Posted by1 year ago
Archived
License Upgrades
- Prime/Free ($0/month, 2 GPUs, ~12 plugins)
- Subscriptions ($20/month, 20 GPUs + all 25 plugins + 2 nodes)
- Enterprise (200 GPUs, 10GE network LAN, NVLink support)
Adaptive Sampling with Octane Render for Cinema 4D Adaptive Sampling with Octane Render for Cinema 4D An excerpt from. OctaneRender 3 Lighting Passes Example 2-2. 5, 2016 (3 years). STAY IN TOUCH. Sign up for OTOY® updates and promotions.
With any of these licenses, you will be able to launch ORC/RNDR jobs; starting from the equivalent of running 2,000 OB per hour (~10x 1080 Ti's) on Prime/Free, up to 200,000 OB per hour (~1,000x 1080 Ti's) on Enterprise. The 20,000/200,000 OB/Hr limits on Subscription/Enterprise will enable massive render jobs to be sent to the network.
Octane 4 upgrades all v3 customers free of charge, so those categories will have a lot more users right off the bat. Yet, the biggest growth will no doubt occur in the Prime/Free category, bringing tens of thousands of new users to the ecosystem and further increasing the potential demand from content creators. Free Tier users can update to Pro/Enterprise via RNDR work, an example of the 'RNDR It Forward' concept first proposed by daffy_ch.
Brigade + Octane
- Octane 4 integrates all of Brigade, Octane's real-time pathtracer
- 100% new scene graph, feels like a game engine!
- Scene loading 10x faster than v3 (practically instant)
- Scene changes 100x faster than v3 (moving heavy objects often 60+ fps (2m07s)
The day that many of us have been waiting for is finally here: The merging of Octane + Brigade into Octane Render 4. For the uninitiated, Brigade was OTOY's first foray into the realm of real-time rendering. We saw little glimpses of it over the years, like this interior scene from 2013 being done on Brigade using unidirectional path tracing. It was running at 1080p / 25fps, relatively low noise on a couple of NVIDIA (Kepler) Titans -- completely unheard of at the time. Looking at the raw specs, in the five years since then we have gone from 2,688 CUDA cores@1GHz (Kepler) to 5,120 CUDA cores@1½GHz (Volta), or a factor 3x increase in perf/watt.
Fast, but still not fast enough for real-time, and this is where AI denoising comes in. An insanely interesting topic in its own right, it's a technique whereby Machine Learning algorithms, having been trained by millions of noisy/final render example pairs, essentially 'dream up' the correct image from having learned to generalize for novel data. This can greatly lower the requirements for reaching real time photorealism in the near future. More about this in 'AI Light + denoiser' below.
Here's the really exciting breakthrough for RNDR miners -- out-of-core (OOC) geometry! This comes back to the 10x faster scene loading. On Octane 3, when texture data would fill up the VRAM, parts of it could be moved to system RAM. This was known as OOC textures. However, the geometry was still confined to the VRAM. And because of the output buffer’s location in the system RAM, you had to constantly shuffle scene data back and forth on the PCIe bus to synchronize the passes between multiple GPUs. The larger the object, the slower it was to move it; almost like the inverse square law of mass.
The new file cache system in v4 will help remove the GPU memory load time bottleneck after first load of large geometry. You can now use use a significant portion of your system memory (as opposed to VRAM) for geometry data! In one stress test, OTOY tested the worst-case scenario of no assets in GPU memory (0 Kb VRAM used) and 40 GB+ geometry in CPU memory -- moving objects in the viewport was still interactive, at 60+ fps to boot! Multiple GPUs are also supported: in one test, OOC geometry scaled to 8x Volta GPUs + 150GB of scene geometry in CPU RAM.
