KAUST-NVIDIA GPU Hackathon 2017

What a GPU Blast!

We had a fantastic hackathon this year with 10 teams participating (from 1 to 3 members per team). 7 teams were from various divisions at KAUST and 3 external 1 from Aramco and 2 from King Abdulaziz University in Jeddah. The teams employed a variety of development ways on GPUs including CUDA (1 group), Libraries (2 groups) and OpenACC (7 groups). The codes were from a variety of domain sciences including seismic imaging , reservoir simulation, earthquake simulation, combustion, CFD and machine learning. More technical details can be found here.

The main development platform was a Cray Storm with 8 K80 GPUs. We also had a system from SuperMicro with 4 Nvlinked Pascal GPUs provided by KAUST IT Research Computing.

In only two days, 8 out of 10 teams were successfully running on K80 GPU, of which, two teams managed to experiment on the latest Pascal system which provided an additional 2-3X performance speedup. Few teams were very close to get impressive results, but the time was not enough. We plan to follow up with them and have a more regular session with them to keep the momentum on using GPUs for computing. 

And the winners are...

After the last report in the second day from the teams, we clearly saw the team working on SOFI3D, a community seismic imaging code, were the most successful using OpenACC with managed memory and got the highest speed up on K80 and even more on Pascal. The winning team was composed of  Nabil Masmoudi and Vladimir Kazei from SWAG group under Prof. Tariq Alkhalifah.
 

Hackathon Grand Prize winners

 
They will be looking at the possibility of improving this work, validating it even more rigorously to share it with the community.
 

Thank you!

Thanks for everyone who joined us for this great event,