Data & Dimensions | Ep. 14 - How smartphones are outgrowing casual gaming

View Full Transcript

Peter Hodges (cold open) — 00:00:00
I really want to see that flywheel effect we were talking about earlier. I want to put developers in a position where the constraints they traditionally fought against—like “what is my mobile graphics budget going to be?” or “what is my graphics budget in general?”—versus what is my creative ambition, can be brought closer together. I want to broaden that envelope.

Neil Barbour — 00:00:28
Welcome to Data Dimensions, a podcast at the intersection of the entertainment industry and interactivity. I’m Neil Barbour, an analyst with S&P Global Market Intelligence, Kagan. Let’s talk about mobile games. We estimate their total addressable market at around $110 billion annually.

But most of the money flowing into mobile games goes through the casual subsegment—think clickers, Match 3, board games.

There are a few reasons mobile has mostly been the domain of low-impact, quick-hit experiences—like the form factor and the input mechanisms. But one of the biggest reasons is performance issues. The high-end chipsets and premium mobile phones can absolutely produce immersive visuals, but that often comes at the expense of battery life and a very hot phone.

That’s where Arm typically comes in: the foundational compute platform. Its CPU and GPU designs underpin much of the mobile market, with a long-standing focus on performance per watt.

Recently, Arm teamed up with Sumo Digital to show off a high-end lighting technique running on a mobile chipset—highlighting the broader opportunity around premium mobile gaming.

To talk about that with me today are Peter Hodges, Director of Developer Ecosystem Strategy at Arm. Peter, thanks for being on the show.

Peter Hodges — 00:01:40
My pleasure. Thanks for having me, Neil.

Neil Barbour — 00:01:42
And Lukáš Medek, Art Director at Sumo Digital and Game Director of Neural Dawn. Great to have you, Lukáš.

Lukáš Medek — 00:01:49
Great to be here. Thank you.

Neil Barbour — 00:01:51
So these two have teamed up to produce Neural Dawn, a production-scale mobile game designed to show lighting effects once reserved for PC and console hardware.

Peter, why don’t we start with you? How did Arm and Sumo Digital come together on this, and what was the north star for this project?

Peter Hodges — 00:02:07
Thank you, Neil. It comes down to a shared question: could we take neural graphics and turn them into something that’s a real asset to game developers—something more than a fancy tech demo?

We wanted something that delivers continued value. With Sumo, they brought game-making craftsmanship and a trusted, tested content pipeline. And just as importantly, the creative pressure test—really helping us answer the question that brought us together in the first place: is this technology genuinely useful in the creation of games, and in expressing visual storytelling?

Arm brings the technology from the mobile GPU perspective—power and efficiency—and our ambition to craft that space in the future, and define the frontiers.

It’s about showing beautiful results, which I hope many people will appreciate in Neural Dawn. And from an ecosystem perspective, it’s about reusable evidence—what works, what’s proven to be hard, and what’s been solved. It’s also about understanding what developers need from SDKs, tools, documentation, and performance guidance, along with what the partner support model looks like. Having all of that in hand—that’s the path to success.

Neil Barbour — 00:03:31
And when you say “crack that space,” you’re talking about high-end, immersive gaming on mobile. You kind of think that’s an underexplored space in the market?

Peter Hodges — 00:03:41
I think so. We find ourselves in a place today where the tools are there for developers to explore. And it’s always been a very challenging environment—ways to differentiate, to stand out, and to tell the stories game developers want to tell—while still meeting the demands of quality and timing that the industry always requires.

Neil Barbour — 00:04:02
Right. Interesting. So setting the flywheel of creating supply to spark demand—creating supply and sparking demand.

So Lukáš, let’s talk about the game itself. It feels like you were careful to find a concept that really fits the tech. Tell us about Neural Dawn and why it’s a showpiece for this technology.

Lukáš Medek — 00:04:18
Sure. The important thing to say is that usually when new tech is being presented or developed, there’s a showcase demo—very short, and mostly focused on the tech itself.

With Neural Dawn, it’s a little different. After initial discussions with Arm—about the project and how we would best showcase the neural graphics plugins—we realized the demo wasn’t enough.

