One of the design goals for Silverstack Lab has always been a fast transcoding engine. We knew that the speed of proxy creation is a crucial requirement for the DIT and her/his responsibility to create dailies on set.
While Silverstack 6.0 already got a new and faster transcoding pipeline, optimizations went on for the subsequent “dot releases” (6.1 and 6.2). Throughout all development phases we have always been asking ourselves the question: “How fast is the transcoding in Silverstack Lab?”
A generic answer to that question is almost impossible due to the many factors that influence transcoding speed. But of course the users and interested customers deserve an answer to it. The comparison of transcoding speeds to existing, other applications is one of the first steps to get closer to an answer.
For a meaningful comparison we have to consider a list of factors that have impact on the transcoding speeds.
Factors for the Assessment of Transcoding Speeds
The following factors play or can play a role in the assessment and comparability of transcoding speeds:
- Computer hardware Machine: The Mac used for the speed tests determines the available hardware resources for the transcoding. So testing identical format combinations on different machines will usually result in different transcoding speeds.
- Source and Destination Format: The different media formats of source and destination will result in different transcoding speeds. The algorithms behind different formats differ quite a lot and RAW formats are often processed with native SDKs e.g. provided by the camera manufacturer. The utilization of GPU and CPU resources also depends on the source format. The white paper about Automatic Resource Management in Silverstack Lab gives you more insights on the use of GPU/ CPU for different processes.
- Source and Destination Resolution: The resolution of the destination in comparison to the source influences the performed scaling and can therefore influence the transcoding speed.
- Transcoding Settings: The detailed transcoding configuration also plays a role. It results in speed differences if the transcoded clip should be created with or without a look baked in or with static or dynamic burn-ins. Debayer settings for RAW formats have to be chosen carefully because they can heavily influence transcoding speeds.
- Disk I/O: The available disk I/O speeds of attached drives can limit the ability for fast transcoding. High-bitrate formats such as ARRIRAW have huge demands for disk I/O, which sometimes can be overlooked and thus can result in a performance bottleneck.
- Operating System: We experienced that different versions of macOS can have noticeable impact on the transcoding speed. This can result in different speeds for the same combinations of software and hardware on different operating systems.
We can see that discussing transcoding speeds needs to dig deep into the specifics of the actual context. Exact format combinations, details about machine, operating system and many more have to be provided to allow for valid comparisons in a specific scenario.
Transcoding Speed Comparison to DaVinci Resolve 14
By taking the above factors into account we created a comparison of transcoding speeds between Silverstack Lab 6.2 and DaVinci Resolve 14. We chose the 2016 MacBook Pro 15″ as the transcoding machine for our tests. Fast Core i7 processors, a dedicated graphics card, H.264 hardware acceleration (e.g. transcoding to H.264 720p with up to 120 fps) and a fast internal SSD storage are some of the features that make the Macbook Pro (although being “just” a notebook) a competitive transcoding machine in a mobile setup.
- Mac Machine:
- MacBookPro 15 inch (Late 2016)
- 2,6 GHz Intel Core i7
- RAM: 16 GB 2133 MHz LPDDR3
- Graphics: Radeon Pro 450 2048 MB
- Source Formats and Resolutions:
- ARRIRAW 3.4K Open Gate (.ari, debayered with half-res)
- ProRes 444 3.2K
- Phantom Cine RAW 4K (debayered with full res default)
- Destination Formats and Resolutions:
- DNxHD 36 1080p
- H.264 720p
- ProRes 422 LT 1080p
- Transcoding Settings: All transcoding speed tests include a
- burned-in look,
- a dynamic (source TC) and
- a static (clip name) burn-in.
- Our tests showed that audio is negligible.
- I/O Speed: Transcoding from a 4-bay SSD RAID with around 1GB/s of read speed to the internal MacBook Pro’s SSD (around 500MB/s write speed)
- Operating System (macOS): macOS Sierra, 10.12
Speed Results Overview
* DaVinci Resolve currently (state: Resolve 14, Nov. 2017) needs to transcode different destination resolutions in two transcoding runs (one for each of the different resolutions).
Speed Result Interpretation
We see from the results that Silverstack Lab has competitive transcoding capabilites with typical format combinations and hardware. As this is just a comparison to one available transcoding tool, we don’t want to generalize with this test that Silverstack Lab is always equally fast or faster for transcoding than other software tools – especially if we consider the above mentioned factors concerning different machines, formats, systems, and so on. But we want to show that Silverstack Lab offers (at least) competitive transcoding speeds that help the DIT to get his work done in time.
Frequent on-set use cases require transcoding to different resolutions, such as DNxHD 36 and H.264 in 1080p and 720p resolutions. Silverstack Lab has the capability to simultaneously transcode to different destination resolutions, and together with performance optimizations for a lot of professional formats, and optimized utilization of the Mac’s hardware capabilities, Silverstack Lab offers great performance embedded in a powerful data and media management software.