GPU Computing Workshop

Friday 25 Mar 2022, 09:00 → 18:00 Asia/Taipei

Auditorium (一樓會議室) (Institute of Physics, Academia Sinica (中央研究院物理所))

09:00 → 10:30 AMD Session

AMD will first give an overview of AMD Instinct GPU, the development history, current technology and visibility for the future. The focuses of the second part are the GPU development environment ROCm, and an open software ecosystem for public access, so as accelerate the computing on Instinct GPU.

09:00 Exscale - AMD Instincts GPU, and an Open Software Ecosystem to accelerate computing

Speakers: Greg Oakes (AMD) , Robert Sheen (AMD)

10:30 → 11:00 Coffee Break

11:00 → 12:30 User Meeting

Speakers: Cheng-Han DU, Dynamic Computing Technology
(Topic: Merging AI/ HPC Research and Service)

Speaker: Yu TSAO, Research Center for Information Technology Innovation, Academia Sinica (Topic: TBC)

Speaker: Ting-Wai CHIU, Institute of Physics, Academia Sinica (Topic: TBC)

Manu George, Ins. of Physics, Academia Sinica (Topic: TBC)

11:00 Merging AI/ HPC Research and Service

Speaker: Dr Cheng-Han DU (Dynamic Computing Technology)

11:22 TBA

Speaker: Dr Yu TSAO (Research Center for Information Technology Innovation, Academia Sinica)

11:44 TBA

Speaker: Dr Ting-Wai CHIU (Institute of Physics, Academia Sinica)

12:06 TBA

Speaker: Dr Manu George (Institute of Physics, Academia Sinica)

12:28 Q & A

12:30 → 13:30 Lunch

13:30 → 14:30 Demo: DGX Staion

14:30 → 16:00 NVIDIA Session

GPU accelerated computing continues to provide excellent high performance computing support to help scientific computing, AI, DL, and ML computing continue to accelerate. Under the development trend in recent years, HPC simulation computing has also begun to use deep learning related methods. To improve computing performance and accuracy more efficiently, we can see that GPU accelerated computing is highly helpful to scientific research. Further research continues, GPU Computing, with the help of NVLink proposed by NVIDIA, has effectively improved the overall Multi-GPU In terms of computing performance, the computing of the multi-GPU servers is also taken care of by the Infiniband high-speed interconnect network that continues to improve performance, helping the GPU Supercomputer to provide better computing support.

Recently, NVIDIA has put more attention to use NVLink to enhance the communication bandwidth between the CPU and the GPU. It is expected that through the new generation of GPU accelerated computing architecture design, help high performance computing can be improved to a better state, and more problems can be solved in the field of scientific research. Accelerated computing requires more than just powerful chips. We achieve incredible speedups through full-stack invention, from the chips and systems to the algorithms and apps they run. In this session, we will share with you new generation chip, architecture design, full-stack software SDK and libraries and HPC compiler related information.

14:30 Trends in GPU Computing – NVIDIA

Speaker: Eric Kang (Solutions Architecture, NVIDIA)

16:00 → 16:30 Coffee Break

16:30 → 17:30 Keynote Talk: Science, Computing and AI