Cédric Lejeune (Workflowers)
Though sustainability is often linked to climate change, it also relates to other topics, such as the availability of technical resources. We already know today that we can’t get all the graphics cards we need, and it’s not only because of crypto miners: it’s also because Taiwan faces a drought and producing chipsets requires a lot of water.
Car manufacturers are halting their production lines, and my friend who shoots documentaries cannot buy a second camera because the model she needs is not available anymore.
The media industry is right in the middle of its second digital transformation: AI, cloud workflows, LED walls for virtual sets. These new technologies use a lot of technical resources – advanced electronics, complex alloys – and despite “digital” becoming ubiquitous, we still use a lot of industry-specific equipment. The move to IP infrastructure already allows us to get rid of a lot of SDI equipment and replace it with commercial off-the-shelf IT devices, though sometimes using them in a very different way to most other industries. Nevertheless, ultimately other industries use more and more video so we may see convergence in the architectures offered by cloud providers.
To reduce emissions and power consumption some providers are installing datacentres in cold countries such as Sweden, where they require less cooling. This is significant because traditional refrigerants can have a terrible greenhouse effect, up to more than 2,000 times that of CO2.
Another approach can be water-cooling, and Microsoft has taken the concept to the limit by creating an underwater datacentre, a strategy that could reduce waste heat, which will be more and more regulated.
Smart use of AI
Lately at Workflowers, we have been testing the use of AI to reduce the rendering time of animation content when creating higher resolutions, rendering 720p from the 3D rendering engine and upscaling that to output 1080p or UHD. While the first results are promising, it also shows that there’s a lot of progress to be made. AI and machine learning take a lot of computing, but they can have a significantly positive impact when used to reduce computing requirements somewhere else. Video
compression is a very interesting field of research for AI, as distribution of more content to different formats requires massive infrastructure. A better understanding of the content before it gets encoded should help the reduction of data rates.
At some point we may also question producing and distributing 4K or even 8K content when TV manufacturers are differentiating themselves on the AI processors they integrate in their sets (although chipset shortages will have an impact on that industry too).
Producing in full resolution and/or HDR could be reserved for premium content, avoiding the need to replace a lot of gear and instead keeping it for a longer time, which is also a great way to optimize emissions.
Our business uses a lot of resources. We may find solutions that have a positive environmental impact, by reducing the need for transportation of people and data, but ultimately that will only have a relative impact on the emissions per hour of content. If we’re looking at the absolute values, the only way to reach our targets for carbon emissions is to change significantly the way we produce and consume content. That leaves a lot of room for innovation, of the disruptive kind, because going +1 or x2 with typical incremental innovation is certainly not sustainable, and very soon we will see the model break. But it the end, is that really a bad thing?
This article was first published in issue 50 of tech-i magazine.