Northeastern University’s group that is mapping number positions in the hexagon won the Computer and Information Sciences award at yesterday’s RISE 2018 expo!
Their GPU/CPU based system has some of the fastest technology available, such as NVIDIA V100 GPUs (the ones on the table a similar too, but not the high end working ones) which, I believe, have 7.5 TFLOPS (double precision) of computing power EACH.
Their project is scalable – so other GPUs could jump in via the web to help map the hexagon. Pi’s powers have been numbers of particular interest to me, so each data point is over three times as large as the last. Before long, the points are difficult to compute even with this monster. So scalability and distributed computing might allow us to find pi powers’ positions at large values.
The project has had some bumps along the way, such as memory overflow issues, but appears to be producing good data now. They’ve not yet overtaken the data I compiled with my basement Mac Pro cluster last year (I’m a little proud of that), but hopefully I’ll soon know if pi^34 and pi^35 roll double – and if any other numbers stand out in my “Numberscape” (more to come on that).
**************************
September 23, 2018
In May, Kaustubh, the graduate student leading the computation effort left for a co-op, and many of the undergraduates helping out left for summer break … and then there was a software bug, and then someone stopped the system for something.
However, Julian Gutierrez, the very first graduate student to work on this has returned, the GPUs are hot at work and he even rounded up 16 more (but slightly vintage) GPUs to begin mapping. Once again I’m hopeful, baring other unforeseen events, we may see a another roll in a few weeks.