7.3.5.0

7 Nov 2024

Arnold 7.3.5.0 is a feature release bringing USD improvements, faster volume intersection, faster baking, and Cryptomatte on GPU.

Enhancements

Optimized Cryptomatte and GPU support: Arnold now has an internal implementation of Cryptomatte that adds support for GPUs and improves performance on CPU. Cryptomatte now consumes less memory for samples, making it easier to use higher AA settings, progressive rendering, and adaptive rendering (when properly used with FIS). In a simple scene rendered at 640x480 resolution with three cryptomattes the below table shows improvements in peak memory. (ARNOLD-11042)

CPU	7.3.4.1	7.3.5.0
fixed-AA	1346MB	442MB
progressive	3433MB	532MB
adaptive	8584MB	441MB

GPU	7.3.4.1	7.3.5.0
fixed-AA	-	799MB
adaptive	-	810MB

Corrected clamping of OpenPBR metallic lobe: The OpenPBR metallic lobe Fresnel previously exhibited energy gain when the specular_weight parameter was increased above 1. This has been corrected so that the Fresnel conserves energy in this case, saturating to a mirror reflection in the limit of high specular_weight. (ARNOLD-15540)

Improved instancing in the alembic procedural: Instanced geometries are now better handled in Alembic. This fixes the known issues with make_instance, which is now enabled by default. (ARNOLD-15209, ARNOLD-8364, ARNOLD-8924, ARNOLD-7818)
Faster OpenVDB volume intersection: Intersecting OpenVDB volumes are optimized for many-cores. Scenes where volume::intersection is a sizable component of render time can see large speedups. One scene where it accounted for 73% of render time on a 128-core machine saw the volume::intersection time drop to 16% of total time, for an overall 3x speedup in render time. (ARNOLD-15448)
Faster uv_camera baking: The uv_camera.grid_size default is now 0 instead of 16, where 0 specifies that the grid_size is automatically chosen. Using the new default, baking a sphere with 800k triangles is now 6.5x faster than the previous default. (ARNOLD-15441)
Add more options settings to image metadata: Images with metadata, such as EXR files, now include arnold/AA_samples_max, arnold/AA_adaptive_threshold, arnold/fis_filter, and arnold/fis_filter_width metadata. (ARNOLD-15521)
Support element linking of array parameters in ramp shaders: Arnold GPU now supports the linking of individual ramp parameter components for the position, value, color, and interpolationparameters. (ARNOLD-7120)
OSL version 1.13.3.0: OSL has been upgraded to version 1.13.3.0. See OSL release notes for details. (ARNOLD-15319)
MaterialX OSL closure support: the OSL closures defined by MaterialX (v1.38) for definition of layered physical materials have been ported to Arnold. For example, all the closures needed to implement the OpenPBR are now available in OSL. See OSL Shaders for more information. (ARNOLD-15047)

_{Coated diffuse material created with OSL MaterialX closures.}

USD Enhancements

Hydra mode in the usd procedural: A new parameter hydra can be used to render the USD data through a Hydra framework. This supports interactive updates, and can be used to reproduce the same behaviour as a Hydra render delegate (usd#2109 usd#2107 usd#2019 usd#2017 usd#2003)
transform_keys primvar: When the primvar arnold:transform_keys is set, it controls the number of time samples used for the matrices of Arnold nodes (usd#2111)
Arnold Cameras in Hydra: Arnold cameras (vr, cylindrical, spherical, uv, fisheye) are now supported in Hydra. You create the Arnold cameras as USD built-in cameras, with a string primvar arnold:camera that specifies the Arnold camera type. (usd#1738 usd#2082)
Improved USD export that better supports Hydra: USD exports now support Hydra, so that scenes from Maya or 3ds Max now render as expected in Houdini Solaris.
- usd#2000 - Write light filters through node graphs so they can be rendered in Hydra
- usd#1965 - Write color manager attributes in the RenderSettings primitive
- usd#2008 - Write spot and photometric lights as UsdLux schemas
- usd#2030 - Write background and atmosphere shaders under a node graph for hydra support
- usd#2031 - Write AOV shaders under a node graph for hydra support
- usd#2025 - Write imagers through node graphs for hydra support
- usd#1987 - Author familyName and familyType in GeomSubsets written as USD
- usd#2080 - Author animated shader attributes in a way that they can render in hydra
- usd#2047 - Shaders exports should be bound to a material
Fixed mismatches between USD and Hydra renders: Several changes were done, either on the procedural or the render delegate, in order to have a consistent behaviour in both code paths.
- usd#1974 - Delegate should only create default shaders when needed
- usd#1959 - Improve translation of normals and primvars in hydra
- usd#1946 - Support color space in materialx for hydra
- usd#2042 - Follow hydra normals skinning behavior in the procedural
- usd#2055 - Support animated curves orientations in hydra
- usd#2053 - Visibility and sidedness attributes not supported in Arnold native hydra prims
- usd#2061 - Support arnold light groups in Hydra
- usd#2086 - Compute FOV in the procedural and hydra in a similar manner
- usd#2057 - Add ginstance support in hydra and fix a data race issue
- usd#2067 - Do not author useless "normals" user data in meshes/curves through the procedural
- usd#2010 - Support TreatAsPoint in spot lights
- usd#1979 - Support treatAsPoint in Hydra photometric lights
UsdPlane support: UsdPlane primitives were previously ignored, and are now rendered in Arnold. (usd#2058)
Improved support for ArnoldProceduralCustom primitives: ArnoldProceduralCustom primitives are now supported like other geometric primitives, including instancing and visibility settings. (usd#2064)
Profiling / Stats: Render settings can now specify filenames to output arnold profiling or stats files (usd#1118)

