The Progressive GPU Lightmapper is a backend for the Progressive Lightmapper which uses your computer’s GPU and Dedicated Video Ram (VRAM) to generate baked lightmaps and Light Probes.
为了使用渐进 GPU 光照贴图，您的计算机必须满足以下最低规格：
如果您正在烘焙的场景所需的 VRAM 多于指定 GPU 上可用 VRAM，则烘焙时间会显著增加。有关可帮助您减少烘焙场景所需时间的信息，请参阅性能。
The Progressive GPU Lightmapper does not support OpenCL CPU devices.
In most configurations, the Progressive GPU Lightmapper performs much faster than the Progressive CPU Lightmapper. The performance depends on the selected Baking Profile.
However, the specifications of the GPU you use to bake and the amount of VRAM dedicated to that GPU both influence your experience of the lightmapper’s performance.
Other applications competing for GPU resources also have an impact on baking performance. Depending on your hardware vendor, reductions in the amount of VRAM available to Unity can slow down the bake process or cause it to fail.
In order to achieve results of the same quality as the Progressive CPU Lightmapper, the Progressive GPU Lightmapper needs up to four times more indirect light samples. The Progressive CPU Lightmapper uses an approach called branched path tracing, which launches more light rays at each bounce. The branched path tracing approach is not suitable for GPUs.
Unity provides a spreadsheet to help you determine how much memory the Progressive GPU Lightmapper needs to bake your scene. For more information see the Progressive GPU Lightmapper Memory Spreadsheet .
The Progressive GPU Lightmapper includes a feature called lightmapper tiling. Lightmapper tiling prevents the Progressive GPU Lightmapper from using all available GPU memory. Instead, Unity allocates some temporary memory on the CPU, then bakes your lighting in tiles that are sized to match the amount of GPU memory available. Unity configures the tiling based on the selected Baking Profile. The Progressive CPU Lightmapper does not include a tiling feature.
When the tiling feature is active, and when the lightmap atlas resolution is 512px or bigger, the baking process is likely to be slower than baking the same scene without tiling.
On macOS it is more difficult to determine how much memory is available. As a result, fallback to CPU is more likely on this platform and it is a best practice to follow instructions for optimizing your bake speed.
有多种方法可以减少烘焙时间并避免烘焙超过指定给 GPU 用于烘焙的 VRAM 的情况。
Close other GPU-accelerated applications. GPU-accelerated 2D image editing and 3D modelling software uses VRAM. Turn off accelerated graphics functionality or quit these applications.
Use Light Probes for smaller GameObjects. GameObjects such as debris or small props use up space in lightmaps but might not contribute significantly to the look of a scene. To optimize bake speed, disable Contribute Global Illumination for these GameObjects’ Mesh Renderers and light them with Light Probes instead.
Designate separate GPUs for rendering and baking. If your computer has more than one GPU, you can designate one for rendering and one for baking. See Configure GPU selection below.
Use a lower number of Anti-aliasing samples. The default value for this setting in the Lightmap Parameters Asset is 8. When you increase it, Unity uses more VRAM. When you are using a lightmap size of 4096 or above, this can quickly exceed the memory of many consumer GPUs.
Use fewer samples (Direct Samples, Indirect Samples, Environment Samples) and use the Denoiser to clean up the remaining noise in the lightmap.
Reduce your lightmap resolution.
如果您至少有两个 GPU，则可以指定一个 GPU 用于渲染场景，另一个用于烘焙光照。如果默认 GPU 没有足够的 VRAM 来渲染您的场景，并使用渐进 GPU 光照贴图对其进行烘焙，可能需要采用这种方法。
要更改 Unity 用于烘焙的 GPU，请执行以下操作：