OctaneRender® 2024.1 Alpha 1 Tweet

[abstrax]

Hi all,

We are excited to bring you the first alpha release of 2024.1.



The minimum NVIDIA driver version that is required by this version is 522. You can launch Octane with an older driver version, but it will fail to render and it will produce an error message in the log.

The minimum supported macOS version is 14. We are planning to bring back support for macOS >= 13.3 in some later version.

As with all experimental builds, we advise to not use this release for production purposes. We are trying to, but we can't guarantee that scenes saved with this version will be compatible in future releases. Changes are still being made daily.



What's New in Octane 2024.1 Alpha 1?

In this release of 2024.1 Alpha 1 we commit fully to hardware ray-tracing and unified the geometry data layout between CUDA and Metal, allowing mixed network rendering between platforms. Besides various smaller changes we also added a new Denoiser system and full texture and shader node support in the compositor:

• New post FX shader graphs: Texture nodes / OSL in the AOV compositor
• Powerful denoise AOV nodes and filter graphs (GPU and CPU)
• Ray-Tracing hardware acceleration speedups on Apple M3
• New geometry pipeline - less memory, better performance:
  
  
  • Smaller geometry memory footprint
  • Mixed platform network rendering between CUDA and Metal GPUs
  • Better performance with RT hardware acceleration on Metal and CUDA GPUs

New Post FX Shader Graphs: Texture Nodes / OSL in the AOV Compositor

You can now use any texture as a layer in an output AOV, with the new "Texture" output AOV layer node. The texture will be evaluated at each pixel, treating the image canvas as a quad with UV coordinates from (0, 0) to (1, 1). You can specify one texture for RGB and another for alpha, and choose the number of samples per pixel for each:



Any texture node can be used here, from an individual image file to an arbitrarily complex node tree. OSL textures can be used, which enables all sorts of custom effect layers and image generators to be used during compositing.

Using an ordinary texture is like blending a regular RGBA image, but you can also use the new "Output AOV parameter" texture node to build effect layers. An effect layer takes an image produced earlier in the compositing process and applies some texture operations to it. For example, you could combine some light passes together, and then use the new texture layer to run the result through a custom color correction OSL shader. The "Output AOV parameter" texture node is a placeholder to be used anywhere in the texture node tree where an image from the compositor should be substituted. These images will be produced during compositing and then made available during evaluation of the texture. Parameters are numbered - parameter 0 is the result of the layer underneath the texture layer, and additional output AOV layers to be made available as parameters during texture evaluation can be added to the texture layer itself.



Please check this demo scene containing the above examples: output-aov-texture-layers.ocs


Powerful Denoise AOV Nodes and Filter Graphs (GPU and CPU)

New denoise AOV nodes - adapted to OctaneRender from the Open Image Denoise framework - are now supported on both GPU / CPU and can be used to denoise any render AOV and create powerful and granular denoising filters in any composition AOV output graph, including per light / per beauty pass denoising and OSL driven denoising effects. It also works with the PMC and info kernels and all AOV outputs.

There are two ways to use the new denoise nodes: The first via a new mode selection "Denoiser" in the denoiser settings of the Imager node. There you choose "Open Image Denoise" and of course enable it. All other denoiser options work the same as with the Octane AI Denoiser with the exception of the option "Denoise volumes" which is ignored when using Open Image Denoise.

We also added an input "Prefilter auxiliary images", which does a noise reduction pass over the auxiliary AOVs (see below) before they are used to denoise the main AOV. This can improve the end result if the auxiliary AOVs are very noisy themselves. Open Image Denoise AOV nodes uses two auxiliary AOVs, which are accessible via the render AOV settings:

• "Denoise albedo AOV" which records the albedo values for the first bounce of the camera path, but also of subsequent bounces if the first bounces are all specular.
• "Denoise normal AOV" which records the normal values for the first bounce of the camera path, but also of subsequent bounces if the first bounces are all specular.

When you use Open Image Denoise in the imager those 2 auxiliary AOVs are enabled, but you can also enable these without the denoiser being enabled, so they can be used later during post-production / compositing.

