New volumetric AI denoiser
A second AI denoiser system is now exposed in the denoiser panel, specifically trained to denoise volumes (and volume passes) that may have required many many thousands of samples to become completely or nearly noise free in Octane 3 (note that AI light also helps AI volumetric denoising even more in scenes with many point light sources). Because volumetric passes have very low frequency details, the Volumetric AI denoiser should not be used with less than 1000 samples in the goal is to preserve details for final render quality that would be expected from a 2K to 10K sample render of the scene. Future versions of Octane will continuously improve volumetric lighting and denoising in through the already successful ongoing user submissions of ORBX test scenes (or ORC jobs) where denoising could be improved due to a corner case the AI has not yet been trained on.
New IOR texture input for Materials
Material IOR in the base layer of Universal Materials is no longer limited to scalar values, and can now be controlled procedurally with textures nodes and OSL shaders connected to a new 1/IOR texture input pin.
New 8:1 GPU texture compression
You can now compress RGB or greyscale image textures for rendering via the image import preferences. Compressed textures use just 1/3 to 1/8 the VRAM of normal RGBA textures while rendering. Compressed textures incur zero speed loss, and are often visually and perceptually identical to their uncompressed counterparts in many cases. Octane can now also load compressed DDS files directly into texture nodes where they will automatically preserve their specified compression settings when stored in GPU memory.
Import of individual layer from multi-layer EXR
Added support for loading an individual layer of a multiple layer EXR input file in image texture node import preferences. This can be used to re-load the multi-layer EXR output file of a render target or render job into an OSL post processing and compositing node graph in Octane. Such a system can further perform granular per render layer beauty and lighting pass compositing, denoising, blending, filtering, relighting/multi-light and color correction - with the full power and flexibility of OSL shader nodes.
New scene exporter
With this new “Render To Mesh” script node in the context menu under “Other” you can use Octane’s internal ORBX scene translation system to flatten and export scene geometry connected to this node to FBX / Alembic files with one click. Note that this system does not yet export volumes or other primitives which are not supported directly in the FBX / Alembic formats. We are working on adding the FBX material export but that work isn’t finished yet.
Fixes and other additions:
Improve OBJ loading speed by between 15-25%.
Fixed AI Light update for Network Render.
Fixed baking camera.
Universal material: Removed metallic influence on calculation of sheen.
Universal material: Fixed incorrect fake shadows with colour even when transmission is black.
Universal material: Fixed some incorrect darkening when using Octane BRDF.
Universal material: Fixed incorrect attenuation.
Universal material: Added 1/IOR setting to the base layer (see above).
Universal material: Fixed artefact on coating layer of universal material when changing roughness of the base specular layer.
Universal material: Split transmission colour from albedo colour into the new transmission pin.
Fixed a GGX sampling error.
Nishita sky: Fixed some very bright sunlight from Nishita sky model when the sun is just below the horizon.
Planetary environment: Improved appearance of star field
Planetary environment: Fixed rendering of sunset at high altitudes
Fixed a couple of issues with gizmos operating on scatter nodes.
Fixed scaling of gizmo handles on some camera projections.
Fixed a crash on the roughness info pass.
Fixed a crash when attempting to render hairs with zero vertices.
Fixed rendering of meshes with an opacity map inside the environment medium.
Updated OSL library to 1.9 and add support for the hash() function.
Fixed integer overflow in GUI library if images with very large resolutions are displayed.
Note nodes are now resizable.
Added CTRL+A shortcut to select all nodes in the node graph editor.
Fixed a hang when using some special characters in the text editors.
UI improvements for picking image files in the image tiles texture node.
Fixed max cores being set in the preferences window when unchanged.
Fixed artefacts when using high roughness on brdfs other than Octane.
Fixed displacement crash on some meshes, mostly alembic.
Added diffuse, refraction, reflection filters for universal material.
When manually compressing textures, these can now be saved as compressed also.
Fixed Beckmann sampling for the specular case.
Fixed universal material wrongly modifying roughness when transmission or metallic is greater than 0.
Fixed some artefacts in compressed textures.
Fixed some memory not being deallocated sometimes when using AI Light.
Some performance optimizations for AI Light processing
UI improvements for the image tiles texture node.
Fixed loading of Y/chroma EXR images
Speed up loading of .hdr files
Fixed universal material when using thin film interference with coating layer.
Added support for handling compressed images in the Lua API.
Fixed a bug that caused the render priority selection in device dialog to not do anything.
Improved how Octane deals with running out of memory, which fixes some cases where caching fails and then a render failure occurs.
A small volume rendering performance improvement of between 5-10%.
