A team from the University of Science and Technology of China (USTC) used 16 NVIDIA GPUs to pull off an SC first, winning both the first place prize for highest overall score and the highest LINPACK run at the SC16 supercomputing show in Salt Lake City, Utah.
Held every at the annual SC conference, teams build the most powerful cluster possible under the competition’s 3,120-watt (26-amp) power limit and then work for 48 hours straight to run them through various real-world applications.
Using eight Tesla P100s and eight GTX 1080 GPUs, the six-person team from USTC generated the competition’s top LINPACK score – a record 31.5 teraflops and more than triple the competition’s previous mark – its efficiency would have ranked it among the top machines on the annual Green500 list of most efficient supercomputers.
Asked for the biggest lesson he would take from this competition, Mingyao Shen, from China’s Huazhong University of Science and Technology, had a simple answer: “If we had more GPUs, we could have done better,” he said with a laugh.
Read more >
Winning Team Sets Record at SC16 Student Cluster Competition
Nov 21, 2016
Discuss (0)
Related resources
- DLI course: Speed Up DataFrame Operations With RAPIDS cuDF
- GTC session: Accelerate Clustering Algorithms to Achieve the Highest Performance
- GTC session: Advanced Multi-GPU Scaling: Communication Libraries
- GTC session: Enable End-to-End AI Infrastructure With ASUS AI Cluster Solutions (Presented by ASUStek Computer, Inc.)
- NGC Containers: milc
- NGC Containers: MATLAB
About the Authors
