Cerny Computer Entertainment
Digital Foundry face to face with PS4 chief architect Mark Cerny.
The arrival of the next generation in console systems is hugely exciting for Digital Foundry. Sony and Microsoft are effectively pressing a giant button that resets everything we know about consoles and, for us, the race is on to find out exactly what's inside the new hardware, what it's truly capable of, how it compares with the current state-of-the-art in gaming technology, and how the platform holders have customised their designs to give their machines the edge over the competition.
Over the last year, we've seen leaks of unprecedented authenticity, we've chased down numerous developer hints, seen prototype devkits put up for sale on eBay, and amidst some remarkable fantasy nonsense, even reader tip-offs have translated into big stories once doubled-sourced with developer contacts. We revealed the near-final PS4 specs prior to February's reveal event, and have followed that up with further next-gen exclusives, but the thirst for hard data on the performance of the new consoles is unquenchable. We will always want to know more. So, what would happen if we had the opportunity to talk face-to-face with the man who holds all the answers about PlayStation 4?
We found out this week, when we had the chance to meet up with chief architect Mark Cerny at the Develop conference in Brighton. This was an opportunity to sit down and chat with the man who not only knows the final hardware design inside out, but who would also be able to talk at length about the entire developmental process - the thinking behind the hardware, the decisions behind the balance of the components, and the alternative designs that never came to be. This is also the man best-placed to tell us how Sony managed to cram 8GB of GDDR5 into the PlayStation 4, in what is believed to be a last-minute yet massive addition to the spec of the new console.
Despite a 23-year career writing about the games business, it's fair to say that I went into this meeting with a mixture of excitement, apprehension and nervousness, but as the interview began to unfold those sensations began to fade, as it became clear that there would be no grand revelations. Acutely aware that he was on the record - with all that represents - Cerny was guarded in some areas, and in terms of the 8GB GDDR5 discussion in particular, we can't help but feel that the true, remarkable story there remains untold, despite his claims to the contrary.
Regardless, what remains is a fascinating insight into the processes and thinking behind the PlayStation 4, the vision behind the architecture and some of the latent potential of GPU compute - something Cerny identifies as an area that may only start reaching anything like its true potential several years into the lifecycle of the hardware. And bearing in mind that Sony carried out a post-mortem on PS3 a year after launch, and actively started development on its successor way back in 2008, we couldn't help but wonder - come 2015, will Mark Cerny return to develop PlayStation 5?
"Well, Ken [Kutaragi] is a genius, it was not possible to duplicate his approach. We had to find a different way of making console hardware."
Well, Ken is a genius, it was not possible to duplicate his approach. We had to find a different way of making console hardware.
In 2007 we started to do a post-mortem on PlayStation 3 to figure out what worked and what didn't and PlayStation 4 didn't really get kicked off until 2008. So at the time that I had my epiphany - that perhaps I had the right skillset for the job - there wasn't much active development.
I'm outside of Sony, so I work with whatever organisation that Sony has in place. I made my pitch just as they were starting to get going.
I think that right now is the best time in the history of gaming to be a gamer. And if you look at it in terms of percentage, you conclude immediately that consoles are small. But that's a consequence of so many more people playing games than have ever played games before. Thanks to tablets and thanks to smartphones, we have this huge audience that now enjoys games. As for what's bringing them to console, that heavy content I was talking about - Skyrim and Assassin's Creed - is of course key to the appeal and I think on top of that we're now going to have the richness of the smaller, nimbler indie titles that are coming in.
What I was looking at on my holiday was just the instruction set of the CPUs and it really seemed to me that things had been changing recently and maybe we were on the verge of a transformation in the x86 architecture and how it would be used. That didn't mean that we would necessarily use an x86 architecture - all that happened as a result of that was that I wondered if we could add it to our list of options. Then, when our hardware project kicked off for real in 2008, I and the ICE team and a number of other technology people within Sony did a very detailed analysis of x86 and how code would work on that versus other processors.
So we looked at the kinds of things we'd been doing on SPUs, we looked at general purpose code. We looked at writing in assembly for that platform versus others. We looked at compiled code and did about 15 different presentations showing off various aspects of the x86 architecture, and our first big meeting on PlayStation 4 was with the first-party teams. We spent most of the day going through our learnings on x86, we'd come to the conclusion that it was useful in a console but we had to ask them if they thought it could be useful in a console.
"The accessibility's had a big pay-off. There are 140 titles in development for PlayStation 4 right now and we have the strongest launch line-up in the history of Sony Computer Entertainment."
This probably doesn't quite answer your question, but there are so many issues involved in working with a vendor. The business relationship is very important. The timelines are very important, because you might be working with the brightest people in the business but if their product doesn't come out in the specific year that you need it, you can't work with them. There were a tremendous number of considerations and the choice of AMD came out of that.
I just chose that story because that's the clearest way I could find in showing our new philosophy. By the PS3 era of thinking, we would definitely have gone with that design. But with the new way of thinking, we wanted to be sure that accessibility was there. We went with this combination of year-one accessibility and then a very interesting feature-set which... there's a Gamasutra article where I went through the major points... for year three or year four. The accessibility's had a big pay-off. There are 140 titles in development for PlayStation 4 right now and we have the strongest launch line-up in the history of Sony Computer Entertainment.
