One machine running Templater Bot is enough until the arithmetic says it is not. A farm beats it in exactly two situations: the batch cannot finish before its deadline on a single machine, even after you have tuned that machine, or a single machine is too risky to leave as your only way of making videos. Much of what gets called a “farm problem” can turn out to be a template problem, a data problem or an asset problem, and a second machine will not fix any of those.
This page is about that decision. It is not about how to size a farm for a surprise burst, and it is not about whether renting minutes costs more than owning hardware. We have separate posts on sizing a render farm for unpredictable volume and on in-house hardware versus cloud rendering by the minute. Here we start earlier. You have a batch machine, or you are about to set one up, and you want to know whether to add a second, a third or a rack of them.
Start with the one machine, and ask whether it is really the limit
Picture a producer at an agency who makes personalized offer videos from a spreadsheet. The Bot runs on one workstation. Last week a batch took far longer than anyone expected, and a colleague said the answer was obvious: buy more machines. Before agreeing, look at what was actually slow.
Our FAQ lists what drives render time. Those are the average length of the video, its resolution, the hardware, the ratio of dynamic frames to static ones, how heavy the effects are, and how the template is built. Only one of those six is about how many machines you own. The other five are things you can change, and changing them helps every machine you will ever add. A heavy template copied onto four computers gives you four computers that are each slow.
The Bot has its own quirk that looks like a capacity problem and is not one. If you leave it enabled for a long stretch, Templater’s versioning can slow down as it keeps receiving more and more data. The cause is the way After Effects caches data when layer properties change. That cache is helpful when a person works by hand. It becomes a drag when a machine runs unattended for hours. Our documentation says to turn off the layer cache in the hidden preferences and to disable disk caching on any machine where the Bot stays enabled. A studio that sees every batch get slower as the night goes on may not need a farm at all. It may need to change two preferences. We walk through the steps in optimizing After Effects for the Templater Bot.
There is also a limit inside the Bot that has nothing to do with hardware. By default it queues at most 20 rows at a time, and that is also the maximum. That sounds like a ceiling but it is a pacing control. Our documentation gives the example of setting ten thousand rows to ready and letting the Bot work through them twenty at a time, which reduces the memory Templater needs. The queue is how one machine handles a huge list without choking on it. Templater is built to break a massive workload into manageable batches and queue the jobs one after another, so a very large batch does not by itself call for a farm. It calls for patience, and patience has a price in hours.
So the first test is simple. Make the template lean, set the caching preferences for unattended running, and measure again. Only then does the number you get describe your real single-machine speed. If you are on Projector rather than Bot for part of the work, there is one more lever on the same box. Projector can render several videos at once, using the processor cores you allocate to it (half of them by default), where After Effects alone works one video at a time. We cannot tell you what that gains on your hardware, because it depends on the template and the machine. Test it before you buy anything.
Before You Buy
Tune one machine before adding more
-
Make the template leanA heavy template copied onto four computers gives you four computers that are each slow.
-
Set caching preferences for unattended runningTurn off the layer cache in the hidden preferences and disable disk caching on any machine where the Bot stays enabled.
-
Measure againOnly then does the number you get describe your real single-machine speed.
-
Pre-flight on a typical row and a heavy rowThe video must be like the others in the set, so do not pick the easiest row.
-
Read the answer against the real deadlineThe deadline decides, the video count did not change, the deadline did.
The deadline test, worked through with a pre-flight
Once the single machine is tuned, the question becomes a clock question. How long does the biggest batch you will actually run take on one machine, and is that inside the time you are allowed?
Our FAQ gives the method, and it is the whole method. Render one video, or a small batch, on a machine comparable to the one you will use in production. The video must be like the others in the set, so do not pick the easiest row. Multiply the time for one by the number of videos. If the videos render in a queue, one after another, that is a fair estimate of the total. Then the FAQ says that running videos in parallel across machines cuts the time in half with two machines, to a third with three, and so on.
Here is an illustration, with made-up numbers chosen only to show the shape of the calculation. Suppose the pre-flight says one offer video takes three minutes, and the weekly list has six hundred rows. Six hundred times three is eighteen hundred minutes, which is thirty hours on one machine. If the offers must reach customers by the next morning and the list arrives in the evening, one machine loses by a wide margin. Two machines make it fifteen hours. Three make it ten. Now you can read the answer off the clock rather than guess at it. If the window is about ten hours, you need three machines at least, and in practice you need a fourth to absorb the ordinary bad luck we describe below.
Now change the story. The same list arrives on Friday and is due the following Wednesday. Thirty hours on one machine fits comfortably in that window, even with a rerun of a few rows. Buying more machines would be spending money to shorten a wait nobody is suffering through. The video count did not change. The deadline did, and the deadline decides.
Two cautions belong with this. First, measure on the machine type you will actually use. A pre-flight on a fast desktop flatters a farm made of older computers, and the slowest machine in a mixed group sets the pace for any row you cannot send elsewhere. Second, do the pre-flight on a heavy row as well as a typical one. The batch will contain both, and the heavy row is the one that blows the estimate. We say nothing here about how much margin to add on top, because there is no honest universal figure. A deadline where lateness is an annoyance can run closer to the line than one where lateness costs you the event. Decide how much failure you can absorb and size to that.
