Cloud has been touted as the next big thing in production for years now, as software makers switch to subscription models and leverage web-based interfaces for more flexible accessibility across devices. However, until just recently, the cost of cloud rendering has yet to reach the tipping point where it made sense for studios to move away from physical resources. Local farms have been the norm for decades and while the prospect of unloading thousands of dollars of legacy infrastructure can be downright scary, it may be the smarter choice in the long run. There isn’t a perfect one-size-fits-all for determining that tipping point metric and it may still be up for debate, but by looking at a few key factors, studios can get a pretty good idea of how cloud rendering might impact their bottom line.
When Cloud Makes Sense
Beyond the initial CapEx requirements for hardware, machines have to be physically housed, powered/cooled, and managed by someone with some degree of technical aptitude. They also have a limited shelf life, and as technology progresses, hardware will need to be serviced, upgraded or replaced. If production is largely steady and predictable, and real estate relatively affordable, local hardware may make the most sense for a studio, though it does limit the ability to scale. Farm utilization is also a huge consideration. In some cases, a farm can sit idle 20-50 percent of the time, which is lost productivity and value. In contrast, a farm that runs 100 percent constantly will encounter rendering backlogs and, as a result, reduced productivity. Companies that purchase additional hardware to complete a job will be faced with underutilization when production wraps; it’s a vicious cycle between having too much or not enough compute power, and a failure to manage this dilemma can (and has) put companies out of business.
Now that physical and cloud costs are mostly aligned, the ability to scale is a major benefit of cloud. When production peaks, studios can scale much more quickly through the cloud than traditional infrastructure build outs, and only pay for resources when used. This flexibility allows studios to better manage rendering costs and also allows them to accept work that they would have otherwise had to pass on because of render farm capacity. And with cloud, setting up a new studio outpost to better service a production or capitalize on tax incentives can be achieved in a matter of weeks and with minimal investment.
In addition to scale, cloud offers machine flexibility. Typically, local render farms have a limited number of variations. Since many options are available in the cloud, each with different costs associated, facilities can select the right machine for any given job. Depending on a production, costs savings can be significant if lower RAM is used, but it will also take 35-40 percent longer to complete a job. When work needs to be done in a specific window of time, it may be worth paying more for faster compute, but ultimately, studios get to decide if turnaround time or cost management are more important for a particular job. While faster returns are generally more expensive, it provides the benefit of allowing productions to work in parallel, with shorter review cycles that minimize delays and better utilize artists’ time, which is generally a studio’s most valuable asset.
Three Golden Rules for Cloud Efficiency
This may seem like a no-brainer, but one of the most important criteria for optimizing cloud usage is for studios to know their work. When starting a job, select and run a few worst-case scenario frames for a scene to benchmark performance and render times, then a supervisor or lead can use that information to determine the appropriate machine spec for a body of work. If there are dramatic variations within a scene, consider breaking it up into chunks of similar complexity and using different machines for each chunk. Start slowly with only a few machines to control the rate of spending when assets, software or other techniques have changed, and sample frames before ramping up full force for a large volume of work.
Knowing the budget is just as, if not more, important as knowing the work. With local render farms, it is too easy to think of compute time as free and not track that usage against an actual budget, but rendering should be included in production charges, whether local or cloud-based. When bidding projects, you can execute select tests on key material to validate and improve on estimates. Those estimates can be used to set limits for spending, whether managed manually or through something offered through your cloud service, like Conductor’s Cost Limit functionality.
As the saying goes, those who do not learn from the past are doomed to repeat it. Studying histories of past jobs and costs provides a wealth of data that can be used to improve the bid process and target areas for optimization. These insights can be mined from local farms, depending on how that data is managed, but in most cases, cloud automatically stores and makes these metrics accessible.
Ultimately, there is no one way to assess and manage cloud use in production. Each organization should be willing to do some experimentation as to what does and doesn’t work, and see if cloud is the right fit for its culture and people. With the right implementation, cloud can be the determining factor in whether or not a studio thrives and grows, let alone survives in this cutthroat industry of razor thin margins. Cloud rendering is already transforming how VFX, and to some degree VR/AR, are created, and I, for one, am excited to be a part of what’s next.