Scene AI technology brings interactive positioning and modifying of geometry in real time, even in scenes with several millions of triangles. Through heuristics & intelligent tracking of topology, it can pull data in and out of core and determine the optimal configuration between system RAM and VRAM (e.g., 85% and 15%) to make the ray/path tracing faster. If it's something that's out of the viewport, it gets cached differently.
AI Light + denoiser
- AI Light is a learning system that improves as you render more samples. It doesn't 'cheat', but looks at the scene before casting any rays and intelligently picks which rays to render. It can dramatically cut down on light sampling time, especially in scenes that have many localized spot lights. AI Light is done in the renderer, fully unbiased, and tracks emissive points in real time. Will evolve over many releases, but already makes a 6x-10x difference in multi-point/spot light scenes
- AI denoiser is doing spectral floating point banners per wave length, per beauty pass before tone mapping or RGB rendering to viewport - very hard to top in a post-processing denoiser. It actually operates on internal perceptual models of material, it 'knows' that a chair is chair and a tree is a tree. Domain specific AI denoising is planned on real time, volumetric passes, lightmap denoiser, refraction, SSS and hair, etc.
- AI Light & AI denoiser are different, but very complementary, and the combination can be magical one-second renders. The magnitude of what is being accomplished here, at this quality, is quite shocking; a couple of GPUs are rendering and denoising a 50 sample per pixel scene, in ~1 second, for a super clean result! Due to AI Light providing information to the AI denoiser that isn't even visible to humans, the Machine Learning system can improve things drastically beyond what's possible with mere denoising.
This work is by no means finished, but expect rapid improvements in the speed and quality of AI denoising going forward.
RNDR SDK
- RNDR is OTOY's future framework for both internal and external development. All of OTOY software and services, including Octane, will be migrating to the RNDR SDK. This means RNDR will replace ORC, LiveDB (asset library) and other cloud services within 2018, in a yet unspecified release. --> all of that demand is going to RNDR miners
- A new toolset using GLSL/WebAssembly. Have backends for CUDA, x86, Vulkan, D3D, RTX and Metal. Octane 2019.x is being ported to the RNDR SDK, w/ 40% done already. RNDR Vulkan and Metal already as fast as CUDA in some early tests of OctaneBench, MacOS and Intel integrated graphics RNDR cross-compiler performance exceeds earlier CUDA cross-compiler work. --> Due to these optimizations, RNDR will support many devices & be hardware agnostic in the future
- Creators, artists and developers can also build and publish services on RNDR with the RNDR SDK -- published on the blockchain in ORBX package (same as ORBX scenes for RNDR jobs)
- In addition to Brigade + Octane APIs, RNDR standard runtime library has access to Light field and Extended Reality (XR) codecs for streaming. w/ all the denoising advances and other AI shortcuts made by OTOY recently, photorealistic rendering will be starting to move more to real time jobs on high-end GPUs to render holographic/volumetric content for mobile XR. OTOY are already half way there (a Tango device w/ an ARKit MR light field stream). This is a growing area of rendering that can use orders of magnitude more computing power than what's available today. Again, very good news for renderers.
tl;dr: Optimizations and new vectors of RNDR demand enabled by v4
- Rendering on Octane 4 is less VRAM bound; GPUs with less VRAM can now tackle more memory intensive scenes (but higher VRAM GPUs still get priority, if available). This means you want to invest in having enough CPU RAM to maximize the jobs that your node can accept
- Octane 4 upgrades all v3 customers free of charge --> more users with high-OB licenses. Free Tier category will bring new users to the ecosystem, further increasing demand from content creators
- Rapid improvements in both the speed and quality of AI denoising will lower the threshold for hobbyists with basic hardware for getting into Octane. --> future flow of jobs to the RNDR network is increased
- RNDR to replace ORC, LiveDB (asset library) and other cloud services in a 2018.x release
- You can RNDR on many devices in the future, support planned for iOS, MacOS & Intel integrated graphics, and more!
