18 September 2024
Maya 2025.3 includes MtoA 5.4.5
MtoA 5.4.5 uses Arnold 7.4.3.1 and is a minor feature release bringing a new OpenPBR surface shader, improvements to volume rendering, OIDN support for Apple Metal and AMD GPUs, and many other improvements and bug fixes.
These release notes include all the updates and changes in MtoA since Maya 2025.2, which included MtoA 5.4.2.1.
Download Arnold for Maya from your Autodesk Account. See Download Arnold for more information about downloading MtoA from your Autodesk Account.
Follow these installation instructions.
OpenPBR enhancements - Improvements to the conversion tools and the OpenPBR shader user interface make it easier to convert scenes that use aiStandardSurface or the Maya Standard Surface to OpenPBR. (MTOA-2104,MTOA-2086,MTOA-2062,MTOA-2054)
OpenPBR Surface specification used for viewport shading - The viewport representation of the aiOpenPBRSurface node now uses the OpenPBR Surface specification to represent the shading (MTOA-2038)
| Smooth Coat | Rough Coat | |
|---|---|---|
| 7.3.3.0 | ![]() |
![]() |
| 7.3.4.0 | ![]() |
![]() |
SSS color matching: The SSS model has been tuned to ensure that the color of dense SSS more closely matches the supplied subsurface_color, as required by the OpenPBR specification. (ARNOLD-15328)
The following images compare diffuse base color with the subsurface colors of Standard Surface and OpenPBR Surface:
base_colorsubsurface_colorsubsurface_color
Scaling of emission luminance from nits to Arnold units: nits_per_unit was added to the options node, and OpenPBR Surface's emission_luminance (in nits) is internally scaled down by this nits_per_unit value (defaulting to 1000) to bring it into Arnold units. This ensures that the default soft-maximum of emission_luminance of 1000 nits corresponds to the former behavior of Standard Surface at emission 1. (ARNOLD-15364)
Faster startup times on GPU: We have made several optimizations to improve startup and scene update time on GPU. These improvements will be most noticeable on scenes with many objects. Startup times on Windows are also now on par with startup times on Linux. On a scene with 1 million instances, we see the following improvements when compared to the previous release. (ARNOLD-15271, ARNOLD-15279, ARNOLD-15310)
| 7.3.3.1 | 7.3.4.0 | Improvement | |
|---|---|---|---|
| GPU update time (Windows) | 204s | 17s | 12X |
| GPU update time (Linux) | 37s | 18s | 2X |
Faster texture opacity lookups in OSL shaders: OSL shaders used in opacity inputs (for example OpenPBR Surface's geometry opacity) now read opacity mask textures faster and with smaller memory footprint. (ARNOLD-15340)
Improved node and parameter introspection with kick
kick -info <node> now groups node parameters by their UI Group metadata, or by whether the parameters are Input or Output. (ARNOLD-14198)
For example, here's the first few UI groups of openpbr_surface:
$ kick -info openpbr_surface
node: openpbr_surface
type: shader
implicit output: CLOSURE
parameters: 65
multioutputs: 0
filename: <built-in>
version: 7.3.4.0
Type Name Unlinkable Default Bounds
------------ ------------------------------------ ------------- ---------
Base
FLOAT base_weight 1 [0, 1]
RGB base_color 0.8, 0.8, 0.8
FLOAT base_diffuse_roughness 0 [0, 1]
FLOAT base_metalness 0 [0, 1]
Specular
FLOAT specular_weight 1 [0, +inf]
RGB specular_color 1, 1, 1
FLOAT specular_roughness 0.3 [0, 1]
FLOAT specular_ior 1.5
FLOAT specular_roughness_anisotropy 0 [0, 1]
Transmission
FLOAT transmission_weight 0 [0, 1]
RGB transmission_color 1, 1, 1
FLOAT transmission_depth 0 [0, +inf]
RGB transmission_scatter 0, 0, 0
FLOAT transmission_scatter_anisotropy 0 [-1, 1]
FLOAT transmission_dispersion_abbe_number 20 [0, +inf]
FLOAT transmission_dispersion_scale 0 [0, 1]
BOOL transmission_transmit_aovs * false
INT dielectric_priority * 0
kick -info <node>.parameter now includes all available metadata for the parameter, including linkability, bounds, help, as well as the type, name, and default value. (ARNOLD-15306)
$ kick -info openpbr_surface.fuzz_weight
node: openpbr_surface
param: fuzz_weight
type: FLOAT
default: 0
metadata:
help: The presence weight of a fuzz layer that can be used to approximate microfibers, for fabrics such as velvet and satin as well as dust grains.
