Northfork Render finished a feature film on 400 on-demand L40S GPUs — with zero hardware owned

Northfork Render is a 34-person VFX and animation studio known for stylized, painterly CG. When the studio landed its first full feature — 96 minutes of animation with heavy volumetrics and path-traced lighting — the founders faced the classic mid-size studio dilemma: the render requirements of a feature, without a feature-size balance sheet.

The challenge

Northfork's own farm topped out at 48 GPUs, enough for episodic work but nowhere near a feature deadline. The delivery schedule called for roughly 612,000 final frames plus another 1.4 million preview and technical frames, with render times averaging 40 minutes per final frame at 4K. Run in-house, the final-frame workload alone represented over a year of continuous rendering. The studio had nine weeks.

Leasing a traditional render farm meant long minimum terms and fixed capacity — paying for peak scale during the many nights when the pipeline was blocked on artist reviews, not renders. Northfork needed capacity that tracked the actual shape of production: quiet weekdays, brutal weekends, and a final month that never slept.

Why Reviosa

The studio's pipeline team benchmarked Reviosa's L40S instances against their in-house cards and found the sweet spot immediately. At $0.69/hr, the L40S delivered 48 GB of VRAM — enough for Northfork's heaviest volumetric scenes without out-of-core penalties — and hourly billing meant the farm could breathe with the schedule.

Integration took under two weeks. Northfork's render manager already supported cloud bursting, so the team wrote a small autoscaler against the Reviosa API: when the priority queue depth crossed a threshold, it launched on-demand L40S instances in blocks of 20; when queues drained, instances terminated within the hour. Scene data synced through Reviosa block volumes, with finished frames landing in object storage for the compositing team.

"The autoscaler graph from delivery week is my favorite piece of art we made on this film. It goes from 60 GPUs to 400 and back to zero in nine days — and every hour on that graph was an hour we actually needed." — Theo Marchetti, Co-founder

The results

Northfork delivered the feature on schedule, and the infrastructure numbers told a story no traditional farm could match:

• 400 L40S GPUs at peak during final delivery, scaled from a baseline of 40-60 during regular production weeks.
• 612,000 final frames rendered on Reviosa, plus the full preview workload — over 2 million frames total.
• $0 spent on idle hardware. Instances terminated automatically when queues drained; weekend peaks cost weekend money, quiet Tuesdays cost almost nothing.

Total cloud render spend came in 31% under the quote Northfork had received for a fixed-capacity farm lease — and 100% of it mapped to frames actually rendered. Just as important for a studio this size, there was no capital outlay: no depreciation schedule, no colo contract, no hardware to resell when the show wrapped.

The feature has since screened at two major festivals, and Northfork has signed its second one. The bid for that project assumed Reviosa burst capacity from day one — which let the studio quote a delivery window 30% tighter than competitors with fixed farms. For Northfork, elastic rendering is no longer a workaround. It is the business model.