Best Cloud Rendering for VFX Render Passes: AOV Setup for Nuke Compositing

Not all render farms handle AOVs the same way — and a dropped render pass can cost you an entire re-render. We set up an 18-pass Redshift AOV configuration (diffuse, specular, SSS, reflection, refraction, emission, depth, P, N, motion vectors, cryptomatte, plus custom utility passes) and rendered the same 100-frame shot across 4 farms. On iRender, all 18 AOVs arrived intact inside multi-layer EXR files — exactly as configured. GarageFarm delivered 15 of 18 passes; it dropped our 3 custom utility AOVs because its pipeline only recognizes standard pass names. Fox Renderfarm renamed several passes during its automated processing, which broke our Nuke template. RebusFarm delivered all passes but in separate single-layer EXRs instead of multi-layer — doubling our Nuke ingestion time.

Render Farm	AOVs Delivered	Multi-Layer EXR	Custom Pass Names	Nuke Template Compatible
iRender ⭐	18/18 ✅	✅ Yes	✅ Preserved	✅ No changes needed
GarageFarm	15/18 ⚠️	✅ Yes	❌ Dropped custom	⚠️ Minor reconnect
Fox Renderfarm	18/18 ✅	✅ Yes	⚠️ Renamed	❌ Template broken
RebusFarm	18/18 ✅	❌ Separate files	✅ Preserved	❌ Manual reassembly

Why Do Some Render Farms Drop or Rename AOV Passes?

SaaS render farms run your scene through an automated pipeline that parses your scene file, distributes frames across nodes, and collects outputs. During this parsing step, the farm’s software needs to understand what each output pass is. Standard AOV names like diffuse_color, specular, or depth are recognized because they’re built into the renderer’s default naming convention. But custom passes — things like utility_mask_hero or custom_holdout_BG — don’t match any expected pattern, so the farm’s pipeline either skips them (GarageFarm) or renames them to a generic label (Fox Renderfarm).

On iRender, this isn’t an issue because you’re rendering on your own machine. There’s no automated parsing layer between your scene file and the renderer. Whatever AOV names you set in Houdini or Maya, that’s exactly what appears in the output EXR. The downside is you’re responsible for everything — if you misconfigure an AOV, there’s no automated check to catch it. We’ve shipped renders with a typo in a pass name and only caught it when the Nuke template couldn’t find the layer.

What’s the Ideal AOV Setup for Nuke Before Sending to a Render Farm?

After testing across farms, we’ve landed on a setup that works reliably everywhere: keep your core AOVs to standard names (the Redshift/Arnold/Karma defaults), and add custom passes only if you’re rendering on iRender or another IaaS farm. Our production AOV list runs 12–15 passes for standard shots and up to 22 passes for hero shots with complex compositing requirements.

A practical tip: render one test frame before committing to a batch. Open it in Nuke, run your Read node, and verify every layer name matches your template. This 5-minute check has saved us from at least 3 full re-renders. For Redshift users specifically, the AOV Manager in Houdini lets you export your AOV preset as a .json file — upload that alongside your scene to ensure consistency between local and cloud renders. The Maxon docs cover this under “AOV Presets” in the Redshift for Houdini section.

Need all your AOV passes delivered intact? → Render on iRender with full AOV control

Frequently Asked Questions

How many AOV passes can I render on a cloud GPU farm?

On iRender’s RTX 4090 with 24 GB VRAM, we’ve rendered up to 22 AOV passes in a single Redshift job without hitting memory limits. Each additional AOV adds roughly 100–300 MB of VRAM overhead depending on resolution. At 2K resolution, 18 passes used about 14 GB of VRAM total. At 4K, the same setup pushed to 19 GB — still within the RTX 4090’s 24 GB limit. If you’re exceeding 20 passes at 4K, monitor VRAM usage closely or split into two render jobs.

Will GarageFarm render my custom AOV passes?

Standard AOVs (diffuse, specular, reflection, depth, normals, motion vectors, cryptomatte) render correctly on GarageFarm. Custom-named passes — anything with non-standard labels like utility_mask_hero — get dropped during their automated pipeline processing. In our test, 3 of 18 passes were missing. If your comp relies on custom utility passes, either rename them to match standard conventions before submitting, or use an IaaS farm like iRender where pass names are preserved exactly as configured.

Should I render AOVs as multi-layer EXR or separate files?

Multi-layer EXR is almost always the better choice for Nuke workflows. A single file per frame with all passes inside is cleaner to manage, faster to load via Nuke’s Read node, and uses less disk space than separate files (roughly 30–40% smaller due to shared header data). The only exception is if your farm doesn’t support multi-layer output — RebusFarm, for example, delivers separate single-layer EXRs, which means you’ll need to reassemble layers in Nuke with a Merge or Shuffle node before your template works.

See more: Best Render Farm for Multi-Pass Rendering: AOV Setup & Cloud Workflow