max: 1
min: 0
Handle update for OSL shader changes: Information about shaders linked to an OSL shader is stored when using interactive render mode, so that any change in those shaders will cause the OSL shader to be automatically updated. (ARNOLD-14983)
Duplicate node name warnings: Warnings are emitted when adding a node with a name that already exists or renaming an existing node to an empty string. (ARNOLD-12596)
image courtesy of Nikie Monteleone
![]() |
![]() |
standard_volume improve how light is scattered in volumes. Using those parameters can help achieve high-scattering looks with fewer volume bounces and without adding energy. (ARNOLD-14482)scatter_diffusion accelerates scattering in volumes.scatter_diffusion_roughness accelerates scattering further on anisotropic volumes and can help achieve realistic looking clouds with fewer bounces. gain and bias curves can further tweak the effect of both parameters
Stochastic volume interpolation: The tricubic interpolation mode of AiVolumeSampleXXX() calls, such as those used by the standard_volume shader, now uses a stochastic method of interpolation that greatly reduces the amount of voxel data read in each sample call. In testing, we've seen speed ups of as much as 1.7x in certain scenes. Note that this technique does not always produce a perfectly matching image as a result, and can be toggled via the stochastic_volume_interpolation render option (default is on). (ARNOLD-3856)
Improved quality with Intel Open Image Denoise 2.3.0: The updated Intel Denoiser improves denoising quality and can produce sharper results with less artifacts on scenes with specular reflections and transmission. For more information about changes in this version, see the OIDN release notes (ARNOLD-15059).


Standins Viewport Display Color: A custom color can be set on aiStandIn nodes to control how they are displayed in the viewport, through the attribute "Object Display/Draw Override/Enable Overrides"
Sphere and Box volume added to aiVolume: A new attribute has been added to the aiVolume to set the volume shape to box or sphere, defaults to existing file mode (MTOA-247)
Support added for camera > backgroundColor: The maya camera attrbiute "backgroundColor" is now supported to change the background color of the current render(MTOA-1727)
OIDN support for Apple Metal and AMD GPU: The Intel OIDN denoiser now supports on macOS all Apple Metal GPUs (M1 and newer) and on Windows AMD GPUs using RDNA2 (Navi 21 only) and RDNA3 (Navi 3x) architectures. This allows for significantly faster denoising: on an M1 Max laptop we see 5-8x faster denoising. (ARNOLD-14855, ARNOLD-14856)
GPU Many-lights support for light linking: Arnold GPU now uses Global Light Sampling for group of lights linked to a shape. This can improve rendering performance for scenes with light linking. For example, the ALab scene now renders twice as fast when you use Global Light Sampling. (ARNOLD-15050)
GPU improved time to first pixel in scenes with many nodes: Reduced time to first pixel with Arnold GPU and scenes with many nodes. For example, we have seen a 18% improvement in time to first pixel on a simple scene with 1,000,000 instances. (ARNOLD-13652)
Motion vector AOV on GPU: The motionvector AOV is now supported when rendering on GPU. (ARNOLD-11046)
3D ramp modes: Ramp shaders have three new modes: 3d_linear, 3d_spherical, and 3d_cylindrical. These modes calculate the ramp input from the distance between the shading point and a user-defined reference point or line. (ARNOLD-15051, ARNOLD-15196)
![]() |
![]() |
![]() |
| 3d_linear | 3d_spherical | 3d_cylindrical |
![]() |
![]() |
![]() |
| worley noise (3d_spherical) | cell_noise (u) | alligator noise (v) |
![]() |
![]() |
![]() |
![]() |
![]() |
Improved AOV memory usage in non-progressive renders: Non-progressive, adaptive renders with the box filter now use less memory. Memory usage is also reduced in non-adaptive renders when using the following filters gaussian_filter, triangle_filter, sinc_filter, catrom_filter, mitnet_filter, blackman_harris_filter. For example, in a 1080p render using adaptive with AA_samples_max 50 and 6 AOVs, the memory used by AOVs was reduced from 14GB to 730MB (-96%). Note also that AOV memory usage now remains constant as AA_samples_max increases, unlike previous releases where the memory would increase quadratically. (ARNOLD-10160)
