Update on Isaac sim port
Hey everyone. I wanted to give a quick update on Void Compute and the mission to get Isaac Sim running natively on AMD hardware. I hit a massive wall with the original plan of full OptiX emulation. After digging into the telemetry logs and the RDNA 3 instruction set, it became clear that trying to emulate an NVIDIA proprietary compiler at the transistor level is just falling for their trap. It is slow and it creates a performance bottleneck that wastes the raw power of the 7800 XT.
I am officially pivoting the project architecture of Phase 2. We are no longer trying to be NVIDIA. We are building a compatibility middleware that bypasses the proprietary OptiX moat entirely. Instead of emulating the silicon, I am targeting the Hydra Render Delegate. This allows the engine to talk natively to AMD Ray Accelerators through the Vulkan RT API. By forcing the Vulkan delegate and using rasterization with a native ray tracing pipeline, we get hardware intersection testing without the CUDA overhead.
I am also ripping out the PhysX backend. PhysX is designed to gatekeep performance on non NVIDIA hardware. I am replacing it with the Newton Dynamics solver because it is backend agnostic and deterministic. For the final visual output, I am using Intel OIDN 2.2 with a HIP backend to handle denoising. This provides visual parity with OptiX while staying within the open standard ecosystem.