This article is intended for use by organisations that have considered purchasing an internal render farm, but have not yet come to a decision in the matter. In writing it, my goal has been to introduce decision-makers to the factors that are worth examining in order that their determination be a solid one.
It should be noted that this writing targets smaller architectural and design studios and freelancers who find themselves under considerable pressure with regard to render times and to whom my own personal experiences apply. The programming environment I have used for this purpose includes Windows, 3ds MAX, V-Ray, Backburner, and Corona Renderer.
In addition, this article may be instructive as a source of additional ideas and views for companies that already own a render farm or rendering software.
The key issues:
Will the profits I gain in increasing render capacity offset the cost of the investment?
The calculations required to answer this question are complicated and the results subject to interpretation, so before I get started, I would like to point out a number of options and issues. To begin with…
With increased render capacity, would I be able to sell my customers more images than I do now?
Ideally, total render time (including testing and draft times) should represent a smaller fraction of the image creation work process than, for example, modelling does. At the same time, the more views of a single scene are required and/or the greater the complexity of the specific jobs involved, the more likely it is that inadequate render capacity affects the image output achievable within a given space of time. In the event it is for this reason you find yourself imposing limits on the number of renders you promise your customers, then it is probable that eliminating the restriciton will lead to increased revenues. In any case, it is worth offering your customers additional camera views in order to find out.
If I had surplus capacity, could I provide my customers with final renders in a higher resolution?
In keeping with the line of thought of the previous response, you might offer your customers a higher-resolution final product (e.g. 4K or billboard size) for a small surcharge per render. This is as easy as inserting a line into your price quotation stating the charge for – say – 4K as opposed to HD, an option that, despite requiring attention to more details within a given, involves a significant increase in render time rather than work quantity.
Would the ability to deliver higher-quality images mean much to me?
Have you ever found it necessary to adjust render quality in order to meet a given deadline, forcing you to turn in work that was not up to your usual standard? I have, several times, and was embarrassed when my customers pointed it out. It can be a great relief in a tight situation to have a capacity that permits you to deliver high – or at least acceptable – quality even when you find yourself short of time.
How much faster could I work if my test render times were significantly reduced?
This is a factor I believe many people forget in assessing this problem. Where extra render capacity is consistently available, test render times decrease, a circumstance that can speed post-modelling work processes considerably. How much time do you spend at your machine waiting for test render results, and how much is that time worth to you? How much money would you save in a month if you could recover “just” a half hour a day of that time?
How much faster could I develop my business if my render capacity was higher?
The ability to work faster with more test renders accelerates learning rates and lets your team gather experience faster, permitting additional experimentation and the trial use of new methods. Though this factor cannot be quantified (at least, I have not been able to do so), it is certainly worth thinking through…
Is it possible that, with enough capacity, I could take on animation rendering jobs?
Because an animation render requires far greater resources than even multiple stills, the final product of an animation job will almost certainly entail the use of an external render farm service. Still, the prep work for animation rendering proves nearly impossible if at least the “draft-level” sequences have not been prepared in advance, a task that requires considerable internal resources. With a little luck, however, along with a carefully considered deadline and/or well-optimised scene, one can gain quite a bit, including considerable financial savings, if at least some of the final calculations are done in-house. In an example from my own experience, I found I was able to complete a 3-minute animation in the time window of a single weekend without resorting to outside help thanks to two full days’ availability and a successfully optimised scene.
Will I find that I can’t do other things – say, modelling – because I have to wait for my rendering to finish?
This concern pertains primarily to those who have only a single work station at their disposal. In such cases, in general, rendering cannot be done in parallel with other work processes.
Though there are certain to be questions I haven’t considered in making this list, I imagine this will at least do to get you thinking, and it could even be that in reading through my responses, you will come up with advantages that have not yet occurred to me.
All in all, what the above shows is that the construction of an in-house render farm can in some cases be an attractive prospect. There is, however, the other side to the coin.
This is, of course, the most obvious consideration. Every additional item of computer hardware and software, and every new accessory comes at a cost.
In this particular case, the expenditure breaks down as follows:
Accessories (e.g. router, file server, service monitor, mouse, keyboard)
The construction and operation of a render farm requires both IT expertise, and a certain investment of labour. Where you have a system administrator to guide you, however, the transition to network rendering can be accomplished quickly.
You will need (if not already owned):
An appropriately configured and installed internal network, with properly set permissions, work groups, and domains – a task best entrusted to your system administrator.
Files and textures saved to and opened from a central server, so that all machines in the network read scenes from the same location.
