Research Computing co-hosts deep learning event at SAS
Months of planning and preparation for the Deep Learning Symposium culminated in two days of sharing knowledge, fostering relationships, building new avenues of inquiry and exploring ways in which these technologies are transforming diverse scientific, engineering and business domains.
The symposium consisted of two components. The educational component featured workshops led by Nvidia instructors on topics including deep learning for health care and image analysis, genomics, image segmentation and classification, and neural networks.
In the second component, a series of presenters shared how these technologies are being used by faculty and students in higher education and by researchers in industry, such as SAS. The presentations spanned biomedical engineering, medicine, physics, nursing, mechanical engineering, and many other fields of study. They included talks on medical applications such as brain imaging, cancer diagnostics, and sepsis detection, as well as interesting applications in shark detection, decision making in gaming, and quantum many body physics.
“We hope this gathering and sharing of ideas and knowledge will generate new possibilities for collaboration and accelerate discovery,” said Michael Barker, Chief Technology Officer and Associate Vice Chancellor of Research Computing.
Research Computing provides UNC-Chapel Hill researchers substantial Nvidia GPU compute capability. Longleaf includes both consumer-grade and enterprise-grade cards. For consumer-grade GPUs, Longleaf includes five nodes each with eight Nvidia 1080GTX cards, comprising more than 100,000 CUDA cores. Longleaf includes 16 nodes each with four Nvidia Tesla V100s GPUs with NVLink—totaling 480 double precision TFLOP/s; or 960 single precision TFLOP/s; or 7680 Tensor TFLOP/s). In their own special purpose cluster, there are three Nvidia DGX-1 boxes (each has eight Tesla V100s with NVlink) and a DGX workstation (having four Tesla V100s with NVlink)—adding 210 double precision TFLOP/s; or 420 single precision TFLOP/s; or 3360 Tensor TFLOP/s.
“While GPUs are broadly popular in the scientific computing community, the Tesla V100s are the state of the art for deep learning use: the Tensor cores are tailor-made for it,” Barker said. “Hence, UNC-Chapel Hill is well-positioned to support our data scientists.”
Key Partner(s): Nvidia, SAS