Kohala is the cumulation of everything from the dual GPU days of yester year. All the pros and cons of the old implementations have been considered. The architecture here is not seen by the OS, the driver, or the game.
We started pulling on the thread we saw hanging in front of our faces. eGPUs with nothing assigned to them. Discrete cards, paired with powerful iGPUs sitting idle, where one never even touches the other. Workstations with secondary compute cards that games don't know exist. Billions of transistors, sitting cold.
Now we will give that back to the community. Regardless of if it may require a period where the project is paid.
Architectural specification and interface contracts drafted before a single line of implementation was written.
Built on Khronos-defined Vulkan layer extension points. No driver hacks, no undocumented APIs, no platform-specific tricks.
Windows and Linux. Both ship simultaneously. No platform gets a delayed or feature-reduced release.
A small group of graphics engineers and systems programmers with backgrounds across AAA game development, GPU driver teams, and low-level graphics research. Everyone here has shipped production graphics code.
Responsible for the overall layer architecture, module interface specification, and the orchestration engine. Previously shipped Vulkan-based rendering systems in commercial game engines.
Owns the frame split, composition, and synchronization subsystems. Deep background in GPU timeline scheduling and cross-queue barrier semantics in Vulkan 1.2–1.3.
Builds the hardware profiling layer, PCIe bandwidth measurement, and external memory transport plumbing that moves data between discrete GPUs without unnecessary copies.
Develops the trained orchestration models in the Pro Edition: compute duration predictors, mode-switching heuristics, and the session data pipeline that feeds them.
Maintains the layer loader, installer infrastructure, and platform-specific driver interaction on both Windows (DXGI interop, WDM) and Linux (DRM/KMS, Mesa integration).
Manages the public specification, module API documentation, and contributor onboarding. Ensures the Community Edition has no artificial gaps relative to the architecture.
These aren't marketing copy. They're the constraints we actually use when we make technical decisions.
The system measures what your hardware can actually do. If the bandwidth isn't there for Split Frame Rendering, it won't pretend it is. We surface the real numbers and act on them. We don't hide them to preserve the illusion of compatibility.
Every subsystem has a defined interface, a defined owner, and can be replaced without touching anything else. This isn't just engineering hygiene. It's how we intend to keep this project maintainable after it ships.
Architecture documentation and module interface contracts are written before implementation begins. If you can't specify it clearly, you don't understand it well enough to build it. This rule has caught more design problems than any amount of code review.
The Community Edition is a complete, honest implementation. It's not a crippled version designed to push users toward the Pro tier. If a feature belongs in the open specification, it ships in the open edition. Professional features are genuinely additive.
If you have deep experience with Vulkan, GPU timeline scheduling, or cross-device memory transport, and you want to build something that doesn't exist yet, we want to hear from you.
COMMUNITY · FREE | PROFESSIONAL · $29 | STUDIO · ENTERPRISE