Blender Cycles-X to be supported by NVIDIA CUDA and AMD HIP
Back in April, the developers of Blender announced that they have begun working on an improved renderer for modern GPUs.
Blender’s Cycles-X renderer will bring various improvements to the architecture and support for modern CPUs and GPU. The primary goal is to prepare the renderer for future development, which was overly complicated with the predecessor. The viewport will now be more responsive thanks to improvements in scheduling, timing and by enabling adaptive sampling. The rendering kernel received improvements for light bounces and shaders, which enables higher GPU occupancy and coherence.
Cycles-X still has many GPU performance related improvements to be implemented, however, the architecture has already proven itself by offering much faster rendering on NVIDIA RTX A6000 with OptiX, 3 to 5 times compared to Blender 2.93.
Render time on an NVIDIA RTX A6000 with OptiX, Source: Blender
The developer already has plans for future optimizations and architecture expansions such as volume rendering, shadow catchers, and multi-device rendering.
Blender confirms that Cycles-X renderer will be fully supported by AMD GPUs. The renderer will primarily focus on OptiX and CUDA APIs, as OpenCL has been deprecated. According to Blender, the OpenCL API which was used by AMD GPUs was always behind other implementations, so it was about time to focus the efforts on faster rendering solutions. Luckily, AMD came up with HIP (Heterogeneous-computing Interface for Portability), which is a C++ Runtime APU that creates portable code between AMD and NVIDIA GPUs from a single source code. This means that Blender developers can easily migrate CUDA code to HIP and enable Cycles-X support for AMD RDNA GPUs.
AMD RDNA2 (RX6000) and RDNA1 (RX 5000) support will come with Blender 3.0 The developer and AMD are already closely working together to bring HIP support for older GPUs and Linux operating systems.