Speaking of compositing, the second - and more interesting - way to use the new denoiser is via the compositor: There we have added a new blend layer node "Open Image Denoiser" which you can find in the category "Effects - post-processing". Because the denoiser can work without auxiliary AOVs it can be applied to any image / compositing result, but you can also give it the denoise albedo AOV and denoise normal AOV to improve the denoise result.

Here is an example scene based on the interior scene of OctaneBench, where the Open Image Denoise blend layer node is used to denoise the main beauty AOV, ambient light AOV, light ID AOV and light ID 2 AOV:
interior-oidn-demo.zip

In this screenshot you can see that it's set up in a separate pass-trough blend layer group so that it can be reused for the various output AOVs:



Giving the following output AOV results:




Ray-Tracing Hardware Acceleration Speedups on Apple M3

Octane X 2024 is being optimized for new ray tracing hardware on M3 Apple GPUS. This is scene dependent, but is already showing 2x to 12x(!) speed gains in heavily instanced scenes:

New Geometry Pipeline - Less Memory, Better Performance

We refactored the geometry pipeline, to speed up rendering, reduce the memory usage, unify geometry data on different platforms and prepare Octane for future developments. Those changes are mostly under-the-hood but also result in the following improvements:

Mixed Platform Network Rendering

Until now we used a different memory layout for geometry data on Metal (macOS) and CUDA. This made the compiled geometry data incompatible between macOS and Windows/Linux, prohibiting mixed platform network rendering. In version 2024.1 we unified the memory layout, allowing now network rendering on both platforms. Below you can see a screen recording where network rendering is enabled which adds a Windows render node with an NVIDIA GeForce RTX 4080 to the Apple M1 Max and bumps up the sampling speed significantly:



Please note that a fast connection between master and render nodes is still important, i.e. you should aim to have at least a 1Gbit/s connection.

We are planning to make regular Studio+ releases for macOS soon again, but until then we are enabling network rendering in macOS Prime builds, so that you can test and experiment with network rendering if you have a Studio+ subscription and a Windows/Linux PC where you can run your net render node.

Better Performance by Fully Utilizing RT Cores for Ray-Tracing

We now use RT cores for all mesh primitives including hair, spheres and displacement triangles. This speeds up rendering in most cases, because it reduces the overhead of having to deal with two different ray-tracing types and maximizes the usage of RT hardware. For example, our test scene based on Cornelius Dämmrich's 6088AD (https://corneliusdammrich.com/6088ad) renders on a GeForce RTX 4080 in about 20% less time (56s vs. 71s):



Since RT hardware really helps with ray-tracing motion blur, sometimes the speed-up can be more, like in the following (extreme) example where the render time was reduced by more than 80% (27s vs. 146s):



Smaller Memory Footprint on Windows/Linux

Not having to prepare geometry data for non-RT core ray-tracing hardware also saves host memory on Windows and Linux (on Metal device and host memory are unified).

As an example when you start rendering the 6088AD test scene, the memory consumption of the 2023.1 Standalone goes from approximately 7GB to about 28GB during geometry compilation and settles at about 19GB during rendering.

In the 2024.1 Standalone the system memory consumption goes from 7GB to about 15GB and stays there. This reduced memory footprint should help especially in Octane plugins since it reduces the memory pressure on the host DCC application.



All Changes Since Version 2023.1

New Features

• Added new blend output AOV layer node "Texture". This allows you to use any texture (including OSL textures) in the compositor (see above).
• Added new texture node "Output AOV parameter" which allows you to access output AOVs in textures (see above). If used outside of the compositor, it produces black.
• Added support for the Intel Open Image Denoiser (see above) to the Camera imager node and as an output AOV layer.
• Added 2 new visible beauty render AOVs which are used by the denoisers, but can also be used in the compositor as inputs for the Open image denoise output AOV layer:
  
