Resources
What is FlipFlop?
Most GPU developers don’t know how much energy their CUDA kernels actually use, or where the waste is. FlipFlop looks at the source code of GPU kernels and identifies patterns that waste energy, things like bad memory access patterns, inefficient execution flow, stuff you wouldn’t catch in a profiler easily.
It works at the code level (static analysis, no runtime overhead) and applies targeted transformations that cut energy without hurting performance. You don’t need to change your build system or runtime setup.
Accepted at ICSE 2026 Research Track.