A little network rendering know-how. (Where Corona or V-Ray are concerned, this can be acquired in just 1 or 2 hours. With Backburner, a little more time will be needed.)
A render farm requires space which, if it is to be operated safely, must satisfy a number of specific conditions. The following list represents the optimal solution. The more machines you plan to install, the more important it is that these conditions are met. (For three computers, of course, you may not even need a separate room.)
You will need:
A separate, air-conditioned room.
A shelving system for the computers.
Adequate electrical wiring (One outlet cannot safely handle two machines).
Fire safety equipment.
And now that we’ve reviewed the factors involved, we can begin assigning some numbers.
To help us, I have prepared a table with an attached calculator employing the above criteria. In it, certain basic values associated with hardware and software are invariable. Clearly, however, the values associated with time and end product will vary from one organisation to the next (not to mention from one country to the next), while different companies use different types of computers and are subject to different operating costs. Accordingly, the fields coloured orange and green can be modified to adjust the calculation.
(The calculation produced by this table does not account for every minute cost, nor include every possible source of revenue – those associated with animation, for example. Instead, it helps determine an order of magnitude. Certainly, before making any final decision, a more detailed calculation tailored to the specifics of your company would be a good suggestion.)
The implementation of a render farm, therefore, involves a larger cost at the outset, including the prices of computer hardware and software licenses. (Of course, since used hardware can be sold, this is not an out-and-out cost or loss, but I’ve regarded it as such for the sake of simplicity.)
The monthly expenditures associated with the investment consist of depreciation on the machines, increased overhead, monthly software support fees, and added labour costs. As the table indicates, the cost of the system administrator’s time will be high at the time of installation. Later, maintenance, including such things as repairs, replacements, and program reinstallations, will require no more than a couple of man-hours per month, with additional hours expended on machine management, powering on and off, spawner and render monitoring.
On the profit side of the coin you have the achievable increase in the number of images sold, surcharges for increased resolution, and work hour savings. My calculation proposes no profits for the first month, and only half the expected value for the second month, as I assume some time will be required for the new work process to solidify.
Though if you have no experience in the area, the numbers in question are difficult to estimate in advance, thinking through the following points can be of help:
Consider carefully how many more images you could produce in a month with several times your current render speed. Calculate using not only the final render time (it won’t be significant compared to that of the entire work process anyway), but also test and draft render times, which the system should reduce. The total should now represent a sizeable slice of the process. The relation between time savings and number of machines is not really linear, but you may assume it is as a rule of thumb (the more machines are added, the more linear it will be.)
Work hours liberated through shorter render times, when used for other projects or other purposes within the same project, constitute a profit for the organisation. To estimate the associated value, you will need to calculate the time spent in test rendering and re-rendering.
These two considerations should not overlap; if liberated work time is applied to producing additional renders, it should figure in that calculation only, while if even more time savings is expected, then this extra time should be entered into the second calculation.
The table encompasses a total time span of 18 months. If the investment cannot be recovered by then, it is probably not worth considering. Hardware and technology change too rapidly for investments recoverable over longer times than that to make any sense. (For results close to the break-even point, however, the possible additional advantages make a positive decision still worth contemplating.)
If the difference between the two is sufficiently large, then it is certainly worth moving ahead. If you are risk-averse, it is a good idea at first to put off buying any larger quantities until you have seen what can be achieved with an array of up to 5 machines. If these bring some financial advantage and a re-evaluation of the above predicts the same for additional machines, you can always increase that number later.
Before ordering machines and software on the basis of the Table provided here, make sure you weigh the following just one more time (and if necessary, use your results to recalculate):
Will I really have enough additional orders to make full use of the extra capacity?
Is there a pressure on me and my team that can be alleviated through added capacity?
Is the inability to render quickly really standing in the way of my development?
Am I building my scenes and setting up my renders as best I can? Is there anything else to optimise, or is hardware really the only thing
standing in the way of increased speed?
If the answer is still yes, then an in-house render farm is probably for you!
Because every studio functions and works differently, it cannot be said with certainty that a long-term investment into an internal render farm is always worth making. Yet my own personal experience suggests it is likely to be. I have found that the more important quantity, resolution, and prompt project reaction speeds are to you, the more worthwhile the contemplation of an internal capacity increase will be.
This article is soon to be expanded: in the next part, we will be weighing various criteria useful in the selection hardware.
https://rendernodemonitor.com/wp-content/uploads/2019/01/internal-render-farm-2.jpg536800Gergely Toth/wp-content/uploads/2018/06/logo.svgGergely Toth2019-01-09 00:04:462019-01-09 10:07:33To Build or Not to Build?