  • Denoise albedo AOV
    
  • Denoise normal AOV
    
• Added a new "Apply LUT" layer for the "Composite texture".
• Added a new "Apply custom curve" layer for the "Composite texture".
• Added option "Direct light rays" to the kernel nodes that trace paths. This option determines how many light samples we calculate for the first high-roughness / diffuse bounce. Calculating more light samples can reduce noise, when many light sources are used and the direct light contribution is significant.
• Added new pin "Volume padding" to all medium nodes to expand the bounding box of a volume, if the "Sample position displacement" input is connected. This is useful if you are using sample position displacement, which can cause the displaced volume to go out of bounds.
• Added new node "Frame index" (under "Values" -> "Utilities") which allows you to fetch the frame index of the current project time plus an offset and use it for other operations.
• Added option "Round edges visibility" which gives control over whether the object layer geometry is visible to the rounded edges calculation.
• Volume to texture nodes can now use position based projections in object space. Loading of older scenes will convert any projections that are used by Volume to texture nodes to world space to produce the same behavior as in older versions.
• OSL: Added new intrinsic color _evaluateDelayedP(color inputTexture, point uvw) which changes the evaluation of the given input texture in the following ways:
  
  • Evaluation is delayed until the control flow reaches this intrinsic.
    
  • The evaluation of textures changes slightly, similar to how it would change for reading a Vectron shader with a Volume to texture node:
    
    • P no longer represents the world space position, but the position where the SDF is being evaluated.
      
    • P will be overridden with the passed-in uvw value.
      
  • Position based projections with the coordinate space set to "Object space" will use the passed in pos value. Projections set to "World space" remain unchanged.
    
• OSL: Implemented OSL functions getmessage() and setmessage(). I.e. you can now set a message of one of the built-in types (int, float, vector-like, color and matrix, but no arrays) and this message can be read in texture nodes that are executed afterwards. This includes non-inlined delayed shaders.
  
  • The use cases for these are very niche, but hard to work around if this feature is unavailable. They include:
    
    • Smuggling extra outputs out of shaders.
      
    • Communicating extra input data to a non-inlined function.
      
  • The implementation in OctaneRender comes with the following caveats:
    
    • The order of evaluation of inputs on built-in nodes is not guaranteed to be left to right. Some texture nodes like Triplanar map may not even evaluate all inputs.
      
    • Only passing messages within one texture evaluation is allowed. Octane doesn't evaluate texture inputs on material trees in a predictable order so passing around messages between different texture evaluations will not be implemented.
      
    • At the moment, all messages together may not take up more than 16 floats. This is to limit the maximal potential memory usage of launching a texture evaluation kernel. Only built-in types int, float, matrix, and vector-like are supported.
      

Improvements

• Overhauled geometry pipeline to unify geometry data layout for CUDA and Metal.
• Added support for network rendering on Apple, but since we are currently only making Prime releases for macOS you will have to run the net render node on a Window/Linux PC with a Studio+ license.
• Fully switched to RTX ray-tracing on Windows and Linux, i.e. all mesh geometry is now traced using RTX/Optix and uses hardware support as much as possible.
• Reduced memory usage during geometry compilation and sped it up.
• Added support for macOS 14, which is currently the minimum macOS version required. We are planning to bring back support for macOS >= 13.3 in a later version.
• Added preliminary support for hardware ray-tracing of static triangles on M3s in macOS 14. This is currently optional and you can enable/disable it in the render device settings dialog.
• Made the volume step lengths relative to the voxel size of the volume, i.e. "Volume step length" in the various medium nodes got replaced with "Volume step %" and "Vol. shadow ray step length" with "Vol. shadow ray step %".
• The environment medium and media in null materials are now treated as volumes, which means they are now affected by the emitter option "Visible on scattering volumes" instead of "Visible on diffuse". We added a compatibility mode "2022.1" that reproduces the old behaviour to all kernel nodes.
• Increased the maximum number of dirt textures per material from 4 to 8.
• Up to 8 dirt and curvature textures can now be used in global texture AOVs.
• Improved wavelength sampling for material/texture preview rendering which can reduce color noise in the preview render in some cases.
• Subtract one ray epsilon from SDF distance values so 0 width SDF objects work. This change was made because there have been multiple support requests where people expect abs(P.x) to render as plane (it should have been just P.x). For most SDFs this will not make a visible difference, unless the global ray epsilon setting is way too high.
• The geometry in multiple Scene node graphs that are created from the same file data and import settings is now treated as multiple instances of the same geometry. This reduces the memory requirements and fixes issues with nodes that behave differently when they are connected to different instances of the same mesh as opposed to different meshes, like the "Random color texture". Note that this matches the behavior already seen when copying a Scene node graph at runtime, which was previously not correctly preserved when saving and reloading the scene or when deleting the Scene node graph copy and undoing that delete.
• With photon mapping, do not restart the entire render when moving the render region after an edit.
• All imaging and post-fx operations are now applied after mixing the denoised and not denoised AOV, which makes results more predictable if the not denoised AOV contains a lot of hot pixels.
• Added input "Lock relative region" to the Film settings node. If enabled (the default), the relative film region parameters are kept constant while the absolute film region parameters get updated if the film resolution is changed. If disable, the absolute region parameters are kept constant.
• Geometry items of the node selection drop down menus were redistributed. The affected menu items are:
  
