Blender 4.1: Cycles

OpenImageDenoise GPU Acceleration

OpenImageDenoise is now GPU accelerated on supported hardware. This makes full quality denoising available at interactive rates in the 3D viewport. (#115045)

It is enabled automatically when using GPU rendering in the 3D viewport. It can be disabled in the denoising settings panel, to reduce GPU memory usage at the cost of slower denoising.

Supported GPUs

• NVIDIA GTX 16xx, TITAN V and all RTX GPUs
• Intel GPUs with Xe-HPG architecture or newer
• Apple Silicon with macOS version 13.0 and newer

For AMD GPUs, it is currently not enabled due to stability issues. However it can be tested in daily Blender 4.2 builds. There it is supported for AMD RDNA2 and RDNA3 generation discrete GPUs.

Benchmark

The denoising time of the Junkshop scene on various devices in seconds (lower is better):

Denoising time on various devices

Device	Time (seconds)
Apple M2 Ultra CPUL	1.035
Apple M2 Ultra GPU (76 Cores)	0.293
Intel i9-13900k	1.224
Intel Arc A770	0.067
NVidia RTX 3090	0.078

Other

• Option to disable bump map correction (36e603c430)
• AMD GPU rendering support added for RDNA3 generation APUs (d19ad12b45)
• Linux CPU rendering performance was improved by about 5% across benchmarks (PR#116663)