So Neural Dawn is a full-scale game, developed by a very small team, and it’s still in development. It’s a third-person adventure game for mobile. The aim is to bring console-level visual quality to mobile.

We chose the genre and third-person viewpoint because it’s commonly used in PC and console games—but it also fits the tech we want to showcase extremely well.

On the narrative side, very quickly: it’s about an android who awakens after a long sleep, explores a huge game world, learns about her past, and tries to finish a research project started a long time ago.

The main character uses flying drones to interact with the environment. And the interaction with the environment is the core mechanic of the game. In almost all cases, those interactions are tied to changes in lighting in the environment.

Neil Barbour — 00:06:29
Right. It’s really cool. I’ve watched some of the videos—I haven’t had a chance to play it myself. It kind of looks like, you know, you said third-person, but I’m going to say Dead Space style, behind-the-shoulder. And to watch all the lights—illuminating different aspects of the environment in real time—it really feels like something we haven’t seen on mobile in a big way before.

So Peter, let’s talk about what’s happening in the rendering pipeline to help all those lights illuminate the space. What’s happening under the hood? What’s the special sauce?

Peter Hodges — 00:07:03
I’ll try to do it justice. Neil, you can absolutely say “ray tracing” when talking about the lighting solution.

The short version is: to render scenes that look this good, it comes at a cost. There’s a computational cost. There’s a power cost. And if you use a conventional approach—pushing that many lighting effects—you spend a long time in the GPU handling shaders and resolving textures into the final presentation.

So we use neural techniques to mitigate some of that expense in a couple ways.

First, we use a super-sampling technique. We render at a lower resolution and use neural techniques to upscale.

Second, we denoise. The reason is that ray tracing is inherently expensive: rays pass through complex scenes, and the sampling is quite incoherent, meaning rays can bounce in unpredictable ways. That also tends to cause memory access patterns to be unpredictable—which costs a lot on mobile performance.

So we test fewer pixels. That saves work, but the tradeoff is that images become noisy. Then we apply a neural denoising technique.

Finally, we insert frames. We probably don’t render the scene fully every frame. Instead, we render every other frame, and the frames in between are generated by interpolation. We’re not doing a full “from scratch” render every single frame at the shader level; we’re taking two frames and interpolating.

Taken together, this is an uplift in power and performance without sacrificing quality—and hopefully, giving players a great experience. And Lukáš can affirm that from what you’ve seen.

Neil Barbour — 00:08:50
It sounds like some of the same tech being deployed in high-end PC graphics cards—DLSS, frame generation. Is it similar? Are they essentially related concepts implemented more efficiently, just at a different chipset level?

Peter Hodges — 00:09:08
It’s a very fair comparison. Solutions like those exist in the PC space. But this is where Arm’s experience really comes into play: we take neural networks that have been used for high-end use cases, and we make them operational within a mobile power and performance budget.

So you could call it a democratization of techniques that have been seen on high-end PCs today.

Neil Barbour — 00:09:32
Lukáš, I want to double-click on MegaLights—that’s still kind of bleeding edge, especially on mobile. My understanding is your team had to work directly with modified engine source, not a fully turnkey mobile path.

What was that experience like?

Lukáš Medek — 00:09:48
Yes. For me, it was amazing—because I’m an artist. Having the possibility to use this kind of high-end feature in the game is very cool.

We’re working with a modified version of Unreal, because MegaLights is supposed to be used only on PC or consoles by default. So we have to modify the engine to make MegaLights more performant so it can run on a phone. And we had to do a lot of optimizations while maintaining visual quality.

So, yes—for me it felt like magic. I never imagined using this kind of tech with ray tracing on mobile.

Neil Barbour — 00:10:52
Yes. So is the idea that Epic would internalize the tools you’ve developed?

Lukáš Medek — 00:10:57
Actually, Epic was pretty amazed that we’re able to run MegaLights on mobile. And the thing is: there isn’t any other released game using MegaLights with this feature set.

So we are basically going to be the pioneers of this tech—on mobile, and at the top end.

Neil Barbour — 00:11:19
Not even on PC and console are there MegaLights games?

Lukáš Medek — 00:11:22
That’s right. That’s right. So we are basically—yes—we will be the first pioneers of this tech on mobile.

