From October 21 to 23, 2021, Beijing time, with the theme of "Intelligent Computing Empowerment · The 2021 CCF National High Performance Computing Academic Conference (CCF HPC China 2021) with the theme of "Win-Win Future" was held in Hengqin, Zhuhai.
Led by 5 academicians from the Chinese Academy of Sciences, the Chinese Academy of Engineering, and the Royal Academy of Engineering, 11 specially invited reports were used to conduct in-depth interpretation and discussion on many cutting-edge topics such as precision medicine, uncertain quantification and scientific computing, parallel sequence diagram algorithm, scientific research innovation based on the integration of memory and computing, supercomputing simulation of full-scale blood flow, earth system model of the Chinese Academy of Sciences, digital twin brain, etc. At the same time, more industry participants, Research institutes and enterprises have carried out a more in-depth combination of technology updates and practice development in many fields.
At the meeting, Li Panxiao, director of R&D of Chaoyi Jilun Platform, delivered a keynote speech. As deep learning is increasingly applied in different fields, Docker and Kubernetes are evolving and iterating as the mainstream choices for container cloud platforms. In the face of the development of container technology, Chaoyi Jilun further efficiently reconciles hardware technology and container products to meet customer needs for higher performance, user density and throughput.
The AI Max deep learning platform independently developed by Chaoyi Jilun, based on Docker+Kubernetes artificial intelligence container cloud platform, can achieve efficient management, scheduling and monitoring of heterogeneous resources, provides a complete process tool from model development, training to deployment, effectively simplifies the complexity of deep learning research, and improves the efficiency of artificial intelligence innovation and research and development.
At the same time, with the wide application of new generation digital technologies such as 5G, artificial intelligence, cloud computing, big data, and the Internet of Things, data centers have ushered in explosive development as an infrastructure for the digital industry and centralized information processing. However, high-performance computing brings super computing power at the same time, but also brings more energy consumption and carbon emission problems, how to improve energy utilization, reduce carbon emissions has become an important goal of the current development of high-performance computing industry.
In order to help more data center users solve the challenges of heat dissipation, energy consumption and carbon emissions, With 42 years of experience in IT overall solutions, Chaoyi Jilun has successfully developed and launched the flagship full liquid cooled GPU server - ServMAX® GL408-X2. The CDU can achieve high power bearing and support the heat dissipation of up to 24 high-power GPUs. At the same time, the GL408-X2 can directly switch the external CDU to a liquid-cooled data center CDU, making it more convenient to put on the shelves.
Cooling:
1. The thermal conductivity of liquid-cooled working fluid is 25 times that of air
2. Liquid cooling + air cooling centralized heat exchange, fan utilization rate is higher
3. Metal welded sealing cold plate, while improving heat dissipation performance, the product is lighter and more reliable, the flow required by the system is smaller, and it is efficient and energy-saving
4. Most of the heat is taken away from the server through the circulating medium, which greatly reduces the air volume required by the server
noises:
The server fan has the lowest duty cycle, significantly reduced noise, and the overall noise is 74dB at full load
stability:
1. The influence of the ring temperature is smaller, the chip working temperature is low, and the continuous high load operation is more stable
2, compared with air cooling, under the same volume, liquid cooling can support higher heating power consumption (that is, support greater energy density)
3. Under the condition of 25 °C, the CPU core temperature is lower than 79 °C, and the GPU core temperature is lower than 73 °C
Material:
1. Metal substrate welded sealing cold plate, high airtight grade, different from the rubber ring sealing products, completely bid farewell to the risk of leakage of aging sealing rings
2, aluminum heat conductive board, to ensure good heat dissipation efficiency under the premise of 52% weight reduction 3, combined with advanced thermal conductivity pad, thermal conductivity 3-5w / m * K, excellent wettability
devise:
1. The cold plate is light and thin, and the foundation thickness is only 7mm, which ensures the GPU spacing and ensures that the ordinary components in the server are ventilated and heat dissipated
2. Modular design, each GPU is independently connected to the liquid supplier, with a quick connector for flexible insertion and removal
3. The CDU refrigeration module and the server are connected by quick connectors, which is convenient for disassembly and maintenance 4, hose connection, ensures the flow rate and heat dissipation effect, and avoids the risk of stress fracture implied by the hard water pipe during the disassembly process