5.1.0.0 - Arnold for 3ds Max
30 Mar 2022
MAXtoA 5.1.0.41 is a feature release, included in 3ds Max 2023, and updating to Arnold 7.1.0.0
Installation
- Download Arnold for 3ds Max.
- Follow the installation instructions.
System Requirements
Autodesk 3ds Max 2020, 2021, 2022, 2023.
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 10.13 or later .
CPUs need to support the SSE4.1 instruction set. Apple Mac models with M series chips are supported under Rosetta 2 mode.
GPU rendering works on Windows and Linux only and requires an NVIDIA GPU of the Ampere, Turing, Volta, Pascal, or Maxwell architecture.
Optix™ denoiser requires an NVidia GPU with CUDA™ Compute Capability 5.0 and above.
- We recommend using the 511.09 (Quadro), 511.09 (GeForce), or higher on Windows and 495.46 or higher drivers on Linux. See Arnold GPU for more information.
- The driver type must be set to DCH.

In the driver page select "Windows Driver Type:" DCHImportant: The cache will also need to be re-populated after installing a new Arnold version, updating to a new NVIDIA driver , or changing the hardware configuration of GPUs on the system. More information can be found here.
Enhancements
MAXTOA-1277 - Triplanar shader now with different texture on each axis
MAXTOA-1281 - Ray depth values should be capped to 100
MAXTOA-1290 - Add flip_on_opposite_direction parameter to triplanar
Improved IPR interactivity: GPU and CPU interactivity is improved. ARNOLD-11981, ARNOLD-12007
Stability improvements: We've improved the stability of Arnold, especially when using multiple render sessions.
Per axis texture in triplanar: The triplanar shader can now optionally use different textures for each axis. ARNOLD-11708.
![]() |
![]() |
| input_per_axis: disabled (default) | input_per_axis: enabled |
A wide range of shading effects such as snow, dust, sand, and moss, can be achieved using input_per_axis (Y).


Compare images with and without triplanar.input_per_axis (Y)
A video that shows how to use the input_per_axis can be found here.
- Flip_on_opposite_direction in triplanar: A new parameter called flip_on_opposite_direction can also be used to control the behavior of the shader on faces facing the corresponding negative axis. ARNOLD-11615.
![]() |
![]() |
| flip_on_opposite_direction: enabled (default) | flip_on_opposite_direction: disabled |
Reduced overhead of running Imagers in IPR: Imagers will no longer always run after every progressive pass or bucket update. This can give a substantial speedup with expensive to run imagers. ARNOLD-11760
Better startup and update times with atmosphere shader: Starting renders and updating parameters when using a high number of AA samples with a volume_scatter atmosphere shader should now be much faster. ARNOLD-5503
Render times of a simple scene with volume_scatter.samples=30:
| AA | AA_max | Arnold 7.0 | Arnold 7.1 |
|---|---|---|---|
| 4 | 50 | 5:28min (node update: 5:20min) | 8 sec (node update: 0 sec) |
Case insensitive output image file extension:
AiFindDriverType()is now case insensitive, so asking, for instance, for a "png", "PNG", or "Png" will all return the png driver. This can be observed withkick -o, which will now work for different case types. ARNOLD-11934Updated MaterialX to 1.38.3: We now use MaterialX 1.38.3. Surface materials defined in a nodegraph are now supported. ARNOLD-10534, ARNOLD-11903
Updated Alembic to 1.8.3: The alembic procedural has been upgraded to use version 1.8.3 of the alembic library. ARNOLD-11973
Faster TextureResolveUseExistingTx: The
TextureResolveUseExistingTxstage is now much faster when many threads are trying to open textures for the first time with the use existing tx option enabled. On a 128-core machine we saw about an 8x speedup. ARNOLD-11689Faster per-pixel imagers: We removed overhead in imagers that can execute independently on each pixel. The exposure and tonemap imagers, for instance, are now significantly faster. This likely will be most noticed when interacting in IPR with small buckets. Note that bloom and denoising imagers require access to all pixels and so will not benefit from this optimization. ARNOLD-12011
Faster bloom imager: The bloom imager has been optimized. In one test we saw a 10% performance improvement. ARNOLD-11568
AOV denoising with the OptiX denoiser: Upgraded from the Optix 6 to Optix 7 denoiser. This newer denoiser supports denoising AOVs in a way that allows the denoised result to be compositable. Prevoius OptiX denoisers had no guarantees that the denoised results would combine correctly. ARNOLD-11726