Peter Hodges — 00:11:30
Perhaps I could add a bit, Neil. The changes we made to implement MegaLights in Unreal Engine on Android were fairly extreme—but in practice, they were relatively small changes to settings files. What we’re looking to do is share learning resources on our site that explain exactly how to do this.

Neil Barbour — 00:11:50
Yes. So Peter, let’s stick with you. You get into the alphabet soup: two technologies in the pipeline.

The NSSD upscales from 540p to 1080p in about 4 milliseconds per frame, and the NFRU doubles frame rates by generating synthetic frames between the rendered ones.

Are these primarily GPU workloads, NPU workloads, or a mix? And how much depends on future silicon versus what’s possible today?

Peter Hodges — 00:12:19
We’re going to have to call it a mixture. But intrinsically, these are graphics pipeline workloads. They exist to render pixels onto the screen at a certain frame rate.

So the challenge—and the solution—has to sit very, very close together on the hardware. We combine frame information such as history, motion, depth, color, and lighting. And it all has to happen in tight coordination.

Because latency is an issue, we don’t really think about offloading this to an NPU or waiting for a result. Instead, the GPU remains at the center of the stage. Our neural technologies are closely coupled—on the same die as the GPU, right next to the execution engine on the silicon.

So the neural models do this complex work within the frame budget, in a way that works in concert with the GPU rather than being some external unit.

Neil Barbour — 00:13:30
Would this run efficiently on, say, a Mali-based chipset from around 18 months ago? Or are you anticipating these solutions will be better implemented on this year’s hardware—or next year’s?

Peter Hodges — 00:13:46
We’ve had a development journey alongside Lukáš and Sumo Digital in general. The current devices are useful for exploration—but to make this a shippable solution, it’s really about future silicon, which we’ll be proud to show later this year.

Neil Barbour — 00:14:04
Okay. Lukáš, from a developer perspective: if you try to bring this to another high-end device on the market—maybe not Mali-based—what changes? Is it not possible on other devices in this class? Or is it specific to the Arm architecture?

Lukáš Medek — 00:14:26
The GPUs that the neural work is aimed for aren’t out yet. So we don’t have an opportunity to test it on the final device.

We’re testing the game on current GPUs on mobile devices. And the biggest issue we see is thermal throttling. Basically, you play for 5 or 10 minutes and the phone gets very warm. And then performance drops quickly, including battery drain.

So yes, we can run the game on current devices, but it’s not very convenient because of performance issues—thermal throttling and fast battery drain.

Neil Barbour — 00:15:30
Got it. So this is underlining more of the next generation of mobile gaming graphics processing than showing bleeding-edge readiness right now.

Peter Hodges — 00:15:39
Yes. It isn’t about flipping a single switch and enabling this on all devices anywhere. It’s about showing what can be achieved at the bleeding edge, and then watching how the market matures—both in the models and the efficiencies we continue to find, and in how neural technologies reach other device tiers.

Neil Barbour — 00:16:01
Peter, what has the reception been like for other developers in your orbit? Are they saying, “Wow, we want to sink our teeth in here and get cracking,” or how have those conversations gone?

Peter Hodges — 00:16:12
It’s been really encouraging, Neil—it’s been fantastic.

What’s been so great about Neural Dawn is that it gives partners something concrete to react to. This isn’t slideware. It’s a real game project—pressure-tested in the way I mentioned earlier, using real Unreal Engine content—showing practical trade-offs that exist in every game development cycle.

That’s what makes the story credible: quality, performance, tools, workflow. That’s what other developers are latching onto.

We’ve had great conversations with developers around the world about what they can do and what this means for us together. Interest is broad—we’re seeing it both from developers themselves and engine-facing teams.

And importantly, it’s moved the conversation forward: it’s not only “does the technology work?” It’s “how do I integrate this into my workflow?”

In that sense, Neural Dawn has been great for the flywheel we discussed earlier.

Neil Barbour — 00:17:17
Lukáš, as a developer, what are you thinking about as your next steps? Do you have games—or game ideas—in mind that would really benefit from this tech?