Improved efficiency of the Alembic procedural: The Alembic procedural now reads data in parallel, which optimizes initialization times of heavy Alembic files with many objects. We observed startup times going from 15 minutes to 4 seconds on a large production scene. (ARNOLD-14796)
Viewport representation for volumes in procedurals: Volumes can now be displayed in the viewport. This is currently only supported when loading a procedural (an aiStandIn) in MtoA. (ARNOLD-14382)
![]() |
Indirect diffuse and indirect specular parameters are now linkable in the standard_surface and openpbr_surface shaders: The indirect_diffuse and indirect_specular of standard_surface and openpbr_surface shaders now support linking of shaders. This should allow better optimization in scenarios where no indirect lighting is needed. (ARNOLD-14204)
Triplanar Nref user param: Triplanar shader now accepts an Nref user parameter, to allow specifying reference normals when in Pref mode. (ARNOLD-15063)
AOV names used for OpenImageIO subimages: Multi-part files previously used default subimage naming for layers/subimages. Arnold now uses the names of AOV outputs for the corresponding subimage. (ARNOLD-15053)
Triangle tessellation memory usage statistics grouped under polymesh memory usage: Arnold logs now list triangle tessellation and polymesh memory usage together in the peak CPU memory used and peak GPU memory consumed. (ARNOLD-15194)
Accessing Save Image Options with MEL: The Save Image options in the Arnold Render View are now accessable via the MEL arnoldRenderView command (ARNOLD-15083)
Deselect snapshots by clicking outside of selection: Clicking in empty space in the Arnold RenderView snapshots panel deselects the selected snapshots (ARNOLD-15080)
User preference for snapshots folder: There is a new user preference for the snapshots folder. You can override this by setting the environment variable ARNOLD_SNAPSHOTS_FOLDER (MTOA-1934)
![]() |
ParallelFor API: A new AiParallelFor function allows API users to easily implement multithreading of work over an array without having to use a dedicated multi-threading library such as TBB. (ARNOLD-14966)
// simple example that squares all the elements in an array
void square(size_t array_index, void* data, void* payload)
{
float& value = *static_cast<float*>(data);
value = value * value;
}
std::array<float, 7> values = {1, 2, 3, 5, 8, 13, 21};
AiParallelFor(values.data(), sizeof(values[0]), values.size(), nullptr, square);
ParallelJobs API: A new AtParallelJobs class allows programs to run jobs in parallel through a parallel job scheduler. This has more overhead than AiParallelFor(), but is more flexible as it allows for dispatching jobs without needing to place all the jobs in an array. (ARNOLD-15091)
AtParallelJobs jobs;
for (size_t i = 0; i < job_count; ++i)
{
payload = create_your_payload_data();
jobs.add(payload, &callback);
}
jobs.wait();
Shader API A new function to query whether an object is opaque or not. (ARNOLD-4919)
AI_API AI_PURE bool AiShaderGlobalsIsObjectOpaque(const AtShaderGlobals *sg);
Enum values You can now define enum values that start with a digit. (ARNOLD-15165)
const char *enum_myparam[] = { "2d_space", "3d_space", NULL };
AiParameterEnum("myparam", 0, enum_myparam);
Critical Section API Added compile guards around native_handle() references in ai_critsec.h because it no longer exists in the STL shipped with Visual Studio v17.8+ (MSVC v14.38+).
Node API Added new API to declare an update dependency between two nodes (a dependency is when a node uses data from another node during its update). (ARNOLD-15211)
AI_API void AiNodeAddDependency(AtNode* node, const AtNode* dependency);
triangle tessellation section. You can now find the memory usage under polymesh memory usage in peak CPU memory used or peak GPU memory consumed. (ARNOLD-15194)node_update function should declare this dependency with the new AiNodeAddDependency API. (ARNOLD-14973)force_update, nodes are expected to declare dependencies on other nodes using the new AiNodeAddDependency API, to optimize scene updates (ARNOLD-14901)force_update metadata is used