Arnold now renders straight in the 3ds Max viewport

Autodesk has released Arnold 7.5.4, and the release's most noticeable change concerns where the artist sees the picture. The renderer outputs the image straight into the standard 3ds Max viewport — switching the viewport into ActiveShade mode is no longer necessary for that. Cinema 4D users get a similar capability: Arnold renders directly into the C4D viewport.

Underlying the change is the accelerated viewport, that is the direct_outputs mode introduced in the July release 7.5.3. In 7.5.4 it has become more responsive during navigation: by Autodesk's description, the delay between camera movement and the appearance of a workable image has dropped, and the picture stays cleaner while the camera moves — so lighting and lookdev can be done without leaving the viewport.

Accelerated viewport: what changed in 7.5.4

During camera movement the viewport returns to an interactive frame rate faster and holds it, and when the camera stops it restores full quality. To keep up, it may drop several quality settings at once and compute at reduced resolution, and if even that is not enough, switch to an unlit preview where every surface is shaded by facing ratio. Two new options govern the behaviour: direct_outputs_max_linear_factor (1 by default, 6 maximum) permits lowering the resolution during navigation, and direct_outputs_ndoteye (off, auto or always; off by default) enables the fast low-quality preview. Both options work only when rendering on the GPU.

The main compatibility extension: the accelerated viewport now works when rendering on the CPU too, not only on the GPU. A graphics card is still required for the viewport processing stage — without one Arnold writes a warning and falls back to ordinary rendering. More AOV types are supported, and the Z, ID, object and shader AOVs returned by AiGetRenderOutput() (the function now works on the CPU as well) get values in every pixel from the very first pass after camera movement, so plug-ins can draw wireframes and overlays from the depth buffer right during navigation. The mode also supports kick, though colour management is not implemented there yet.

One limitation is worth knowing in advance. The direct outputs mode takes AOVs directly, bypassing the usual CPU infrastructure of imagers and drivers — meaning imagers, including the denoisers and the new Shader Imager, do not apply in this path.

What else version 7.5.4 brought

• Shader Imager — an imager that references a shader network and evaluates it at the imager pass stage. OSL shaders or the image node read AOVs by names of the form "aov:diffuse" and "aov:albedo", so a simple composite can be assembled interactively, right during the render. Intermediate imager results when rendering on the CPU now appear earlier and more often.
• Pausing and resuming a render within a session — without losing what has been computed; in the interactive kick window the pause is toggled with the P key.
• A render time limit. The max_render_seconds parameter stops a batch render roughly after the specified number of seconds has elapsed and writes the frame in the state it managed to compute; the progressive sampler is used for this so that the frame is always complete. The default is 0, meaning the limit is off, and interactive sessions ignore the setting for now.
• Rim Light has been split out into a separate node: it produces the same rim lighting as the Toon shader, but it can be mixed into any shader network as a separate RGB component. The light_samples_indirect option sets the number of light samples on indirect reflections (1 by default, as before).
• A rename. The specular_weight parameter of the specular BSDF node is now called reflectivity; the old name is accepted as a synonym, so existing scenes work.
• Performance. By Autodesk's measurements, deforming a subdivided character with displacement (17 million polygons after subdivision) got 43% faster on both CPU and GPU, and deforming all the meshes of the Caldera USD scene of 27,000 objects by up to 78% on the CPU and up to 75% on the GPU. Preparing a GPU scene with 100,000 instances dropped from 660 to 214 ms, with a million instances — from 9.8 to 4.1 s. In a test scene with 39,000 moving objects with differing numbers of motion keys, render time fell from 3.76 to 0.28 s.

Where to get it and what it runs on

Arnold 7.5.4.0 came out on 22 October 2026. Per the release notes, the renderer runs on Windows 10 and newer, on Linux with glibc 2.28 and above and on macOS 12+; GPU rendering and OptiX denoising are available only on Windows and Linux and require an NVIDIA card of the Maxwell architecture or newer.

The cloud service Flow Render remains in technical preview with a quota of 2400 render minutes per month per user, and in this release has learned to work with custom OCIO configurations, UDIM tile tokens in texture names and a per-frame time limit.

Why rendering is moving into the viewport

A separate interactive preview window was a compromise of an era when ray tracing could not keep up with navigating the scene. Today that compromise looks superfluous: Blender computes Cycles right in the viewport, Redshift launches an interactive render in the Cinema 4D viewport, Houdini lets you pick Karma as the viewport's render delegate. Against that backdrop Arnold looked like a laggard — above all in 3ds Max and Cinema 4D, whereas in Maya the plug-in got the accelerated viewport back in the July release 7.5.3. So 7.5.4 rather levels the integrations with one another.

The technical basis for that levelling was laid gradually. In 7.5.3 Autodesk added the direct_outputs mode, which takes AOVs directly, bypassing the CPU infrastructure of imagers and drivers, and opened an API allowing plug-ins to copy the render result into an OpenGL texture via CUDA-GL interop — in the release notes this is called outright a way to build fast interactive viewports. A release later the mechanism has reached 3ds Max and Cinema 4D, and Houdini and Katana users have grounds to expect the same.

There is a flip side too. The fast viewport so far lives apart from imagers — that is, from the part of the pipeline where Autodesk put the Shader Imager and rough compositing in this very release. The result is two interactive paths with different capabilities: a responsive viewport without post-processing and a slower preview with it. If the trend continues, bringing them together is the next logical step, otherwise modellers and lighting artists will have to choose between navigation speed and the look of the final frame.

What matters more in the Arnold update?

• Rendering straight in the viewport
• Dropping ActiveShade