In a 2021 hackathon structured around utilization of one of NVIDIA’s GPU acceleration packages, Senior ML Engineer Alan Reiner used CuPy to show that he could perform fractal generation and searching with significant speedup (>100x) over similar CPU-based processes, reusing much of the same syntax.

Novetta has hosted annual internal hackathons where employees create prototypes focused on technologies relevant to our customers. In our most recent hackathon, I utilized NVIDIA’s CuPy library for GPU-accelerated fractal generation and searching. By using a feature that allows users to write raw CUDA C++ code in a Python project without needing to create a C++ project or interact directly with the CUDA C++ compiler from NVIDIA, the CuPy library gives users access to much of the power of CUDA C++ in a much easier format. Fractals serve as a good example of the many kinds of custom GPU-optimized computations that can be implemented with minimal C++ experience.

In this report I will discuss how to generate fractals with CuPy and compare it to the more traditional CPU-based approach, demonstrate interactive fractal-searching, and (most importantly) generate an appealing wallpaper for your desktop… (continued in report)