Director and General Manager at AWS Creative Tools Antony Passemard told us about the company's solutions for game developers and discussed the future of cloud workflows.
Could you please introduce yourself?
I'm Antony Passemard, the General Manager for a service called Creative Tools, or rather a set of services. It's part of a group called Creative Tools within AWS. I've been at AWS for a little over two years, although I've spent time at AWS before, so I have about four years of overall AWS experience. I'm fairly new to the industry, as the last two years have been my first professional experience, although I have had a long-time passion for it. But as a job, it's a new space for me.
I come from a background mostly in cloud infrastructure. I've worked in the cloud for 15-20 years, primarily in software and cloud at companies such as Salesforce, Google, Amazon, and others.
I guess the biggest question our audience has when it comes to Amazon is what does Amazon have to do with visual effects, real-time graphics, and computer graphics in general? What's the interest there? What's the offering?
Amazon is highly involved in content production in general. We have five main solution areas. We have content production, in which I'm a part of. There's media and supply chain. There's data analytics, as you'll see in Formula 1, for example, or NFL. We provide all the real-time analytics when you watch the show. We have streamers, direct-to-consumer, and streaming and broadcast. That's Netflix, Hulu, HBO Max, and Prime Video, all powered by the cloud. The fourth one is archiving. I believe it's archiving, so there are five main pillars that we've been in for several years, probably quite a lot, where we've supported many workloads in general.
In terms of content production, the adventure started mostly when we acquired Thinkbox, which was about five or six years ago. Thinkbox and the whole portfolio of tools, along with the Deadline Render Manager, which is the biggest offering from Thinkbox, began about five years ago. This was part of EC2, the compute workload. And I believe three years ago, we established a dedicated team called Creative Tools to concentrate on Thinkbox and a few other offerings we had, such as Studio in the cloud. We really tried to focus more on how to create content in the cloud. Initially, it was more about VFX and animation. The Thinkbox team and some other teams we brought in had VFX backgrounds.
We have a fairly extensive portfolio of offerings and solutions that have been in place for many years and support some of the larger workloads you see out there. Avatar rendering is a significant example of scale. "The Crown" is a good example of speed, and "The Lord of the Rings" is a significant example of a distributed workforce with 1500 artists spread across the globe to accomplish this.
Tell us a little bit about this idea of having your studio in a cloud. What is the demand from the market like?
Yeah, it's been something that's been happening in the last three, four years, driven by two things. The first one was driven by the demand for content, like the streamers, in particular, were asking for more and more content. The number of productions that were happening was just growing massively, and so there was a huge demand for content that the studios could not deliver just with their own on-premise infrastructure. So they were turning to the cloud to say, "How can I get more content? How can I win that project and actually deliver it when I have a small farm I can't use? I need to burst to the cloud and use that." That was the first driver, the demand for content from broadcasters.
The second driver was the pandemic, because when the pandemic happened, people were not on-premise anymore, not in the same studio. You had demand for content during the pandemic became even bigger. People were actually watching TV a lot more, and you had to have distributed teams, and you can't do distributed teams without the cloud. Like, technically you can, but it's much easier to do when you have a studio in the cloud. So that drove demand drastically. The other thing that drove demand, that's less known, is the crypto surge. It started to be big, and they were grabbing all the GPUs that were available, so you couldn't buy a GPU. There was a supply chain issue. You could buy GPUs to get on-prem. The cloud had GPUs available for you. So those three factors kind of came in and really started that whole notion of a studio in the cloud.
Studio in the cloud is where you have a workstation, usually GPU-based, in the cloud. You have your data in the cloud that can be distributed between regions, and you have your rendering capabilities in the cloud because you don't want the data to go up and down from the cloud every time you have something to render. So when you have this trifecta of workstation, storage, and compute in the cloud, you can be completely global, you can have virtually unlimited scale, and be really flexible. Also, when you see the writer strikes that happened recently, most of the studios that were in the cloud just turned off their machines, no cost, versus if you have on-premise infrastructure where you have your data, you have your leases going on, you're still paying even though you don't have jobs. So that was really a demonstration of the value of that elasticity that you get with the cloud and the fact that you can get artists from anywhere in the world.