If you want to test the whole idea before you commit to any hardware, the free trial of Templater contains all features for all versions, with a watermark on dynamic layers. That includes the command line. You can run a real pre-flight without buying a key first. The watermark means the output is a test, not something to send.
What the second machine actually costs you
Suppose the deadline test says one machine is not enough. A farm is not just more of the same. Each added machine brings work that the first one never needed, and the honest comparison counts that work.
The data source needs rules. If several Bots read the same spreadsheet, they have to know which rows are theirs. Our documentation says to add a column named render-status, give every Bot a unique name in the Bot Name field in Preferences, and add a column named bot where you put the name of the Bot that should process each row. You mark a row ready by typing ready into render-status, and Templater moves it through queued, processing and done. It marks failures with codes such as ERR::MEDIA for a problem with a media asset and ERR::AFXHOST when After Effects itself hit an error. The documentation adds a warning about the ERR::AFXHOST case. After Effects will likely have stopped operating normally on that machine, so a person has to go and look. Setting this up is a small job once, and a larger job every time somebody edits the sheet carelessly. Our post on setting up one Bot or many on one data source covers each column.
There is a trap in the same documentation that matters more with several machines than with one. The Bot overwrites existing files if it renders to the same output file name. With one machine you might get away with sloppy names. With four machines writing to the same output folder, two rows with the same name will quietly replace each other and nobody will see an error. Give every row a unique id or output name before the farm exists.
Render Status Codes
What each Bot status value means
| Status value | Set by | Action needed |
|---|---|---|
| ready | You, in the bot column | Templater picks up the row |
| queued | Templater | None, row is waiting in the queue |
| processing | Templater | None, row is actively rendering |
| done | Templater | None, row completed |
| ERR::MEDIA | Templater | Problem with a media asset |
| ERR::AFXHOST | Templater | A person has to go and look, After Effects likely stopped operating normally |
A URL feed adds a further wrinkle. When the source is a URL feed, the Bot cannot update the render-status value, because the data source belongs to the user’s own system. You register a custom script on the Post Job, Post Batch or On Bot Disable events so that “ready” rows are not rendered twice. On one machine that is an annoyance. On several, a missing script means several machines pick up the same row.
Every machine is another license activation. The standard Templater license is node-locked. It is activated on one machine at a time, and you move it by deactivating it in one place and reactivating it in another. For a farm of fixed machines, that means one activation per machine. A floating license works differently. Its key activates a license server you install, called DataclayLM, and the Templater machines lease a seat from the pool and hand it back when After Effects quits. Jon Christoffersen, who leads our support team, gave us the working rule: if at most six are active at the same time, the floating key needs six activations. More client machines can have the floating build installed and check out a seat as seats return to the pool. The key type is chosen when you buy, and the two builds are not interchangeable, so settle that before the purchase, not after. For machines that come and go, the floating option is the more comfortable one, and our documentation points out that it suits virtual machines in particular, because a node-locked license has to be deactivated and reactivated by hand every time a VM is rebooted or a new one is created. We explain the choice in what a floating license is.
Every machine has to be a faithful copy. The same After Effects, the same Templater, the same fonts and plugins, and the same footage reachable at the paths the project expects. After Effects stores paths. A machine that has the project but not the fonts will produce a different video, or none. This is general practice. Build the first machine, then copy its environment on purpose rather than rebuilding it from memory.
Renting machines is possible, with a condition. Our FAQ says you can run Templater on one or more Windows Server machines in a cloud environment from one of the large cloud providers, and that no special Dataclay license is required. It also says Templater needs After Effects installed and running there, and that Adobe’s standard end user license agreement does not permit After Effects on a server. You have to speak with an Adobe salesperson to get the right license. Settle that conversation before you plan around the option.
Multi-Bot Setup
What to settle before running several Bots
- Add a render-status column and a bot column to the data source
- Give every Bot a unique name in the Bot Name field in Preferences
- Give every row a unique id or output name before the farm exists
- Register event scripts on Post Job, Post Batch or On Bot Disable if using a URL feed
- Settle the license plan, node-locked or floating, before purchase
- Build the first machine, then copy its environment on purpose: same After Effects, same Templater, same fonts and plugins, same footage paths
If you use QUE, you become the scheduler. Our FAQ states that QUE does not offer load balancing of jobs between satellites at this time. You can assign jobs to specific satellites yourself, and you have to check that the fonts, plugins and other assets each job needs exist on whichever satellite you choose. Your plan also comes with a certain number of satellite slots, and more can be purchased. For a small fleet that you know well, manual assignment is fine. For a large and changing fleet, it is a standing chore. Read what a QUE satellite is before you design around it.
The risk test, and when one machine should stay one machine
The second case for a farm has nothing to do with speed. It is about what happens on the night something breaks.