Consoles hit above their weight when you look at the price/performance. There's a few reasons for that. One is that the software layer in-between the game creators and the hardware is much thinner than it is on PC. Another reason is that because the hardware doesn't change, you can focus in just on that one particular architecture and learn a lot about it and then the games over the years... You'll see the benefits in terms of the graphics and the quality of the world simulation and the like. On top of that we did significant customisation of the GPU to make sure that there was that extra target for people to go after in 2016/2017. Put that all together, I think the future's looking very bright for PlayStation 4.
I think of it a super-charged PC architecture, and that's because we have gone in and altered it in a number of ways to make it better for gaming. We have unified memory which certainly makes creating a game easier - that was the number one feature requested by the games companies. Because of that you don't have to worry about splitting your programmatical assets from your graphical assets because they are never in the ratios that the hardware designers chose for the memory. And then for the GPU we went in and made sure if would work better for asynchronous fine-grain compute because I believe that with regards to the GPU, we'll get in a couple of years into the hardware cycle and it'll be used for a lot more than graphics.
Now when I say that many people say, "but we want the best possible graphics". It turns out that they're not incompatible. If you look at how the GPU and its various sub-components are utilised throughout the frame, there are many portions throughout the frame - for example during the rendering of opaque shadowmaps - that the bulk of the GPU is unused. And so if you're doing compute for collision detection, physics or ray-casting for audio during those times you're not really affecting the graphics. You're utilising portions of the GPU that at that instant are otherwise under-utilised. And if you look through the frame you can see that depending on what phase it is, what portion is really available to use for compute.
"The point is the hardware is intentionally not 100 per cent round. It has a little bit more ALU in it than it would if you were thinking strictly about graphics."
Knack is a small title. Knack will not be leading the way in terms of utilisation. We'll be looking much more towards teams that have done that in the past like Ubisoft with Assassin's Creed and Watch_Dogs, Naughty Dog with their next titles or Sony Santa Monica with theirs. I think Killzone is going to be doing that. All we are doing ultimately is setting it up so that developers can take it any way that they want go. So if increased visual fidelity is what is most interesting to that team, that is what they can focus on. We just wanted to be sure that they could also dig in on GPGPU [general purpose computing on GPU] as well.
Well, it is true that performance per watt is quite good for Jaguar - that has been helping us out in general. We knew that our design space would accommodate either four or eight so we proceeded accordingly.
To be honest, when we asked people, we heard absolutely every answer you could think of. One developer even told me their technology could accommodate a thousand cores! But for us it was very important to find out what the mainstream would be, what the overwhelming majority of the games industry would need.
That comes from a leak and is not any form of formal evangelisation. The point is the hardware is intentionally not 100 per cent round. It has a little bit more ALU in it than it would if you were thinking strictly about graphics. As a result of that you have an opportunity, you could say an incentivisation, to use that ALU for GPGPU.
Such a toolchain does exist. It's AMD's HSA [Heterogeneous System Architecture]. That's very exciting but our current strategies are about exposing the low-level aspects of the GPU to a higher-level language. We think that's where the greatest benefit is in year one.
There's dedicated audio hardware. The principal thing that it does is that it compresses and decompresses audio streams, various formats. So some of that is for the games - you'll have many, many audio streams in MP3 or another format and the hardware will take care of that for you. Or, on the system side for example, audio chat - the compression and decompression of that.
"Latency in GDDR5 isn't particularly higher than the latency in DDR3. Also, GPUs are designed to be extraordinarily latency tolerant so I can't imagine that being much of a factor."
So the GPU is a natural place to do a number of different types of audio processing. It really does come down though to the amount of parallelisation that is natural to perform for that algorithm, and that does vary greatly depending on what you are doing specifically in your audio processing. I think that as you go forward we will see a hybrid approach in a couple of years where certain aspects of the audio are being done on GPU.
Actually the audio ray-casting I was talking about it much simpler. We'll cast rays to see if you can shoot something. You really need to know because it affects the reticule. It'll be a different colour if you can shoot an enemy or not shoot an enemy. For audio, you need to check whether you can hear a source or not. Many games cast a couple of hundred rays. It's actually quite time-consuming to do those calculations and it turns out to be a pretty natural fit for the GPU. You can take your couple of hundred rays, have them processed in bundles of 64 on the GPU and then the game designers get to make up new and more creative ways to use that ray-casting.
That was really a case where our developer-driven process worked. So we received feedback, we listened to the feedback, we altered the hardware as a result. As far as how late it was in the process, actually what you're seeing there is just the developers were very, very good about keeping information confidential. As near as I can tell that all came from one rogue hacker, who wasn't even disclosed by us, how managed to hack into a developer and extract both the Microsoft and Sony documentation.
Latency in GDDR5 isn't particularly higher than the latency in DDR3. Also, GPUs are designed to be extraordinarily latency tolerant so I can't imagine that being much of a factor.
Interestingly, the target bandwidth required depends on the length of the game. The longer the game is, the less bandwidth you'll need to be loading in the later stages of the game. There's another factor there as well. Are you playing continuously as a download or do you play for a few hours and then put the system on standby for the rest of the evening where it can continue to download and then resume the next day?
Certainly the broadband infrastructure - not just for the user, but for the country - is relevant when you are designing systems. Our focus has been primarily on coming up with the over-arching concept that will be compelling to users.
[Laughs] Can you just put "Mark laughs" in there?
Update: A couple of hours after publication, Mark requested some minor grammatical changes made to his responses, which we've incorporated into the piece.