When I wrote this story, I based it on the Verge’s coverage and statements regarding the capabilities of the PS4 Pro. Eurogamer’s discussion on the same topic with Mark Cerny contains more information and much clearer quotes. Based on his statements, it seems clear that the GPU inside the PlayStation 4 Pro is based on Polaris, which is what we’d expected all along.
“Polaris is a very energy efficient GPU architecture that lets us boost the GPU power pretty dramatically, while keeping the console form-factor roughly the same. DCC – which is short for delta colour compression – is a roadmap feature that’s been improved for Polaris. It’s making its PlayStation debut in PS4 Pro,” Mark Cerny told Eurogamer,, confirming in the process that this feature was not implemented in the standard PS4 model.”
I apologize for any confusion.
Original Story Below:
Ever since news of the Sony PlayStation 4 Pro leaked, gamers have wanted to hear more about what the system could do and what kind of benefits it would offer over and above the standard PlayStation 4. Sony has shared some of those details and been shy on others. But with the launch fast approaching (and the need to build hype around the system), the chief architect of the PS4 and PS4 Pro, Mark Cerny, has been more willing to share details on the platform’s capabilities.
Cerny recently sat down with The Verge to talk about the new console, and some of what he had to say is quite a bit different from what we were expecting. For starters, take the GPU. General expectations around the Internet (including our own) have predicted a single unified Polaris-class GPU with 36 CUs. According to Cerny, that’s not what Sony built. Instead of a unified GPU core, Sony plunked a second GPU down right beside the first. It’s possible the GPU inside the new PS4 Pro isn’t actually new, and the implication is that this core isn’t based on Polaris at all. Alternatively, it’s also possible both GPUs are now based on Polaris — but the interview doesn’t present any data to suggest this is the case.
The PS4 Pro will also have more RAM available for running games — while the 8GB total system memory is unchanged, Sony has made another 512MB of it available to developers. Finally, the console will have 1GB of conventional RAM to serve as a cache for applications that aren’t currently in use. This may also help explain how Sony was able to squeeze another 512MB of RAM out of the PS4 — if applications like Netflix can be shoved into this 1GB DRAM buffer, it would leave more of the PS4’s GDDR5 main memory pool available for other tasks.
As we’ve previously discussed, the PS4 will use checkerboard rendering at 2160p to offer near-4K rendering in some cases, while some games will run in an 1800p mode “using similar techniques.” This implies that the PS4 is using a checkerboard pattern to simulate 3200×1800 and possibly upscaling the image thereafter, as opposed to rendering in native 1800p. Game developers will also have the option to render in 1080p and scale up to 4K using hardware built into the PS4 Pro.
As for existing titles, they’ll see only very marginal improvements on the PS4 Pro unless specifically patched for it. This is where Sony’s GPU strategy ensures backwards compatibility. By default, firing up a PS4 game on the PS4 Pro won’t take advantage of any of the new GPU hardware. Developers will have to patch their titles to add that functionality, and if they don’t, you won’t see performance boosts (though you may still get HDR support if you have a TV that can take advantage of it). The clock speed improvements in PS4 Pro hardware should still play a small part in improving visuals, but Sony is telling gamers not to expect much.
PSVR won’t automatically benefit from PS4 Pro
When Sony held its PlayStation 4 Pro launch event, we wondered why there was no mention of PSVR. That was surprising, given that the PS4 Pro promised far more graphics performance than its predecessor, and VR needs that boost. Again, however, these improvements won’t be baked into the underlying hardware. “We’re just asking for them to take advantage,” Cerny said. “We’d like for them [developers] to take a look at what the hardware can do and do something with it.”
In this case, we think developers probably will patch their games to take advantage of the PS4 Pro, since Sony’s VR push is so new and only a handful of games are currently available, but those gains aren’t automatic.
Other details: Cerny said there won’t be a 2TB SKU in the near future, the company left streaming out because it doesn’t see a future for 4K Blu-ray support, and it has no plans to introduce a PS4 Elite controller. All in all, it was a surprising discussion — and not in any way that favored Sony.
Did Sony play it too safe with the PS4 Pro?
When the first news of the PS4 Neo broke cover, it sparked extensive discussion of whether or not Sony was taking a radical chance, assuming that gamers would be willing to upgrade to new CPUs and GPUs as part of a mid-cycle transition, rather than the 5-7 year cycles we’ve observed previously. Now that Cerny has shared more about the platform, we’re left wondering if Sony has actually played things too safe. The most straightforward reading of the Verge interview suggests that the PS4 Pro relies on two GCN 1.0 GPUs. It won’t improve VR or game performance without specific patches, it won’t feature 2TB SKUs, Elite controllers, or other features that are available on Xbox, and its upgraded hardware won’t be nearly as much of a leap as the Xbox One Scorpio.
It’s also not clear how effectively the GPUs will work in tandem. We won’t assume AMD will have the frame-pacing issues here that it has had in Windows, because the operating systems and APIs are vastly different. But multi-GPU configurations typically also require more CPU overhead. This could practically limit how much performance Sony can squeeze out of the new system, and may explain part of why it doesn’t seem to be aggressively talking about pushing the performance envelope.
The PS4 has two CPU SoC blocks, but one unified GPU. If the new PS4 Pro has two GPUs, it’ll be interesting to see how those GPUs are connected within the SoC — does each CPU cluster handle one GPU?
We’ll be watching with interest to see how this competitive situation develops, but what Cerny shared here raises as many questions as it answered.