One of the questions we get, especially when we work with cloud services, is what about coverage? Are you represented in different regions?
Yeah, it's a question that we often receive from customers: how will the latency be when I'm rendering and then attempting to work in real-time? Yeah, trying to work in real-time with Maya or Blender requires low latency. Currently, we have 102 available within 32 geographic regions around the world, and we actually have plans to add 12 more availability zones and four more regions in Canada, Malaysia, New Zealand, and Thailand. But I can provide you with a map right now.
This is expanding rapidly to achieve low latency for virtual machines. One good example is that we have studios in LA, and in LA, we don't have a region, but we have what we call a local zone. These are smaller zones that bring the network and the machines closer to the customers. So in LA, if you have your full machines, you'll use the local zone and experience super-fast round trips. It's almost indistinguishable from a local workload if you're working in any of the DCCs. We're expanding, but we're not everywhere yet, obviously. So when we talk to customers, the first thing we do is conduct a little test to determine the best region to access and the best fiber we can connect to. We're trying to figure out the best data center region they can connect to for their workloads.
So tell us a little bit about how exactly does it work? So let's say, what do I do? Do I buy a Maya license to run in the cloud? Do I just sort of leave the computer in the cloud? Do you work with Autodesk and all the other partners? Tell us a little bit about it.
There are two sides to that question. There's the desktop and workstation side, and there's the rendering side. Two different ways of managing your licenses. On the desktop side, you will go to EC2 and create an Amazon Machine Image (AMI) that has all the software, including Maya and all that. That's your machine. You spin it up, and you have all your software on it. The machine needs to connect to a license server for Maya, as an example. So, you buy that from Autodesk, set up your license server, and configure your machine to connect to that server. That's on you as a customer to manage the license on the desktop side.
On the rendering side, Thinkbox has what we call usage-based licensing. When we created Thinkbox in the cloud, we wanted to allow customers to scale quickly. If they need 600 machines to do something, they don't have to go to Autodesk and buy 600 Maya licenses and put them on the workers. That could take several weeks potentially. So we have usage-based licensing where we manage all the Autodesk licenses. You can spin up and down your workload, and we charge you for what you use. That's a different model that we work with other vendors to enable for customers, so they don't have that restriction on how many machines they want to use.
When you're approaching customers, and you're talking with them, what are the friction points?
It's interesting because the M&E industry is extremely innovative from a technology standpoint. Every new movie has a new algorithm, a new AI, a new this or that. But at the same time, their architecture is still very old-fashioned, if you will. It's still very much on-prem. They buy servers, they buy racks, and they buy storage that they have in their own data center. It's kind of interesting to see that it's driven by the way the budgets are made for a movie. You have a certain budget to do a certain amount of shots, so that certainty in how much money you're going to spend is something that's been a hindrance for cloud adoption, where it's much more flexible. You're going to pay for what you use, and some of those studios have a hard time making the calculation and the total cost of ownership of running an entire studio in the cloud.
We have to be educating the market on that because we've seen studios very successful. One studio, for example, has been starting from the cloud and has been ramping up and down in the cloud as projects were coming in, and they're very successful with that business model. But they embraced it from the start. Versus studios that have an existing infrastructure and you're like, "Well, when you go to the cloud, where we have this thing, so we know that." And we meet customers where they are. For example, Deadline Render Farm allows you to run on-premise and in the cloud and burst in the cloud. We totally allow those hybrid use cases, and we want those hybrid use cases to be enabled and exist because we want customers to realize and be able to learn a little bit about what's the cost of running the cloud and what's the total cost of ownership. That's really important.
And how do you battle rejections to the point where they see too many dependencies?
The dependency is not strong. That was my talk today, all about open source and how we push that to remove those dependencies as much as possible.
