After letting Microsoft have free reign for almost three months, Sony and their chief architect Mark Cerny finally lifted the lid on what everyone wanted to know about the PlayStation 5: install times, backwards compatibility, how 3D audio works in practice, GPU and CPU speeds, ray-tracing, architecture and what that means for video games.
In a premiered video over YouTube, Mark Cerny stood in front of a quiet room, starting by saying “there’ll be plenty of opportunities later this year to talk about PlayStation 5 games”. Rather, the livestream was an opportunity to talk about the console’s architecture and what that meant for developers.
“About once every two years, I take a tour of the industry,” Cerny said, talking about spending “weeks on the road” as he visited more than 24 developers and publishers talking to them about the PlayStation console and ecosystem.
Cerny spoke about a custom engine for audio. It wasn’t a feature requested by developers, but something Sony was pushing for internally (and something that it’s shown off publicly before at CES).
The most requested feature from publishers and internally at Sony, however, was an SSD. “A lot of developer time was being designed around slow load speeds,” Cerny said. The PS4’s platter hard drive can take between 2 to 50 milliseconds seek time for a gigabyte of data, with a read speed between 50mb to 100mb/s depending on what edge the data is located, which works out to be an overall load time of about 20 seconds per 1GB.
“My rule of thumb is that the hard drive is spending two thirds of its time seeking, and only a third loading data,” Cerny said. And this is just for 1GB of data, which is bugger all these days.
The target for the PS5’s SSD hard drive, Cerny noted, was to hit at least 5GB/s bandwidth at a minimum. The load time for 2GB of data is 0.27 seconds, meaning players don’t have to deal with loading times, reloading after dying, and fast travel becomes “blink and you miss it” type of quick.
“We might even have to slow that transition down,” Cerny said, although he argued the SSD’s main advantage is the freedom it gives developers.
Another strategy Cerny mentioned was the ability to load all the data behind a player instantly, reducing the amount of data that has to be streamed. Install times “as you know them today” will be gone, and system memory will be more efficiently used with the PS5. The SSD itself is “in some ways more like RAM”, and that the console will ship with 16GB of GDDR6.
But extra abilities like reducing install times, streaming bottlenecks and data streaming issues aren’t all resolved with an NVMe-level SSD. There’s also bottlenecks in the existing PS4 architecture, which is why adding an SSD yourself to a PS4 doesn’t automatically resolve all these issues.
So Sony worked on a custom flash controller, using a 12 channel interface that meant the natural size of the SSD would be around 825GB. “The friction caused by reinstall or redownloads would be quite low, so we locked in on that 825GB size,” Cerny said, adding that Sony would allow gamers to add more storage if they wanted.
After his tour of studios in 2017, Sony adopted a new decompression standard called Kraken that was either already in use at every major studio, or about to be evaluated. So when building a custom I/O unit, which forms part of the Zen-based system-on-a-chip from AMD that sits on the same die as the GPU and CPU.
But what if users want more storage? Cerny mentioned that users would be able to support “certain M.2 SSDs” that could be installed in a bay on the PlayStation 5. The drive will have to be a PCIe 4.0 drive, however, and those commercial NVMe drive won’t have the advantages of the custom flash controller that Sony created.
Cerny noted that Sony would be doing “compatibility testing” with various M.2 drives, and they’ll supply a list of compatible M.2 drives that are not only fast enough, but are physically capable of fitting within the specified bay in the PS5. “Please hold off on getting that M.2 drive until you hear from us,” Cerny said, noting this list would most likely be published after the PS5’s launch.
As for the GPU, it’s a RDNA-2 custom design from AMD, Cerny stressed, adding that users should not use compute unit counts or teraflops as a measurement for the power of RDNA 2 GPUs against the PlayStation 4.
“The PlayStation 5 GPU is backwards compatible with the PS4,” Cerny said, noting that this was handled by including legacy modes for the base PS4 and PS4 Pro inside the PS5. “Once backwards compatibility is in the console, it’s in. It’s not as if a cost backdown will remove backwards compatibility,” Cerny said, noting it took AMD years to get backwards compatibility going.
Almost all of the top 100 games on PS4, sorted by playtime, are expected to be playable at launch on the PS5 using this boost mode, Cerny said. Sony has to do individual game testing to be sure for other games.
The PS5 uses what’s called an intersection engine – the same technique AMD will use on their upcoming GPUs for PC soon – to enable real-time ray-tracing on consoles. Developers have full reign to decide what amount of ray-tracing they want to use: they can invest hardware time into global illumination, ray-traced shadows and reflections, or focusing on ray-traced audio.
“I’ve already seen a PS5 title with ray-traced reflections in complex scenes with only modest cost,” Cerny said.
To make all of this possible, Sony have tried to push the speed of the GPU much faster. Cerny deployed another example how running fewer compute units at a higher frequency wasn’t necessarily comparable to more compute units at a slower speed, because of how the frequency of a GPU affects all parts of a game. The PS5 will have 36 CUs – although each CU is larger than their PS4 equivalent – and the GPU has a variable frequency. “We continually run the GPU and CPU in boost mode,” Cerny said.
What happens is that the console runs at a consistent power draw, and the frequency of the CPU and GPU will vary based on the load require. AMD’s SmartShift technology lets the console send any unused cycles from the CPU to the GPU, and with the variable boost strategy Sony ended up capping the frequency of the GPU at 2.23GHz. The CPU, meanwhile, spends “most of its time” at 3.5GHz, although not all games will run at these high speeds all the time.
Cerny mentioned the most “dramatic progress” with audio over the PS4’s lifecycle was the audio in the PSVR. The main goals for sound on the PS5 was to support “hundreds of advanced sound sources”, make sure audio was part of the console and an improved experience for all users in all environments, and to improve the locality and presence of audio of games.
Using Dead Space of all games, Cerny pointed out it was easier for users to deduce where a final enemy was if you were using headphones as opposed to real TV speakers. With the PS5, the goal is to provide more precise, accurate 3D audio.
The audio engine is called Tempest, and the Tempest Engine is a hardware unit inside the PS5 that’s very similar on the audio units that were inside the PS3. “The goal being to make possible near 100% utilisation,” Cerny said.
Cerny mentioned that using an existing system like Dolby Atmos wouldn’t have been sufficient, because it wouldn’t have supplied 3D audio for all, and Sony couldn’t be sure what sound hardware certain users would have, not to mention issues with proprietary tech. But to actually get 3D audio working for regular humans, Sony has created five presets based that you’ll pick from using a configuration tool.
Cerny’s presentation didn’t include a teardown of the console, any upcoming features from the PS5’s controller or upgrades to the PS5 system software or ecosystem, although those details are expected to come as we get closer to launch.
This post is being updated live.
Leave a Reply