SIGGRAPH 2024 BoF recap and demos

[randypacker]

Hello again!

It was another great SIGGRAPH this year, and our Birds of the Feather, kindly hosted by the Academy Software Foundation went quite well! Although it was not recorded, we wanted to share some of the highlights we presented in that session for folks who couldn't attend or would like to reference what we discussed.

First, we gave an update as to where Open MoonRay stands.

• Open Source over a year, we continue moving forward on variety of features and fixes for our productions and as reported and contributed in the community.
• We are regularly pushing updates at a good cadence
• Community reports of MoonRay running successfully on Fedora, RHEL, Ubuntu, Windows Subsystem for Linux
• Validated hdMoonRay Hydra delegate against public USD assets

Most encouraging are the number of folks who’ve have MoonRay running on systems completely outside of the DreamWorks studio environment and tooling. These efforts that flex MoonRay outside of our studio walls has contributed directly to improved build, configuration and documentation. Thanks to everyone who reports issues and solutions!

We’ve made great progress on improving our release process to streamline things both for internal and external contributions, and we continue to work on making it easy as possible to merge and release new updates.

We’ve added full support to MoonRay for Rocky9, and deprecated our build of CentOS7.  This keeps us up to date with the latest VFX reference platforms.  In support of that, in this list are a few of the platform updates that have come in over the past year to our most recent release as of today.

• Rocky9 Support
• Embree 4.2
• VFX Reference Platform CY 2020 through CY 2023
• USD 0.21.8 through 0.22.5

In terms of features, many of these are driven by specific production needs, but you can also see that we are are continuously looking to improve our performance characteristics for high-quality and high-demand rendering, such as with our CPU-affinity control and execution modes. We've also upgraded OpenImageDenoise to the latest version, which supports multiple GPU hardware in addition to the existing CPU denoising, and really allows artists to reach their time to first decision faster, while iterating on a scene.

We’ve added a new FishEye camera, as well as we’ve also added as new attributes to the SphericalCamera, for rendering projections onto the outside of a Las Vegas Sphere, for example.  We’ve also made several improvements to our utility toolset, including rdl2_print, rdla_filter, moonray and moonray_gui.

• New adaptive light sampling scheme, up to 5x faster in some scenes
• CPU-affinity control and custom thread pool
• Added GPU mode for denoising via OIDN
• Exposed the execution mode (auto, xpu, vector, scalar) in hdMoonray
• Multiple improvements to DwaToonMaterial for non-photorealistic looks
• New FisheyeCamera model
• Additional new features to the SphericalCamera
• Added some Geo preview settings to hdMoonray
• Improvements to several command-line tools
• Performance and accuracy improvements to our math lib

We’ve added a new telemetry overlay to multi-machine renders, which allows us to inspect the state of rendering in real-time. Here’s an example of the telemetry overlay, which provides specific stats to monitor the transfer of data between nodes, the mcrt progress, and memory and core utilization on all render hosts.  This is extremely useful for inspecting an Arras node cluster and its performance.

Contemporary Bathroom, authored by Mareck, curated by Benedikt Bitterli, rendered in MoonRay

To reiterate: our focus is carefully blended between improving render performance, and providing functionality to support production artists.

Additionally, with support from the Metal Ecosystem engineering team at Apple, we have MoonRay up and running on Apple Silicon with full support for XPU mode using Metal. Thanks to Apple Silicon and its large unified memory pool that both the CPU and GPU can access, we have the ability to render large assets that are difficult to render when the BVH doesn’t fit in the VRAM of a discrete GPU. We released this major update at SIGGRAPH in OpenMoonRay version 1.6.

We've begun work towards supporting the MaterialX standard nodes as native MoonRay shaders. This is the first step in our venture into MaterialX and more details will be coming in later releases as we continuously push updates. This chess set example shows MaterialX materials being rendered in Moonray in vectorized mode.

The Open Chess Set, authored by Moeen and Mujtaba Sayed, contributed to MaterialX by SideFX, rendered in MoonRay

We continue to iterate on our XPU mode and are adding support to process direct and indirect rays, as compared to our current implementation that processes occlusion rays.  This moves more of the processing onto the GPU and thus has a greater potential speedup in rendering.  We have achieved output matching with non-XPU mode, and are working on performance optimization work now.  This work is continuously integrated into our releases as we further our development on it.

We updated all of our core set of user documentation and improved everything, adding videos and images to show off various features and attributes.   We’ve added in search functionality to our documentation website, and this has become our new go-to for all things MoonRay, internally. 

We released a convenience utility to automatically convert textures in a USD scene to tiled, mipmapped format, as Moonray requires mip-mapped textures for rendering performance.  It handles most common image formats.

We also released a branch of Moonray that includes support for PresenZ, a technology that allows rendering of a feature quality scene into a format for playback on virtual reality devices.  Unlike standard 360 degree video, PresenZ renders all points of view visible from a user defined box which allows for full parallax as the viewer’s position changes.

Switching gears, we then went into demo mode. Our colleague @toshi-kato gave three live demos of MoonRay rendering, including this example running MoonRay on Windows Subsystem for Linux on a 2015 Dell Precision 5510 laptop.

• MoonRay on Windows (WSL Rocky9)

Followed by this demo running MoonRay on a M2 MacBook Pro.

• MoonRay on M2 MacBook Pro

And finally, the following two demos running MoonRay in multi-machine mode, using the two aforementioned Dell and Apple laptops as Arras dispatch, computation and merge nodes in a local, private wifi network.

• Multi-machine MoonRay/Arras Staircase scene (with new Fisheye camera)
• Multi-machine MoonRay/Arras Bathroom scene

More information about the setup for each of those demos can be found in the linked presentation file below.

Following those live demos, Toshi then spoke in more detail about a couple of interesting improvements to MoonRay.

1. A new shared memory buffer interface, which opens the door to creating completely decoupled applications without any dependency on MoonRay. Allowing for example an interactive, web-browser based GUI front-end.
2. Adding non-uniform memory access (NUMA) architecture support to MoonRay. This is a result of our continuous testing of high-end, new render servers and noticing a performance degradation in vector mode under NUMA architecture. We've successfully addressed and are releasing the adaptation for this architecture in a coming release and to be well-positioned for future hardware and performant rendering.

The presentation for these two topics, plus more information on the demos is in this attached file.

That covers the gist of the BoF this year and thanks for reading! We look forward to more improvements, new features and future events as we continue the rendering journey with OpenMoonRay!