For example, last year we did two things. We released all our Thinkbox tools, like Katana, XMesh, Frost, and the other one I'm blanking on the name, all of those tools for particle simulation and particle rendering. We released all of them as open source. These are for Maya and 3DS Max mostly. So, they are all open source, and we also made Thinkbox Deadline free. It's still not open source, but it is free, and we've had thousands of customers downloading it since then and rendering with it. So it's pretty amazing to see.
The goal of the open-source initiative that we have right now is to remove dependencies. Right now, you are dependent on Thinkbox to release the submitters for Maya and Max every time there's a new version. We want to get out of this. We want the customers to have the freedom to release their own submitters if we don't have one, modify the submitters to have their own add-ons or their own specificity and customization, and really not wait on us for that.
What I was talking about in the session was a new format that we're working on for job submission. It's a way to specify the job that you want to run on a render farm in an open format, which is human-readable and in text. You can push that to any farm that supports that format, and Deadline is going to be the first one to support it for sure. But we want others to support it to make sure that you only write those submitters once, and that you have control of those submitters. You write them once and then you can push them to any render farm you want, not just Thinkbox, but any render farm you want, in the cloud, on-prem, and other vendors, as long as that format is supported, that specification is followed. You'll be able to describe your job, the steps, the dependencies, the assets you want to bring in, the iteration, describe all of that, and it will run everywhere, with any vendor. That's something that we're pushing and we're very passionate about. That was the main subject of my talk. We're trying to get feedback right now from different vendors to make sure we're aligned and we're not forgetting anything, and we're going to push that out.
Can you tell me a little bit about who are your customers?
In the VFX offering, you talk about the big players like Weta or Netflix on the streamer side. HBO Max is there, Hulu is there, Peacock, and all the streamers. We have fairly massive VFX studios, and production houses that are customers of AWS. We have a pretty long tail of customers that we know they're there because they download Deadline, but we don't really hear from them. So it covers the whole gamut, from tiny studios probably doing that on old machines to very, very large customers.
So, overall, from Amazon's perspective, what should we expect in the future? Do you expect to see even more demand for content? Do you think this demand is going to shift from larger companies to smaller companies? Are you expecting consolidations in the market? Anything you could share from your top perspective?
I'm definitely expecting more content to happen. That's not slowing down, that's not what we're seeing. Obviously, the writer strikes have slowed down things, but we're expecting quite a growth when things start to restart, which is good. We're expecting a lot more content happening. What we expect is that with proper cloud solutions, we're going to see smaller studios being able to take on more. That's exactly what Untold's story is.
When Untold started, there was a small team, and they told us they were able to get those bigger projects that they could never have done if they had to buy an entire render farm on-prem to do it. But because they were able to scale as the demand came in, they were able to take on those bigger projects. They've done fantastic content with that. So I'm expecting this to happen more. Smaller studios participate in broader, bigger projects more easily.
And when the project is done, they just turn off the infrastructure, and that's fine. They can go to the next project and then turn that backup. So, I'm seeing that flexibility and the ability for smaller studios to do more. But I'm not expecting content to be reduced; it's going to be more and more.
The need for VFX is now in almost every movie, even romantic comedies have some VFX. It's kind of interesting. So that's not going away. There's also a bit of convergence with real-time, which is an interesting trend as well. We have a session this afternoon you can watch that. Deferred rendering is not going away for sure. There's always a need for super high-end quality effects or composition that needs to happen, which can't be done in real-time. But real-time is getting more and more present in the pipelines for previz, even for rendering at the end. That's an interesting trend that we're seeing with real-time rendering.
Is this real-time rendering something you're also embracing, I guess, like through your virtual desktops?
Yes, virtual desktops, but also Unreal Engine and other game engines. 3D is one that is open source that we released. What we see is people do real-time on the virtual desktop or their own desktop, and at some point, they want to do a cinematic for a game that will go as a different rendering. So this integration between Unreal and Deadline, for example, is happening. This is what the session is about this afternoon. You can do your whole real-time, but when you need to render at super high-end quality, you just push a button, and it goes to Deadline and renders its frame like a traditional deferred rendering. So we see those two components coming together.