As we have already said, many people exploit “quiet time” on their workstations for rendering; they submit frames for rendering either out-of-hours or opportunistically, and this is a great, cost-effective option. However, there are other factors, such as rendering performance, which mean a dedicated render farm may work best:
- Artists are more creative – When scheduled rendering is not an option (maybe because of tight deadlines and the need to keep the creative process moving) a dedicated render farm won’t draw on the workstations’ processing power or RAM, leaving applications snappy and responsive.
- Extreme rendering performance – A dedicated render farm is optimised for fast rendering, so the completed job is returned as quickly as possible.
- Accommodate new projects with ease – The rack-mounted hardware used in dedicated render farms is fully scalable, so if you start a new project you can “scale out” your render farm by adding new render nodes. In an emergency you can also pull in some workstations as additional nodes.
- Low profile but maximum processor density – There are hardware options that offer thousands of cores per rack, meaning only around two feet by four feet of expensive floor space for a powerful render farm.
- Protected by server room facilities – Most dedicated render farms can be located alongside other computing facilities and tend to enjoy the protection and security of uninterrupted power, cooling, industrial grade power feed, and restricted physical access. In other words, maximum uptime.
- Workstations – Regardless of which platform you’ve chosen, whether it be Boxx workstations or Mac Pros, and whether they’re optimised for professional 3D animation or games development and visualisation, a render farm can be assembled to match.
- Ethernet switches – Connecting the render farm to your workstations, we can recommend options for high performance or close integration with your existing network infrastructure. Gigabit Ethernet is the most common choice today, although 10 Gigabit is available. Leading brands include Cisco, Juniper, HP ProCurve and 3COM.
- Render nodes – Using rack mounted blade servers it’s straight forward to build a hugely powerful farm. Great choices here include the Boxx RenderBOXX 10200 system or a bespoke HP C-class Blade system.
- Storage networking – It’s important to identify the right technology for shared back-end storage. The best option for you will depend on your exact workflow. These options include iSCSI, Fibre Channel, FCoE, InfiniBand, and simple NAS over Ethernet. Each has its specific pros and cons that we can cover.
- Storage – There are some fantastic options for storage. For example, the Isilon IQ series scales performance right up to 20Gb/sec with 3.45TB of shared storage. Even if the numbers are not your thing, the simplicity of management is compelling. For example, more storage can be added, as you need it, without downtime.
The table below shows some examples of the amount of storage that you may require:
10 minute long footage (14400 frames total, 3 passes) = 43200 images
Resolution Open Exr File size Total size of project files
1920 x 1080 30MB (per image) 1296000MB 1280 x 720 14MB (per image)
604800MB 640 x 480 5MB (per image) 216000MB
Points for Diagram:
A) An artist clicks the render option in his 3D application and is immediately free to continue with his design work. The job passes to the Supervisor node of the render farm.
B) The supervisor node breaks the animated sequence into frames and allocates them to specific render nodes.
C) The render nodes pull files they require from high-speed shared storage and process the frames.
D) Now complete, the rendered sequence is pulled together and made available to the artist.