The problem
Five idle machines and one that renders all night.
The render sits on one workstation for six hours while every other machine in the building does nothing. Everybody knows this, and everybody lives with it, because the answer has always been a render farm: a server, a database, a service running on each machine, and somebody whose job is to keep all of it alive.
On a show, nobody has that somebody. You are rigging, you are cutting, you are delivering. So the farm never gets built and the workstation keeps rendering all night.
The answer
The machines run the farm themselves.
There is nothing to install but the app. Submit a comp from the panel inside After Effects and it is cut into parts. One of the machines running Presto quietly takes the role of pool server: it knows the whole queue, hands each free machine its next part and pushes every change to the others as it happens. One of them assembles the finished segments into the final file.
Nobody picks that machine and nobody looks after it. It still renders like the others, and if it leaves, the next one in line takes over within seconds. Open the app on a machine, point it at the folder, press start: that machine is now part of the farm. Close it and it is not. No address to type, no node to register, no administrator to ask.
The principle
Nothing in this picture is a Presto component.
The machines are your machines, the NAS is your NAS, and the folder is a folder. The only thing Presto adds is an app on each machine, and one of those apps also serves the pool. The folder stays the memory: each part is a file whose name carries its state, and a machine taking a part renames it. A rename either happens or it does not, which is the one operation every network share on earth already performs reliably.
So the machine serving the pool can vanish in the middle of a render and nothing is lost: no part disappears, no part is rendered twice. The next machine in line reads the folder once and carries on.
Add a machine to the pool and it joins the render. Take one away and the rest carry on.
Who renders what
Fast machines first, and nobody waits for the slowest one.
Machines do not race for parts. The machine serving the pool hands them out one by one, in queue order, to whoever asks. Three things decide who gets what: the priority of each render, the score of each machine, and what happens when only the last parts are left.
Urgent renders go first
Move a render ahead of the others at any time, from the app or from your phone. It takes the next free machines, and nothing already rendering is lost.
Every machine has a score
The score is a delivery time: how long the machine takes to render a reference part and land it on the share. It starts from a quick benchmark, then learns from real jobs, comparing each machine with the others on parts of the same comp. A laptop on Wi-Fi scores far below the same laptop on Ethernet, because that is what it costs in real life.
The last parts go to the fast machines
While there is work for everybody, everybody renders. When fewer parts are left than machines, a slow machine only takes one if no faster machine will be free soon enough to finish it first. It never waits more than three minutes on a job: a slow render always beats a stalled queue.
Backup render at the very end
One part still crawling on a slow machine, a fast one sitting idle: the fast machine renders the same part in parallel, and the first to finish wins. Nobody is stopped on a guess. It only kicks in when it clearly pays, at least a minute and 30 % of the remaining time saved, and it can be switched off in Settings.
The end of a job is where farms lose their time: nine machines done, one still grinding. Presto spends its effort there, so the final file lands when the fast machines say it does, not when the slowest one gets round to it.
Key features
What it actually changes for you.
Keep working
The render leaves your machine and runs on the others, from a frozen copy of the project taken at submission. Each machine chooses its roles, so a workstation in use can still lend a hand.
More machines, more power
Need it faster tonight? Open Presto on another machine and point it at the same folder. It joins the job already in progress.
Mac and PC together
The Mac suite and the Windows machines in the truck work the same job at the same time. A part rendered on Windows is combined on a Mac without anyone thinking about it.
Installed in two minutes
One app per machine, one folder to point at, one button. The machines find each other through the folder: no address to type, nothing to stand up, nothing to maintain.
Built to survive the night
A machine that crashes, a laptop that walks out, a share that drops: the job carries on and the lost part goes back in the queue on its own. Even the machine serving the pool can go, another takes over within seconds. Whatever fails says which machine and why, and your phone gets a push on delivery and on failure.
Nothing leaves the building
The project, the media and the finished file never go near a third party. No upload, no account, no cloud storage holding an unreleased show. Even the live monitor stays on your network.
Movie files out
The final file is the file you asked for, assembled automatically. No image sequence to babysit and no manual pass at the end.
Live monitor, phone included
Served live by the pool itself: every part fills its bar as it renders, on any screen in the building. Scan a QR code in the app and it is on your phone, where you can kill, restart or reprioritise a part. Red means an error and nothing else.
One detail that decides more evenings than raw power: a part is not finished when After Effects stops, it is finished when the file has landed on the share. Thirty gigabytes travel in thirty-three seconds over ten gigabit and in seventeen minutes over Wi-Fi. A laptop on Wi-Fi and the same laptop on Ethernet are not the same machine, and Presto does not score them the same.
Where the others stop
Everything else shares out jobs. Presto shares out one comp.
Event work has its own weight class: LED wall resolutions, HAP, NotchLC or ProRes, deliverables measured in tens of gigabytes. Anything that answers a render problem by moving those files somewhere else has already lost.
What each one actually does
- Render Brain. Easy, cheap, and the right answer when ten comps are waiting: it hands a whole render item to each machine. One comp still stays on one machine, so a three hour render is still a three hour render. The pool is metered too, two to twelve machines depending on the plan.
- Studio render managers. These do split a comp and run on your own network. They also want a server, a repository, a service on every machine and somebody to keep it alive, which is a pipeline department. And what comes out is a sequence of still images.
- After Effects on its own. Adobe is explicit: several machines cannot produce one movie file together, only a sequence of still images. Hundreds of files crossing the share, then a second pass to assemble them.
- Cloud render farms. Upload eighty gigabytes of ProRes from a venue, over whatever connection the site provides, and download it before doors. The maths does not work, and it means handing an unreleased show to a company you have never met.
- RenderGarden. The reference for years, and gone: its maker wound the company down. Free and unsupported is not something to build a delivery on.
| Presto | Render Brain | Render managers | After Effects | |
|---|---|---|---|---|
| One comp split across machines | Whole jobs only | |||
| Comes back as a movie file | Image sequence | Image sequence | ||
| Machines included | Unlimited | 2 to 12 by plan | Unlimited | Unlimited |
| No server to install or maintain | Server required |
Render Brain is the tool to reach for when the queue is long. Presto is the one for when the clock is short.
Availability
Coming soon.
Presto is in daily use on a live event pool of Mac and Windows machines, and it is being finished with that pool as the test bench. It is not downloadable yet.
One licence, for the machine that sends the render. Every machine that renders is free, so the pool can grow without the bill growing with it.
If your machines spend their nights idle while one of them renders, say so and you will hear about it the day it ships.