Today, Sony released a new technical presentation from PlayStation architect Mark Cerny, in which he took an in-depth look at the PS5 Pro – particularly the system’s new AI upscaling technology, PlayStation Spectral Super Resolution. During the conversation, Cerny also announced that Sony and AMD will be entering into a strategic partnership called Project Amethyst to advance gaming graphics through machine learning – the technology behind Sony’s PSSR and AMD’s FSR.

Last month I had the opportunity to visit PlayStation headquarters and take a look at today’s presentation. I also sat down with Cerny to get answers to some of our lingering questions about the PS5 Pro, as well as more details about Project Amethyst – and how it might work not only for the future of gaming graphics on PlayStation, but also on Xbox and PC .

The following conversation has been lightly edited for length and clarity.

IGN: You’ve talked about how there’s a difference between RDNA 2 and RDNA 3, and that you wanted to stay with RDNA 2 because you didn’t want to create a lot of work for the developers when upgrading to RDNA 3.

Mark Cerny: So it’s 2.X, right? That’s why we asked what features you can adopt from RDNA 3 that don’t create a lot of work for developers. And so we took them.

My question is why is this a problem on console while on PC you can just jump in with PC games? New generation GPU and it just works.

Drivers in the console world tend to be very, very thin. This is considered one of the advantages of consoles – that you can take full advantage of all hardware features. But that means there’s more work as you move from generation to generation. But problems would also arise in the PC world. If you had a new card, you would either have to compile all of your shaders live, which of course can cause problems, or you would have to have a strategy to deploy all of those shaders compiled to the new GPU.

I love the term flopflation. Did you come up with this term? Or is this something from the marketing department?

Yes, I made that up. I’m not sure how much credit I want to take, but yeah, that’s what I came up with.

But how do you feel about the general teraflops arms race on this topic? I know there are many different factors that affect gaming performance, but then again, the general public wants to have a solid number to stick to. What do you think of it?

Actually, I don’t think that will happen in the PC world. I don’t think consumers in the PC world talk much about teraflops. It seems to only apply to consoles and only to the newer generations. As I’ve said many times, teraflops are not a good indicator of GPU performance.

Tell me more about the Amethyst partnership. What type of information is exchanged between companies? And how does this partnership differ from the existing relationship between SIE and AMD in terms of how you built the PS5 on their hardware and previous consoles? What’s new?

Therefore, I should first explain the nature of the collaboration. There are two goals we are working on with AMD. One of them is: better hardware architectures for machine learning. This isn’t about developing proprietary technology for PlayStation – the goal is to create something that can be used broadly across PC, console and cloud. The other collaboration is on these lightweight CNNs for game graphics. So you know what kind of things are used in the PSSR and what kind of things may be used in the future FSR.

So that is the comprehensive nature of the collaboration. What type of information is then exchanged between companies to achieve these goals?

We work directly together on both goals.

Does this mean we can expect the results of this collaboration to be reflected in future AMD hardware that isn’t necessarily PlayStation hardware?

Absolutely. This is not about developing proprietary technology or hardware for PlayStation.

To take it a step further, does this mean it could potentially be used for Xbox hardware?

It can be used by anyone who wants to use it.

Who initiated the partnership? Did Sony want to compete more with AMD? Or did they come to you and ask how we can learn from your experiences in gaming?

Well, they are long-term partners and it was very clear that we had similar goals.

Let’s get back to the PS5 Pro. I want to talk about some of the internals. You didn’t mention the CPU at all in the presentation today. Is there a difference in the CPU of the PS5 Pro from the base PS5?

There are some minor improvements throughout, and CPU clock speed is one of those improvements. If you want the CPU to run at a 10% higher clock speed, you can run it at a 10% higher clock speed.

And that is?

It would be 3.85 GHz.

Architecturally. It’s still running…

Zen 2. It’s the same Zen 2 CPU.

As for the AI ​​upscaler you use for PSSR – is that a separate piece of hardware or is it integrated into the GPU itself?

We needed hardware with this very high performance for machine learning. So we modified the shader core to make this possible. Specifically, what you’re talking about on the software side, there are 44 new machine learning instructions that take a freer approach to registering RAM access. In fact, you are using register RAM as RAM. And also implement the math required for the CNNs.

In other words, we’ve improved the GPU. But we didn’t add a tensor unit or anything like that.

You mentioned frame generation, but it sounded like that wasn’t the focus of the machine learning aspects of the PS5 Pro. Does the PS5 Pro have frame generation technology?

On the PS5 Pro we currently only have PSSR, i.e. super resolution. We’re very interested in what machine learning can do to improve game graphics. And I think it’s pretty obvious to everyone that frame generation, frame extrapolation and ray tracing noise reduction are also very interesting targets.

Ray tracing is significantly more power-intensive than traditional screening and traditional reflection and lighting techniques. Do you think that the gaming community’s interest in ray tracing, as you mentioned in the talk, is worth expanding further in this direction?

Ray tracing isn’t one thing, it’s many things. You can use ray tracing for audio polling – which is pretty inexpensive – or you can use ray tracing to improve your lighting, which is a little more expensive but not particularly expensive. You can do reflections, and since the reflections can be done at a lower resolution, this is also possible without breaking the bank. And then as you move on, you end up with path tracing, where your technology is essentially based on ray tracing, and unless you have incredibly powerful ray tracing hardware, you’re not going to get very far with that. So what we’re doing is providing tools to developers and giving them the opportunity to figure out where they want to be on that spectrum.

In the development of the PS5 Pro, which began about four years ago…Knowing what you know now about the progress you’ve made along the way and everything that’s happened in the four years since then, what are the most important ones Things you’d do Are you excited to get into the PS5 Pro on a technical level that you weren’t able to do and is now “next on your hit list”?

I can tell you, it was an incredible education developing the machine learning hardware for PlayStation 5 Pro. And instead of saying, “Oh, we would have done it differently,” I look at it as we now have a good understanding of how this works, and anything we do in the future has significantly more potential.

You also mentioned the prospect of building it yourself rather than buying or outsourcing the technology. Could you elaborate on the thought process there?

A very simple way to look at this is: do we take the next AMD technology roadmap or do we actually try to design the circuits ourselves – and we chose the latter. And we chose the latter because we really wanted to start working in this space ourselves. It was clear that the future would be very ML-driven. And in that, you know, the world is talking about LLMs and generative AI, but I’m really just looking at game graphics and the push we can get for game graphics. We wanted to work in this area on this basis.

More broadly, what do you think of the recent global fascination with AI as a buzzword?

Well, we live in very interesting times.

That’s for sure.

I have to say, these are very different topics. So you might see something like a smartphone that has AI capabilities or a laptop that has AI capabilities, and it may not be immediately clear how these features will benefit you as a consumer. But the console space is pleasantly simple, because there are some things that have already been proven to be of great use – if you only have sufficiently powerful hardware.

Looking ahead, you said console development was about a four-year process. Does this mean we can assume work has begun on the PS6?

We are not currently discussing PS6.

Can we expect this in about four years?

Same answer.