Session VII: Into The Wild: Integrating non Von Neumann Computing Into HPC

A perfect storm of semiconductor lithographical challenges combined with power package limitations due to the death of Dennard scaling have resulted in the end times for von Neumann architectures.

But there is hope by using novel ways to compute.

Well researched and risk adverse approaches involve using fixed-function accelerators, including reconfigurable (FPGA) accelerators, for key kernels.

More risky are using machine learning techniques to drive surrogate models of expensive computation.

Even more risky are techniques that compute in analog.

And then there are emerging “fringe techniques,” including leveraging open-system thermodynamics, physical Ising-model based optimizations, and of course gate level quantum computing.

In this talk, I will review these novel computing approaches and discuss their applicability to HPC.

Download Presentation

Location: Grand Ballroom C Date: March 29, 2019 Time: 10:30 am - 10:50 am Tom Conte, Georgia Tech