  • Geometry|Scatter -> Geometry|Scatter tools|Scatter (moved to the other Scatter tools)
    
  • Other|Geometry exporter -> Geometry|Utility|Geometry exporter
    
  • Geometry|Group -> Geometry|Utility|Geometry group (moved and renamed)
    
  • Geometry|Placement* -> Geometry|Utility|Placement
    
  • Geometry|Joint* -> Geometry|Utility|Joint
    
  • Geometry|Material map -> Geometry|Utility|Material map
    
  • Geometry|Object layer map -> Geometry|Utility|Object layer map
    
• The *Object Layer* menu items have been moved to the *Geometry* sub-menu in the node selection drop down menus. Affected menu items:
  
  • Object layer|Object layer -> Geometry|Utility|Object layer
    
  • Object layer|Utility|Object layer switch -> Geometry|Utility|Object layer switch
    
• OSL: Improved accuracy of function expm1() for values close to 0 (approximately less than 0.1) on macOS.
• OSL: Improved performance for certain OSL shaders that declare constant arrays.
• OSL: Using Triplanar UV projection now works in node trees connected to inputs of OSL nodes with delayed input evaluation.
• OSL: Clearer error message if an OSL texture or OSL camera shader has an _sdf output.
• Standalone: --no-gui can now be used without setting --script if you provide a scene and a render target node to render instead. If --no-gui is set, Octane will exit after rendering the render target of the specified scene, regardless of whether --exit is set or not.
• Standalone: --sign-out can now be used in combination with --no-gui.
• Lua: Changed default compressionType of octane.render.saveRenderPassesDeepExr() and octane.render.saveRenderPassesDeepExr2() from octane.exrCompressionType.ZIP to octane.exrCompressionType.ZIPS. Compression format ZIP was actually invalid and the Deep EXR export was falling back to ZIPS so this change has no observable impact.
• Lua: The octane.render.PROPS_RENDER_PASS_INFO table (returned by octane.render.getRenderPassInfo()) has a new field exrLayerName, which specifies the default EXR layer name that should be used when the AOV is saved in a multi-layer EXR image.
• Lua: Added new function octane.render.deepPassesEnabled() to check whether the kernel option "Deep render AOVs" is enabled.
• Lua: Exclude Cryptomatte AOVs when using octane.render.saveRenderPassesDeepExr() and octane.render.saveRenderPassesDeepExr2().
• Standalone: "Cloud" menu entry "Open RNDR..." has been renamed to "Open Render Network...".