- Support for light fields enabled by the RNDR standard runtime library. 10x larger than lightmapping jobs
- Mobile XR experiences will be real-time rendered starting as early as next year
- Creators can build and publish services/modules with the RNDR SDK; rendering systems mixing AI/Brigade/Octane possible
7 comments
With Octane 3, the latest version of OTOY’s GPU based unbiased renderer, it really does feel that Octane is part of the big time with it’s relatively easy-to-use settings, adaptability and, of course, blazing fast render speed. With a road map which includes much better cloud rendering capabilities and the ability to use Octane on computers without NVidia cards (aka current macs), coming with v3.1., Octane has the potential to hugely expand it’s user base. Here are some top tips for working with Octane in Cinema 4D.
01. Get lots of graphics cards
Octane is the reason I built my own PC for the first time in 15 years. Octane is a GPU based renderer, which means that it uses the power of a gaming card for it’s render speed. Octane’s license is per machine, but you can build PC’s with up to 12 GPU’s (graphics cards) in them. At the moment, the cards need to be Nvidia, but they can be mixed and matched based on what is available
02. Network Render
Try to have a minimum of two graphics card in your computer when working with Octane, one to drive the screen, and one for Octane to use exclusively. What if your machine has only one GPU? For example, if you work on a laptop. Octane’s network rendering works really well. As long as you have a Octane License for every computer on your network, Octane can use the GPU’s on all the machines for every part of it’s process from Live Preview through to Final Render.
03. Use Octane standalone for material building
The Octane standalone application is definitely an acquired taste. However, that doesn’t mean that it should be ignored. Especially as due to the .ORBX file format, files can be swapped from the host application (in my case Cinema 4D) to the standalone app with ease. One the great things about the standalone is the UI which it provides to build materials which can be used across all Octane application instances.
04. Use Mix Materials
Mix Materials are hugely powerful as they allow combinations of simple materials to be overplayed into one ‘super material’. The mixing of the material can be controlled by procedural tools within Octane or image maps, and Mix Materials can be used within other Mix Materials to give near limitless control when creating textures within Octane.
05. Use LiveDB
When time is too tight to create a material from scratch, it is great to know that you can use the ‘LiveDB’ database of materials that come with Octane to create a starting point for what you need. From metals, fabrics, liquids and skin there will be a Material preset available within LiveDB to download into your scene to help you on your way.
06. Volumetric powerhouse
Octane 3 can now support volumetric systems like gas, and fire with full emissive and shadow capabilities. This is game changing when combined with tools like Turbulence FD for Cinema 4D. As both of these plugins are GPU-based, they are blisteringly quick allowing many more iterations of sim and render all within one PC.
07. Explore all of the Kernels
Don’t assume that the three different Kernels in Octane equate to fast, pretty fast and slow. Direct Lighting is my go to Kernel for animation, but it can create gotcha’s such as weird shadows. This is where I can explore PMC and Path Tracing, although on the surface both of these options seem slower, reducing the sample count can still create good results. They also both provide more realistic results, and offer extra features such as caustics.
08. Make the most of depth of field
Usually when using depth of field in CG Renders it sends the render times through the roof. This isn’t the case with Octane. At most, I have only ever seen an increase of 5-10% in render time by adding depth of field, but usually it’s a lot less and because its in the render, it carries through into compositing passes as well.
09. Use low samples for client preview
When working in animation, I try and use a few samples as I can when working, even for client preview work, as the great thing with Octane is that there are no tedious render buckets, just a complete iteration of frame with each sample. There are various post tools such as Red Giant’s Denoiser which are great for cleaning up noise, and when the work is signed off, just increase the sample rate.
10. Use the Node Editor
Learning to use the node editor, whether it’s in the Octane Standalone app, or more importantly within the host application, can make scenes much more efficient. This is especially true in Cinema 4D where the lack of a true nodal material system is a real hindrance. Also it makes it easier when working with Octane in different applications as the node structures are then consistent across your work.