A single batch machine is a single point of failure. Suppose it hits an After Effects error halfway through a run, or an operating system update restarts it, or its disk fills. Every video that was meant to ship from that machine waits for a person to notice and fix it. That might be perfectly fine for an internal weekly report. It is not fine for something customers expect on a schedule, such as a set of results the morning after an event.
With two machines running the Bot against the same data source, the work can continue on the other one while someone repairs the first. We want to be careful here, because we should not promise more than the documentation does. Rows you assign to a named Bot belong to that Bot, so if you divide rows by name, a machine that stops leaves its rows waiting until somebody reassigns them. Whether a second machine rescues the batch depends on how you divided the work. That is general practice, not a feature we can describe in detail. Think it through before the night it matters, and write down who reassigns rows and how.
There is a related point about how fast anyone finds out. The Bot is a background process with no front end of its own. Event scripts can send an email or a text message when something happens, but nothing does that by default. A farm without an alert is a farm that fails silently in more places. Our post on configuring the Templater Bot covers the settings that give you a chance to hear about trouble early.
Now the cases where we would tell you to stay with one machine. Stay if your largest batch fits inside its deadline with room to spare. Stay if the volume is occasional and a long overnight run is acceptable. Stay if nobody on your team wants to keep a second computer patched, licensed and identical to the first. A fleet is a small piece of infrastructure, and infrastructure needs an owner. If you do not have one, a second machine becomes a liability that fails at a worse moment than the first would have.
Go to a farm when the pre-flight arithmetic fails on a tuned machine, or when a stopped machine would cost you something you cannot recover. And if your work is occasional with sharp peaks, a mix can be better than buying capacity for a peak you reach twice a year. QUE Meter is built for sporadic campaigns, and our FAQ notes that any active Rig, Pro or Bot key can be connected to it and draw from a shared pool of credits. A cloud rendering service that bills by the minute is also a fair choice for teams who would rather not run machines. We describe where each fits in the posts linked at the top, and for a team that wants no infrastructure at all, a hosted platform is a perfectly honest answer.
A last note on terms. Everything above assumes you drive videos from a data source with Templater and its Bot edition. If your videos do not come from an After Effects project at all, none of this applies, and a different kind of tool will serve you better.
The bottom line is short. Tune one machine first, because it improves every machine you add later. Do the pre-flight on a typical row and a heavy row, and read the answer against the real deadline. If one machine fits, keep one machine and spend the effort on a canary row and an alert. If it does not fit, or if a single failure is too expensive, add machines with the data source rules, the license plan and the identical environment settled first. The farm is the right answer when the clock or the risk says so, and not before.
We covered a lot of ground. Here are the questions people tend to ask us when they reach this decision.
Frequently Asked Questions
How many machines do I need to cut a batch in half?
Two. Our FAQ says that rendering in parallel across multiple machines running Templater Bot cuts the total production time in half with two machines, to a third with three, and so on. That assumes every machine renders at a similar speed and the data is ready for all of them. Do a pre-flight on a typical and a heavy row, multiply, and divide by the number of machines. Then leave room for reruns and for a machine that is busy when you need it.
Can two Bots work on the same spreadsheet without colliding?
Yes, with setup. Add a render-status column and a bot column to the sheet, give each Bot a unique name in Templater Preferences, and put that name against each row. Mark rows ready and Templater moves them through queued, processing and done. Also make sure every row has a unique output name, because the Bot overwrites existing files that share a name. With a URL feed the Bot cannot update render-status itself, so register event scripts to do it.
Does a node-locked license work on every machine in a farm?
Each machine needs its own activation under a standard node-locked license, which runs on one machine at a time and moves by deactivation and reactivation. A floating license suits machines that come and go, because they lease a seat from a pool and return it when After Effects quits. Jon Christoffersen, who leads our support team, told us that if at most six are active at once, the floating key needs six activations. The key type is fixed at purchase.
Can I run the farm on rented cloud machines?
Our FAQ says Templater can run on one or more Windows Server machines in a cloud environment from one of the large cloud providers, and that Dataclay requires no special license for it. After Effects must be installed and running there, though, and Adobe’s standard license agreement does not permit After Effects on a server. You will need to speak with an Adobe salesperson about the right license. Settle that before you build anything.
Why does my Bot get slower the longer it runs?
Our documentation explains that when the Bot stays enabled for an extended period, Templater’s versioning can slow as it receives more data, because of the way After Effects caches layer changes. The fix is to turn off the layer cache in the hidden preferences, set the purge frames to zero, and disable disk caching on any machine where the Bot stays on. Try that before concluding that you need more hardware.
Does QUE spread jobs across my machines for me?
Not at this time. Our FAQ says QUE does not offer load balancing of jobs between satellites. You can assign jobs to specific satellites, and you must confirm that each satellite has the Autograf, fonts, plugins and other assets the job needs. For a small fleet that is workable. A large one means more manual scheduling.
Can I test a multi-machine setup before buying?
Yes. The free trial of Templater contains all features for all versions, and it adds a watermark to dynamic layers. Jon told us the command line works in the trial as well. You can run a real pre-flight and see how your data source behaves. The watermarked output is for testing, so it is not something to send to anyone.