Lukáš Medek — 00:17:24
Yes, absolutely. And especially with the overlap that’s starting to become more standard—visual quality like PC or console, but in mobile games. That means we can make fewer compromises when developing for mobile, and aim for a higher standard of visual fidelity.

For example, with MegaLights and dynamic lighting, we can speed up development. Traditionally, mobile games often use what we call baked static lighting—where the lighting is precomputed into the assets, and lights aren’t actually used dynamically at runtime.

With a dynamic approach, in the Unreal Editor, what you see gets into the game immediately. That’s a huge advantage.

And it’s not only about games that benefit. This could touch other fields too—architecture visualization, automotive visualization—anywhere you need precise dynamic lighting.

So there’s a broad range of projects that can benefit.

Neil Barbour — 00:19:23
Yes, it does feel like you’re pulling me toward spatial computing in that framing.

Peter, we talked about bringing console-quality visuals into a mobile power budget. Could the same efficiency story go the other direction too—for laptops, handheld PCs, maybe even consoles—where power, thermals, and bill-of-materials costs are becoming major issues?

Peter Hodges — 00:19:51
I’m inclined to think it does. We already have proof points in PC gaming.

The same core idea is that these techniques can deliver reconstructed images—improved image quality and higher frame rates—along with more ambitious lighting that developers want, but within a practical performance budget.

When we scale this down, the way we think about constraints changes. We take into account the power constraints and memory constraints seen on mobile and on other Arm-based devices. We look at power, thermals, memory bandwidth, and silicon area.

And those considerations translate into benefits across the board—back to laptops, handhelds, tablets. And even further, into automotive systems and consoles. From Arm’s perspective, it’s about taking this broader picture—making efficiency the deciding factor, and making it practical for developers to keep expanding that flywheel.

Neil Barbour — 00:21:15
Yes. The big choke point for moving Arm more deeply into gaming has historically been x86 legacy compatibility issues. Maybe spend a moment talking about emulation or other layers to make gaming palatable across architectures.

Peter Hodges — 00:21:40
We can certainly look across the ecosystem and see places where instruction sets and APIs are emulated. You can find examples of that in hardware. So in that sense, this is already an unfolding story—one that will continue to develop.

Neil Barbour — 00:22:00
Lukáš, from a developer perspective: does technology like this change platform calculus? If mobile can support more premium-looking experiences, does that bring more developers to mobile—especially developers coming from console where user acquisition can be a challenge, but mobile already has billions of users?

Lukáš Medek — 00:22:21
From a business perspective I can’t really tell. But from an artist and developer perspective—yes, definitely.

With this, we can have much higher-quality visuals and visual fidelity, while keeping the budget low because the development process can be faster.

So yes, I think it can bring more developers to the platform—especially teams that have been afraid of being limited by performance issues and performance constraints.

Neural plugins and these kinds of high-end features widen who can participate in the market.

Neil Barbour — 00:23:28
Peter, let’s assume mission accomplished here. What’s the next big breakthrough from Arm? What should we look forward to from this “brain trust”?

Peter Hodges — 00:23:36
I look forward to so much.

Neural Dawn has given us tangible proof points—something really useful for showing content to developers around the world. It helps us all rationalize together about what the future of gaming could be.

I really want to see that flywheel effect. I want developers to be in a place where the constraints they traditionally fought—like “what is my mobile graphics budget?” or “what is my overall graphics budget versus my creative ambition?”—are less of a barrier.

Making it possible for more and more people to take advantage of these tools, and getting meaningful feedback that keeps driving our SDKs forward—improving engine integrations—and making sure people are set up to succeed—that’s what excites me. That’s what keeps me moving on the continuing mission.

My goal is to make neural graphics a normal, developer-accessible part of the game creation cycle.

Neil Barbour — 00:24:40
Luka, what’s next for you? What is Sumo doing next?

Lukáš Medek — 00:24:44
We’re focusing on finishing Neural Dawn. And I’m pretty sure it will end well, because even now the results are really promising.

And yes—being pioneers of new tech is amazing. We’re hoping for another project like that in the near future.

Neil Barbour — 00:25:10
Great stuff. Thanks, guys. Thanks so much for joining the podcast.

Lukáš Medek — 00:25:13
Thank you. Thank you so much. Bye.