MX1 uses high bandwidth, dual power supply and other technologies, and can be compatible with various AI acceleration chips that comply with OAM specifications. The MX1's chip interconnect has a total bandwidth of 224 Gbps, and provides two interconnected topologies: Fully-connected and Hybrid Cube Mesh. It is convenient for users to adapt to different neural network models according to the needs of chip communication. Design chip interconnect solutions.

In addition, for different power consumption differences of AI chips, MX1 has two independent power supply schemes of 12V and 54V. The maximum power of 12V power supply is 300W, and the maximum power of 54V power supply is 450W-500W. It can support various high-power AI accelerators. .

At the same time, the MX1 single-node design supports eight AI accelerators and can support up to 32 AI accelerators with high-speed interconnected Scale-up extensions, which can be used for AI computing requirements of ultra-large-scale deep neural network models. (Dingxi)

Source of information: NetEase Intelligence

Statement: this information is reprinted from authoritative news media. Reprinted for the purpose of transmitting more information and academic exchange, it is not used for commercial purposes, and does not mean to agree with its views or confirm its description. The content of this article is for reference only. If you violate the rights and interests of a third party, please contact us and we will deal with it as soon as possible.