Fixes

• Fixed incorrect rendering if a Standard surface material node was connected with multiple material inputs.
• Fixed incorrect medium extinction for light rays towards the backfacing side of a standard surface material.
• Fixed noise on materials with diffuse transmission in shadow AOVs.
• Fixed missing emission in volumes when the absorption is very low.
• Fixed render failure occurring in scenes with analytic lights and scattering media.
• Fixed render failure on macOS when you try to use a greyscale texture with a number of pixels that is not a multiple of 4.
• Fixed a bug where disc analytic lights are too bright.
• Fixed a possible render hang with the Volumetric Spotlight when the throw distance is zero.
• Fixed visible seams on triangle edges where the UV mapping changes direction when using strong bump maps.
• Fixed precision loss causing jagged rings in the image when using small camera distortion values.
• Fixed incorrect motion blur of VDBs. The underlying issue was that the motion vectors were applied in the incorrect coordinate space. To reproduce the old (incorrect) behaviour, in case some older scenes rely on it, we added a compatibility mode to the Volume node.
• Fixed scale of volume SDF, and mesh to volume nodes if not using metres as import unit.
• Fixed rendering of scenes with multiple copies of a skinned mesh and its joint nodes (e.g. because the Scene node graph containing them has been copied). Previously, only one of the copies was rendered correctly.
• Fixed occasional missing render updates after changing the custom AOVs or custom AOV channel in an object layer node.
• Fixed render failure when using interactive render region with deep image or Cryptomatte enabled.
• Fixed incorrect post processing chromatic aberration result on macOS.
• Fixed the post processing render AOV containing the full beauty render AOV when post processing chromatic aberration is enabled, instead of just the difference between the beauty render AOV with and without post processing.
• Fixed issue causing the first render device to always be enabled for rendering in render nodes, regardless of the device selection on the command line.
• Fixed loading of packages on Windows and macOS which contain asset file names where only the case of some characters is different.
• Fixed insufficient error logging during uncompressing and CRC checking of the downloaded cuDNN library.
• Standalone: Fixed crash when starting Octane without a physical display, even when --no-gui was provided. On Linux this crash could also happen if the DISPLAY environment variable did not refer to a valid X server.
• Standalone: Restored options "Uncompressed" and "RLE compression" options when saving deep EXR.
• Standalone: Removed spaces from Cryptomatte EXR layer names when batch rendering.
• Standalone: Fixed the first frame of batch rendering being wrong if the current frame was not 0 before running the script.
• Standalone: All deep pixel AOVs are now saved, if kernel option "Deep render AOVs" is enabled and Save all enabled passes" and "Save additional deep image" in the batch render script and batch render job. Before that change, only the deep pixel data of the main AOV would have been saved.
• Standalone: Output and denoised AOVs are now excluded from deep pixel exports.
• Standalone: Fixed saving of duplicated frames from render jobs.
• Standalone: Fixed project being flagged as dirty right after loading, if it was saved at a nonzero frame / time.

2024.1 Features in Active Development

• Light Field Baking: high performance reflectance field caching for rendering Octane scenes at 120 fps with final frame quality - used for Apple Vision Pro, mixed reality,virtual production
  
  https://www.youtube.com/watch?v=ixjwQIf7iqo&t=1634s
  
  
  
• Brigade Temporal Denoiser - real time filtering for path traced rays (in 2023 core, but 2024 will expose to plug-ins, starting with UE5, Blender 4.0 and C4D)
  
  
  
• Meshlet Texture Streaming
  
  https://youtu.be/pMcDrWKuAtc?t=121
  
  
• OctaneGPT at your service!
  A large Language Model service that can provide help inside Octane - covering all documentation, tutorials and even code - the service can create nodes, scripts, shaders and generate scene and assets interactively form text, voice and even screen sharing:
  
  

  

  
  

  

  
  
• Localized language support
  
  

  

  
  
• Material X in core - and MATX looks and deeper USD and Hydra 2.0 integration in progress:
  
  

  

  
  

  

  
  
• Unreal Engine style decal support in core
  
• Rest position nodes

What's on the 2024.x Roadmap?

• Headless rendering with multi-render and multi-engine live linking (Blender, UE, C4D and standalone)
• Neural rendering: NeRF, Splats and generative AI scene nodes
• Meshlet geometry streaming
• New Render Network manager for live and offline GPU workloads
• Volumetric live compositor for virtual production and mixed reality and light field displays
• Improved BRDF and shader support for faster single sample quality output
• Improved LensFX system building on 2023 framework and 2024 shader graph system for post processing

Downloads

Downloads for Studio+ Subscription Users

OctaneRender Studio+ 2024.1 Alpha 1 Standalone for Windows (installer)
OctaneRender Studio+ 2024.1 Alpha 1 Standalone for Windows (ZIP archive)
OctaneRender Studio+ 2024.1 Alpha 1 Standalone for Linux
OctaneRender 2024.1 Alpha 1 Standalone for MacOS (not ready yet, will be provided as soon as possible)

