Skip to content

Details

CAPOW is a continuous-valued cellular automata explorer with update loops that look GPU-friendly: one work item per cell, small-neighborhood stencils, ping-pong buffers, and pointwise nonlinear rules.

This month, Richard Thomson will present a case study in using Alpaka (An Abstraction Library for Parallel Kernel Acceleration) to move a CAPOW-like update rule onto CUDA without porting CAPOW itself. We'll walk through the spike: CMake/vcpkg setup, CUDA language support, the CPU reference path, the Alpaka CUDA kernel, correctness checks, timing results, and the parts that hurt.

The first pass showed zero CPU/GPU error and strong update-only speedups, including about 102x at 1000 x 1000 and 174x at 1920 x 1080 for the diffusion rule.

We'll also discuss whether Alpaka is earning its keep compared with writing direct CUDA.

This will be an online meeting, so drinks and snacks are on you!

Join the meeting here: https://meet.xmission.com/Utah-Cpp-Programmers

Watch previous topics on the Utah C++ Programmers YouTube channel:
https://www.youtube.com/@UtahCppProgrammers

Future topics: https://utahcpp.wordpress.com/future-meeting-topics/
Past topics: https://utahcpp.wordpress.com/past-meeting-topics/

You may also like