

Top-down rendering inverts Nuke’s classic rendering method, rendering the graph node-by-node from the top of the graph down, rather than scanline-by-scanline on demand. Performance in Nuke is variable, depending on the script you are rendering, but in our internal testing scripts render 20% faster on average and some scripts render as much as 1.5x faster. Top-down rendering provides significant performance improvements to Nuke’s 2D graph processing. In this release we have introduced a new way for Nuke to render its node graph. See Train Neural Networks to Replicate Effects Using Machine Learning for more information. You can now run Cop圜at from the command line using the -X flag without any need to write additional Python scripts. If you have four GPUs and you set the Batch Size to two, Cop圜at only uses two GPUs.Ĭop圜at can now support more than four channels, as many as your GPUs and Nuke can handle, allowing you to train networks for a variety of more advanced use cases.įinally, we have streamlined the ability to kick off training on remote machines. For example, if you have two GPUs, Cop圜at's Batch Size must be set to two or higher to use both GPUs. Note:Multi-GPU processing requires a Batch Size greater than or equal to the number of GPUs you want to use. You can either speed up training by running Cop圜at on all your GPUs simultaneously or run different training sessions on each GPU, allowing you to train multiple models at the same time. Not only is single GPU training up to 30% faster, but you can also take advantage of setups with multiple GPUs. We have also removed any limitation on the number of training images and enabled the ability to train on headless Nuke with the -X flag. In this release, we have continued to improve our Machine Learning tools, enhancing the Cop圜at node with faster training, support for multiple GPUs and multi-channel training. Faster Machine Learning with Multiple GPUs See Visualize Unreal Scenes in NukeX and Nuke Indie for more information.

This helps with cross-operating system support when writing to shared network drives.


Monitor Out - AJA and Blackmagic Design.Import and Export Timeline Data with OTIO (Beta).Adjust Geometry Faster with Improved 3D UX.Faster Machine Learning with Multiple GPUs.What's New in Nuke, Nuke Studio, and Hiero 13.2.