OctaneRender Studio+ 2024.1 Alpha 1 Node for Windows (installer)
OctaneRender Studio+ 2024.1 Alpha 1 Node for Windows (ZIP archive)
OctaneRender Studio+ 2024.1 Alpha 1 Node for Linux

Demo Downloads
OctaneRender Demo 2024.1 Alpha 1 Standalone for Windows (installer)
OctaneRender Demo 2024.1 Alpha 1 Standalone for Windows (ZIP archive)
OctaneRender Demo 2024.1 Alpha 1 Standalone for Linux (ZIP archive)


Happy rendering
Your OTOY team

[merid888]

Thank you for new update OctaneRender team.

[Zay]

Is this only for RTX cards? Octane froze when I tried with my GTX 1080Ti. Although it worked in Direct Lighting but not in Path Tracing mode.

[john_otoy]

Is this only for RTX cards? Octane froze when I tried with my GTX 1080Ti. Although it worked in Direct Lighting but not in Path Tracing mode.

I just tested path tracing on a 1080Ti using driver 537.58 and it's working. Are you on a recent driver?

[Zay]

I just tested path tracing on a 1080Ti using driver 537.58 and it's working. Are you on a recent driver?

I'm using 546.01. My system has a RTX 3080Ti (which works) and 2 GTX 1080Ti's (one on a PCI-E extension cord).

[john_otoy]

I just tested path tracing on a 1080Ti using driver 537.58 and it's working. Are you on a recent driver?

I'm using 546.01. My system has a RTX 3080Ti (which works) and 2 GTX 1080Ti's (one on a PCI-E extension cord).

I installed 546.01 to check and this is working as well. It could be related to your multi-GPU setup.

Is it happening if you start a new scene with a render target set to path tracing? Are you seeing any log errors?

[Zay]

I installed 546.01 to check and this is working as well. It could be related to your multi-GPU setup.

Is it happening if you start a new scene with a render target set to path tracing? Are you seeing any log errors?

New scenes seems to run fine.

It seems to be a denoising problem but only on heavy scenes. So far I can render without enable denoising, but as soon as I enable Octane AI denoising it renders a few frames and then freeze up. I manage to get an error while "denoising" was happening and froze:
OctaneRender Studio+ 2024.1 Alpha 1 (14000001)

denoiserThread2: CUDA_ERROR_ILLEGAL_ADDRESS: an illegal memory access was encountered. Restart required
CUDA_ERROR_ILLEGAL_ADDRESS: an illegal memory access was encountered
denoiserThread2 : Exited with a error. Restart required

But usually it never gets to "denoising" before it freezes up. No errors show when that happens.

There are no problems in v2023.1.

[Zay]

I installed 546.01 to check and this is working as well. It could be related to your multi-GPU setup.

Sent you a scene in PM.

And here's another from the error log:

OctaneRender Studio+ 2024.1 Alpha 1 (14000001)

OptiX error for device 2: [ERROR] Failed to copy internal launch buffer (CUDA error string: an illegal memory access was encountered, CUDA error code: 700)
Error synchronizing on an outstanding request (CUDA error string: an illegal memory access was encountered, CUDA error code: 700)
Error recording resource event on user stream (CUDA error string: an illegal memory access was encountered, CUDA error code: 700)
OptiX API error for device 2: CUDA error
device 2: path tracing kernel failed
denoiserThread2: CUDA_ERROR_ILLEGAL_ADDRESS: an illegal memory access was encountered. Restart required
CUDA_ERROR_ILLEGAL_ADDRESS: an illegal memory access was encountered
denoiserThread2 : Exited with a error. Restart required

[Hesekiel2517]

Thank you very much. This is all very exciting!

In the 2024.1 Standalone the system memory consumption goes from 7GB to about 15GB and stays there.

It's the other way around, right?

[ricky_otoy]

Thank you very much. This is all very exciting!

In the 2024.1 Standalone the system memory consumption goes from 7GB to about 15GB and stays there.

It's the other way around, right?

Hi there, no for that particular scene, it doesn't go over 15GB during geo compilation in 2024.1
Usually, you'd find RAM usage would go up quite substantially before dropping off - whereas in this case it should use the appropriate amount.