Original beauty Denoised beauty Denoised diffuse AOV Denoised specular AOV Denoised diffuse + denoised specular Upgraded OIIO to 2.4.0: This allowed for fixing several defects listed in the Bug Fixes section. ARNOLD-11651, ARNOLD-11785, ARNOLD-11790
Invert view and display transform for textures using OCIO: When using an explicit OCIO or the default color manager, it is possible to specify both a display and view transform when writing outputs, using the View ( Display ) syntax. It is now possible to use the same syntax for input color transforms as well. For example, reading a texture using the ACES 1.0 SDR-video ( sRGB ) color space will correctly invert a file with a view transform baked in. (ARNOLD-11961).
GPU Enhancements
- Improved GPU IPR interactivity: GPU IPR rendering is now more interactive. ARNOLD-11981
USD Enhancements
- Point Instancer improvements: Per-instance primvars and nested instancers are now handled both in the procedural usd#944 as well as the render delegate usd#482
- Mesh builtin normals: The procedural now reads the builtin USD description for mesh normals usd#962 usd#986
- Curves UVs: The procedural now supports UVs on curves, for per-strand primvars usd#975
- Light-linking: ignoreRoots wasn't supported correctly for light / shadow links in the procedural usd#987
- Shadow parameters: The procedural now reads shadow parameters (enable, color) from USD lights usd#945
- Shader connections to array attributes: The render delegate now supports rendering shader connection to an array attribute. This is needed e.g. to have shader connections on ramps usd#970
- Render tags: Render tags are now taken into account by the render delegate usd#883
- Procedural search path: USD is now properly considering the options procedural_search_path when loading nested procedurals #usd942
- Motion blur with UsdSkel: the procedural now extends the interval used to bake the skinning, so that it includes surrounding integer frames usd#951
Bug Fixes
MAXTOA-1067 - Let the user select the AOVs' output type
MAXTOA-1264 - No more warning about the OCIO config file
MAXTOA-1288 - Crash when rendering a sequence with a scene that contains a RenderTimeInstanceObject
MAXTOA-1291 - The RenderTimeInstance_Translator causes warnings to be logged when it tries to access node parameters that don't exist
ARNOLD-11958 - Color correct shader produces incorrect values with small gamma values
ARNOLD-11852 - Memory leak with imagers when rendering multiple frames
ARNOLD-11835 - Full frame imagers crash with negative render region
ARNOLD-11827 - Intel Alder Lake CPUs did not always use high performance cores in Windows
ARNOLD-11795 - Spotlight penumbra should affect the specular intensity
ARNOLD-11667 - GPU and CPU+FIS filtering generated sampling patterns that could alias
ARNOLD-11242 - 'n' key was no longer toggling AI denoisers in kick IPR
ARNOLD-10400 - Volume motion blur advection computation was incorrect
ARNOLD-10391 - Volume motion blur used incorrect motion/shutter intervals
ARNOLD-11889 - Maketx --monochrome-detect should generate Y not R image for exr formats
ARNOLD-11968 - AiUniverseGetRenderSession returns invalid pointer instead of null if the session was previously destroyed
ARNOLD-11982 - Texture invalidation in Linux and macOS occurred even when texture was not modified
ARNOLD-12055 - [GPU] Crash switching from CPU to GPU with optix denoiser
ARNOLD-10577 - [GPU] Textured skydome_light leaks memory over multiple frames in a session



A video that shows how to use the input_per_axis can be found 
