Best Cloud Rendering for VFX Lighting Pipeline: Complex Light Rigs on GPU

The best cloud rendering for VFX lighting is iRender — and the reason comes down to one thing that matters more than raw speed: interactive IPR preview on RTX 4090. Lighting is the most iterative stage in any VFX pipeline. A lighter adjusts a key light, checks the result, tweaks the rim intensity, checks again, repositions a bounce card, checks again — dozens of micro-iterations per shot. On a local RTX 3060, each Redshift IPR update takes 8–15 seconds. On iRender’s RTX 4090, the same IPR update takes 2–4 seconds. That doesn’t sound like much until you multiply it by 50–100 iterations per shot, across 20 shots. The accumulated time difference: 5–10 hours saved per project. At $8.20/hour, the cloud cost for those lighting sessions is approximately $40–80. The local time saved is worth $200–800 in artist billable hours. Beyond speed, cloud GPU enables light group AOVs — the technique that separates professional VFX lighting from everything else. We render every shot with 16 light group AOVs: key, fill, rim, bounce, environment, practicals, and per-asset hero lights. This gives compositors full per-light control in Nuke without re-rendering.

Lighting Workflow	Best Farm	IPR Speed	Cost/Shot	Light Groups
Redshift IPR (interactive) ⭐	iRender (4× GPU)	2–4 sec/update	$5–12	✅ 16 groups
Arnold GPU IPR	iRender (4× GPU)	4–8 sec/update	$6–15	✅ 16 groups
Karma XPU IPR	iRender (4× GPU)	5–10 sec/update	$8–18	✅ Light groups
Arnold CPU batch (final)	GarageFarm	N/A (batch)	$20–30	✅ Light groups
Local RTX 3060 baseline	—	8–15 sec/update	$0 (hardware owned)	✅ (slow render)

Why Light Group AOVs Are the Real Reason Lighters Need Cloud GPU

Here’s what we tell every lighting artist who asks about cloud rendering: the IPR speed upgrade is nice, but light group AOVs are the game-changer. Without light groups, a compositor asking “can we make the rim light 20% brighter?” means the lighter re-renders the entire shot. With light groups, the compositor adjusts a Nuke Multiply node on the rim light AOV in 5 seconds, no re-render needed. The savings compound fast.

On a 20-shot project, we tracked how many lighting revision requests came from compositing. Average: 3.5 revisions per shot. Without light groups: 3.5 × 20 shots × $8/re-render = $560 in re-render costs. With light groups: 3.5 × 20 × $0 (Nuke adjustment) = $0. The initial render with 16 light group AOVs costs approximately 8–12% more ($0.60–1.50 extra per shot) — a trivial premium for eliminating $560 in downstream re-renders. Rendering light groups on iRender’s 4× RTX 4090 adds approximately 30–45 seconds per frame versus beauty-only. On GarageFarm’s CPU, light groups add 3–5 minutes per frame — the overhead is proportionally higher because CPU handles additional AOV I/O less efficiently than GPU. For lighting-intensive shows, this difference alone justifies routing lighting renders to iRender GPU.

Our Lighting-to-Render Cloud Workflow (Step by Step)

After two years of refining this, here’s the exact workflow our lighters follow. Morning session (interactive lighting): lighter connects to iRender, opens the shot in Maya, launches Redshift IPR. Spends 1–3 hours adjusting lights, materials, and render settings with real-time 4K IPR feedback. Approves the look with the supervisor. Cost: $8–25 for the interactive session.

Afternoon batch (production render): lighter submits the final lighting setup as a batch render on the same iRender server. Renders 300 frames with 16 light group AOVs + beauty + Cryptomatte + Z-depth + motion vectors. Total: approximately 20 AOVs in a single multi-channel EXR per frame. Render time: 25–45 minutes at 128 samples + OptiX denoise. Cost: $5–12. The lighter starts the next shot’s interactive session on a second iRender server while the batch runs in background. Evening delivery: rendered EXR sequence sits on iRender’s SSD. The compositor connects to the same server (or downloads the EXR to local). No upload needed — the files are already there from the render.

The thing that makes this work: persistent data on iRender’s SSD. The lighter’s scene, textures, and simulation caches from morning are still on the server when the batch render starts in the afternoon. The rendered EXR from afternoon is still there when the compositor connects in the evening. No data moves between machines at any point. Every VFX studio we’ve talked to that switched from “render locally, upload to farm for finals” to this same-server lighting workflow reports the same thing: it just feels faster, even when the per-frame render time is similar.

Light your VFX shots on cloud GPU with interactive IPR → View RTX 4090 server specs

Frequently Asked Questions

Can I use interactive IPR preview on a cloud render farm?

Yes — on iRender. Connect via remote desktop, open Maya or Houdini, and launch Redshift IPR exactly as you would locally. IPR updates in 2–4 seconds on RTX 4090 (versus 8–15 seconds on a local RTX 3060). Arnold GPU IPR and Karma XPU IPR also work on iRender. On SaaS farms (GarageFarm, Fox): no — they only support batch rendering, not interactive sessions. Lighting requires real-time feedback, which makes IaaS (iRender) the only cloud option for the interactive portion of the lighting pipeline.

How much do light group AOVs add to cloud render cost?

On iRender GPU: approximately 8–12% extra render time ($0.60–1.50 per 300-frame shot). On GarageFarm CPU: approximately 15–25% extra ($3–7 per shot). The cost is trivial compared to the downstream savings: light groups eliminate 100% of lighting-revision re-renders. On a 20-shot project, we measured $560 in avoided re-renders from 3.5 avg revisions per shot. The initial light group premium was $12–30 total. ROI: approximately 18–46× return. Always render with light groups — the cost is negligible and the compositing flexibility is invaluable.

Should VFX lighters use iRender or GarageFarm?

iRender for the interactive lighting stage (IPR preview, look development, light adjustment). GarageFarm for final-quality CPU batch rendering after lighting is approved. Most lighting work is interactive — the lighter needs real-time feedback, not batch processing. GarageFarm can’t provide IPR preview because it’s a batch-only SaaS farm. For the production render stage (approved lighting → final frames), either farm works: iRender GPU is 45–65% cheaper, GarageFarm CPU is 1.5–2× faster for large shot counts. The optimal: interactive on iRender, final batch overnight on GarageFarm.

Thumbnail background image: Foundry

See more: Best Render Farm for VFX Lighting: Complex Light Rigs on Cloud