When Samsung SARC set out to build a new open-source GPU profiler for Android developers, the goal was to provide real-time insight into GPU behavior across devices and workloads. The team wanted to give developers deeper, more practical visibility into GPU behavior across real-world workloads and devices. That meant solving a problem that has long plagued graphics development: understanding what the GPU is doing over time, not just in a single frame.
To make that possible, Samsung SARC built Sokatoa on LunarG’s GFXReconstruct.
GFXReconstruct is the capture and replay engine that records graphics API calls with high fidelity and reconstructs them for deterministic analysis. That foundation enabled Sokatoa’s defining capability: multi-frame profiling. Instead of analyzing isolated moments, Android developers can now step through multiple frames of GPU rendering simultaneously, revealing performance issues and rendering problems that can otherwise be difficult and time-consuming to diagnose.
Why Multi-Frame GPU Visibility Matters
GPU bugs rarely exist in isolation. Performance regressions emerge over multiple frames, and visual defects are often triggered by earlier state changes. Without the ability to capture and replay exactly what the application specified, teams are forced to rely on guesswork and repeated trial and error. This challenge is not unique to graphics. Studies show that software developers spend an estimated 35–50 percent of their time validating and debugging code.* In GPU development, where issues are harder to reproduce and diagnose, that burden is often even heavier. GFXReconstruct removes that uncertainty by making GPU behavior observable, repeatable and inspectable.
Sokatoa’s support for Exynos, ARM and Qualcomm GPUs reflects the role GFXReconstruct plays as a hardware-agnostic foundation for modern GPU tooling. The engine supports workflows that extend well beyond profiling, including defect reproduction, interactive debugging, regression testing and bring-up for new hardware designs.
Built to Integrate. Designed to Pay Off.
What sets GFXReconstruct apart is the way it is designed to be used. It’s built to integrate directly into existing toolchains and developer ecosystems. LunarG works closely with hardware and software vendors to customize and embed the technology, allowing teams to deliver differentiated tools without rebuilding complex infrastructure from scratch.
The financial impact is substantial. Debugging, testing and verification are estimated to consume 50–75 percent of total software development budgets, representing more than $100 billion in annual costs.* Reducing uncertainty and time spent chasing elusive issues is not just a technical improvement; it is a material business advantage.
For hardware vendors and platform teams building GPU development tools, this approach is increasingly relevant. The Samsung Sokatoa collaboration demonstrates how the world’s leading hardware teams are approaching GPU tooling today: building differentiated developer experiences on proven, extensible infrastructure rather than reinventing core systems. To learn more about LunarG’s deep expertise in customization and integration, visit the official GFXReconstruct GitHub repository or review Vulkan SDK resources at vulkan.lunarg.com.
Learn More About GFXReconstruct
Our ongoing series, Mastering GFXReconstruct, can help you quickly get up to speed.
