Hacker News Reader

Comment by uguuo_o | Hacker News Reader

The amount of scientific work that could be done with such a machine would be staggering, specially if it can be local. I use 2-10k cores on a daily basis, having a dedicated machine with 1k would make debugging large scale problems much faster.

View on HN

> 2k-10k cores on a daily basis

could you tell more please? sorry, not in the loop on what a typical “scientific workflow” is like

Large-scale simulations of phisical systems tend to have a variety of length/time scales that require vast numbers of CPUs to model in a timely fashion. This could be anywhere from a few hundred cpus to tens of thousands. Fluid dynamics, solid mechanics, fluid-structure interactions, electrohydrodynamics, etc. simulation tools can all benefit from having a larger number of highly interconnected CPUs, we can simply resolve engineering problems better with more computational resources for better understanding of problems.

what language/frameworks orchestrate work across 10k cores? is that like, 10k threads? 10k processes?

It is not a single language. c, C++, Fortran are the most commonly used. The communication happens using one of many versions of the message passing interface [0] (MPI). MPI essentially spawns child processes and a communication ring between nodes using whichever interconnect is available (usually infoniband).

[0] https://en.m.wikipedia.org/wiki/Message_Passing_Interface

Reply on news.ycombinator.com