API changes

Node API: Added new API functions to declare an update dependency between a consumer node and a specific parameter on a producer node (a consumer can depend on a producer through multiple parameters). Added another API to clear the dependency between two nodes (for all parameters). Finally, the previous API is deprecated. The same behaviour can be replicated by using the new AiNodeAddDependency API and passing an empty or null string for the parameter, to create a dependency that covers all parameters (this should be used only when absolutely necessary). (ARNOLD-15483)
```
 AI_API void AiNodeAddDependency(AtNode* consumer, const AtNode* producer, const AtString param);
 AI_API void AiNodeClearDependency(AtNode* consumer, const AtNode* producer);
 AI_API AI_DEPRECATED void AiNodeAddDependency(AtNode* consumer, const AtNode* producer);
```
Limit node update recursion: Previously, node updates were propagated up the parent chain for all node types. We now do this for shape nodes only. This can improve node update time. (ARNOLD-15492)
Avoid unnecessary node updates: Resetting a node parameter to the current value will no longer cause a node update. (ARNOLD-15402)
Per-parameter update dependencies: Previously, a node update dependency caused a node to be updated whenever the dependency node was modified. We now define these dependencies based on parameters, so the node update is triggered only when one of the specified parameters on the dependency node was modified. (ARNOLD-15483)

Incompatible Changes

Missing dependencies: Custom nodes must declare all dependencies. Now that Arnold uses per-parameter dependencies, any missing dependencies will result in invalid data that may cause crashes or hangs. Also, shaders that depend on render camera values must declare a dependency on options.camera so that the shader is updated when a new camera is set. (ARNOLD-15483)

Bug Fixes

ARNOLD-11891 - Random crashes when using the Maya hypershade
ARNOLD-15542 - Crash when modifying light group array on a mesh with instances
ARNOLD-15478 - [GPU] Hangs and crashes with uv_camera
ARNOLD-13968 - Add execute permission to ADPClientService on Linux
ARNOLD-15454 - Support overriding a user property on an instance if defined already on parent
ARNOLD-11794 - Vector type shaders render black when assigned to a shape
usd#1982 - Fix subdivision when primvars are set in parent primitives
usd#1997 - Apply correct amount of transform keys when xformOp is set on parent prims
usd#1972 - Ensure subdivision is disabled when subdiv iterations is equal to 0 in Hydra
usd#2015 - Support hydra points with empty widths
usd#2084 - Imagers should be applied to all drivers
usd#2105 - Ensure the Arnold scene isn't modified after a Hydra batch render started
usd#1961 - Random curves width in Hydra when radius primvars are authored
usd#1977 - AOV shaders not set properly in hydra mode of the scene format
usd#1984 - Cylinder lights not taking normalization into account through USD
usd#1994 - Fixed hydra errors with varying topologies, and incorrect velocities in first renders
usd#1992 - Support hydra skinned positions with more than 3 keys
usd#2027 - Fix faceVarying normals interpolation in the procedural when the mesh is left handed
usd#1837 - Fix motion blur of instanced skinned geometry with animated parent matrix
usd#2037 - Improve instances and objects motion blur coherence
usd#2078 - Ensure the hydra render callback is always invoked
usd#2094 - Support material interactive updates in the procedural
usd#2092 - Fix interactive update issue when prims visibility is tweaked in the procedural
usd#2102 - Remove hydra warning subdiv_iterations: use type BYTE, not INT

System Requirements

Windows 10 or later, with the Visual Studio 2019 redistributable.
Linux with at least glibc 2.17 and libstdc++ 4.8.5 (gcc 4.8.5). This is equivalent to RHEL/CentOS 7.
macOS 11 or later.
x86-64 CPUs need to support the SSE4.1 instruction set. Apple Mac models with M series chips are natively supported.
GPU rendering and OptiX denoising work on Windows and Linux only, and require an NVIDIA GPU with the Maxwell architecture or later.
- On Linux, we recommend 535.104 or higher drivers.
- On Windows, we recommend 537.13 or higher drivers.
Intel OIDN GPU support is limited to:
- MacOS
  - Apple native CPUs (M1 and newer)
- Windows
  - Intel Xe dedicated and integrated GPUs
  - NVIDIA GPUs using Turing or newer architectures
  - AMD GPUs with RDNA2 (Navi 21 only) and RDNA3 (Navi 3x) architectures