Fixed camera presets when there are volumes in the scene and some meshes are changed in the viewport using eg gizmos.
Fixed a significant performance regression from 3.08, where scenes with hundreds of unique meshes rendered significantly slower.
Render speed improvements - in general v4 is now faster than v3, and info channel kernels are between 15-25% faster.
Fixed internal connections being inconsistent when collapsing nodegraphs into a node pin.
Fixes and improvements for handling large nodes in the node graph editor: multi-connect now has a scrolling panel, and allows searching in the list of input pins. You can now expand pins by Ctrl-dragging items out of their pins.
Inside node graphs you can use ctrl+drag to create input and output linker nodes
Node graphs with a single output pin can now be collapsed.
Restored shortcut keys for collapsing and expanding.
Fixed a rare render failure when modifying meshes in the viewport either in scatter/placement, or with gizmos.
Fixed a permanent hang and a render failure when using emission on displacement triangles
Re-did performance tuning for all compute capabilities.
Fixed crash caused by invalid primitives with emission.
We no longer distribute the cuDNN library with Octane - when octane first loads, we download cuDNN if it has changed.
Hi, yes, the latest Blender for Octane use Octane 4 RC 2 but standalone is on RC 4.
You need a perpetual license to work with it, iirc.
I think the next update for BfO will wait until Octane 4 Stable.
Was only some bug fixing in RC 3 and 4.
Starting from this release, we no longer include cuDNN in the Octane download due to its size. Octane will now download it once automatically when started.
Image tiles texture node will now work for displacement.
Texture compression will no longer block the user interface, but when starting render, any compression jobs will complete before rendering begins.
Fixed incorrect behaviour when more than 64 render passes are enabled, which also caused a lot of error messages.
Fixed texture preview.
Fixed a crash when loading very small images.
Fixed light linking passes 7 and 8.
Fixed volume rendering when there are multiple volumes in the scene.
Fixed some specular material differences from 3.08.
Fixed universal material special case when BRDF is not Octane BRDF, IOR is 1.0, transmission is 1.0, with black diffuse.
Fixed denoise results when affect alpha option is enabled .
Fixed AI Light with animated triangles in the scene.
Fixed random colour texture.
Fixed a performance issue when using direct light kernel with a medium or volume where absorption is very low.
Improve OSL compiler performance.
Lua API: fix print() for node graphs.
Fixed an issue when exporting scatter nodes to FBX.
Fixed denoiser output when using affect alpha without an alpha channel.
Fixed a case where network render GPU count was incorrect.
You are probably going to get very different answers for that question, depending on what people are doing and what they want to accomplish. Some might complain about the cost which right now can be a barrier, although hopefully soon, Octane will be free for up to the usage of two GPUs.
I like Octane for many reasons, mostly because I jumped on board to using it very early (2010), this was before Cycles came out and the only other option for Blender was Indigo (and maybe Luxrenderer?). I really wanted to use an un-biased renderer and Octane was much cheaper than Indigo and much faster too. I couldn’t afford Indigo, which was too bad because I really liked it and had been using it since 2006.
The other reasons I stay with Octane and not move over to Cycles is that Octane is updated more quickly and in general runs faster. Octane supports more features than Cycles like out-of-core geometry, AI scene loading/denoising/lighting, Osl on the GPU, etc.
I really like how Octane handles displacement, where you don’t have to subdivide your polys to have it look good, it handles it automatically and you just choose what resolution you would like. It’s really fast and works well.
This is not to say that Cycles is bad, far from it, a lot of people use it to make really good images.
Just to note, Cycles now has adaptive diving via the OpenSubDiv library as of 2.79. All you need to do to switch the feature set from “supported” to “experimental”. It is still experimental because it currently lacks a couple of features like a geometry cache and subdivided UV’s.
Octane is now trying to prevent the computer to go to sleep while rendering.
Made a change in the render pipeline that hopefully increases the refresh rate at the beginning of rendering.
Octane prints now a warning in the log if the environment variable CUDA_VISIBLE_DEVICES or CUDA_LAUNCH_BLOCKING is set. The effect of these is that not all GPUs are visible, causing rendering to be slower.
Added support for emitters with “surface brightness” disabled to hair geometry.
Fixed bug causing hair geometry emission to not to work at all.
Fixed incorrect values displayed in the not-enough-memory error message of the denoiser (see viewtopic.php?p=346310#p346310).
OSL: Fixed detection of derivatives ( Dx and Dy intrinsics) which are not supported.
OSL: When you save an OSL node or scripted graph which currently refers to a file in an ORBX package, do not show an error dialog. Instead clear the file name attribute and save the code.
