A remarkable story unfolded last April where it seemed that an intrepid explorer of the 3DMark benchmark database stumbled upon preliminary testing for a new gaming processor from AMD dubbed 'Gonzalo' - almost certainly a codename for work-in-progress PlayStation 5 silicon. The notion of a PC benchmark database yielding top secret information about an upcoming next-gen console seems implausible - but further leaks over the last few days not only back up the Gonzalo story but also deliver new details about the graphics core of the new machine. On top of that, the leak also contains tantalising hints about the technical make-up of the Xbox Series X GPU too.
The scale and scope of this latest leak is remarkable and the origin of the new information seems even more far-fetched than the Gonzalo story, leading many to believe that the entire thing may be a work of fiction. However, having looked into the situation and independently verified the source, the overwhelming evidence is that the data does indeed originate from AMD - and hasn't been doctored. We're lacking crucial context for sure but the reasons to doubt the veracity of the leak are somewhat thin on the ground.
From what I can gather, someone at AMD's ASIC validation department used GitHub to store fragments of internal testing data from a range of work-in-progress Team Red projects. The leaks include testing of next-gen desktop and mobile Ryzen APUs along with some deep-dive testing on the PS5 chip, now codenamed Oberon. While the data is not public, it's clear that the GitHub test data has travelled far and wide: further details from the leak mentioned in this article are being discussed at length on ResetEra, for example. The genie is out of the bottle.
My understanding is that this data was first stored on GitHub around six to seven months ago - and looking back over noted leakers' timelines on Twitter, the source seems to have been picked up on as early as August. While this may suggest that the testing data doesn't reflect current next-gen console specs, it's important to remember that developing a microprocessor of the complexity we're talking about here tends to be a multi-year effort. Testing and validating a chip to ensure that it meets performance targets and that it passes debugging is in itself a lengthy process - and making changes to the architecture of the chip at this point is unlikely. Tweaks to clock speeds or accompanying memory are a possibility but the timeline we have suggests that Sony already took the decision to push GPU clock speeds higher by the time the leaked testing took place.
|PlayStation 5 (Unconfirmed)||PlayStation 4 Pro||PlayStation 4|
|CPU||Eight Zen 2 Cores with SMT - clocks undisclosed||Eight Jaguar cores clocked at 2.1GHz||Eight Jaguar cores clocked at 1.6GHz|
|GPU||36 custom Navi compute units at 2000MHz||36 custom GCN compute units at 911MHz||18 GCN compute units at 800MHz|
|Memory||GDDR6 at 448GB/s (Possibly 512GB/s) - capacity undisclosed||8GB GDDR5 at 218GB/s||8GB GDDR5 at 176GB/s|
The Gonzalo leak back in April suggested that PlayStation 5 would feature a Zen 2-based CPU cluster running at 3.2GHz paired with a Navi graphics core running at 1.8GHz. Slightly less concrete evidence linked to PCI Express identifiers suggested that AMD was referring to the GPU as 'Navi 10 Lite' - inferring heavily that the GPU would have the same 40 compute units as the PC-based Navi 10 part found in the RX 5700 and RX 5700 XT (with four CUs likely disabled for improved yields from the production line). The test leaks that emerged in recent days tell us nothing about the CPU component but confirm 36 available compute units running at 2.0GHz - which while unconfirmed would give us a 9.2 teraflop GPU for PlayStation 5. At this point, it would be remiss of me not to point out that performance from a Navi teraflop is much, much improved over older generation GCN equivalents.
The leak also suggests that PlayStation 5 uses GDDR6 memory - the same as the RX 5700-series Navi cards in the PC space. Rated bandwidth is 448GB/s but select tests seem to indicate that bandwidth could be as high as 512GB/s. This may indicate that 14gbps GDDR6 is upgraded to 16gbps on the same 256-bit memory interface or it may simply be the case that performance was improved on an internal cache. An upgrade to what is currently premium-level memory may well be beyond the scope of PlayStation 5's balance between price and performance - but of course, it's not as if Sony hasn't upgraded RAM in the past.
So how can we be confident that this processor is actually a semi-custom AMD product for Sony and not another partner like Microsoft? The giveaway is the fact that the GPU can be switched to three different modes in order to provide hardware backwards compatibility with PS4 and PS4 Pro. While a 2.0GHz GPU clock is used for what is described as the fully unlocked 'native' or 'Gen2' mode, the processor is also tested in what is referred to as Gen1 and Gen0 modes. The former is explicitly stated as running with 36 compute units, a 911MHz core clock, 218GB/s of memory bandwidth and 64 ROPs - the exact specifications of PlayStation 4 Pro. The latter Gen0 mode cuts the CU and ROP counts in half and runs at 800MHz, a match for the base PS4. The indications are that back-compat is an integral part of the silicon, which in turn raises some interesting questions about the makeup of the Navi GPU and the extent to which older GCN compatibility may be baked into the design.
|Xbox Series X (Unconfirmed)||Xbox One X||Xbox One/ Xbox One S|
|CPU||Eight Zen 2 Cores with SMT - clocks undisclosed||Eight Jaguar cores clocked at 2.3GHz||Eight Jaguar cores clocked at 1.75GHz|
|GPU||56 custom Navi compute units at approx 1700MHz||40 custom GCN compute units at 1172MHz||12 GCN compute units at 853MHz/914MHz (S)|
|Memory||GDDR6 at 560GB/s - capacity undisclosed||12GB GDDR5 at 326GB/s||8GB DDR3 at 68GB/s (Plus ESRAM)|
If the leak is remarkably specific about the PlayStation 5's basic GPU specification, where does this leave Xbox Series X? As we understand it, the documents consist of fragments from a much larger data set that we don't get to see. There is mention of a processor referred to as Arden, which is a highly likely candidate for Xbox Series X silicon. However, there is nowhere near as much testing done on this - or the data simply isn't up to date enough to include Arden validation testing of any note. Alternatively, the Sony and Microsoft semi-custom designs may be tested in a different way with a different set of criteria.
In reaction to the leak, there has been an argument suggesting that the PS5 specs are invalid because there is no mention of hardware accelerated ray tracing, whereas Arden has it confirmed (along with VRS - variable rate shading). However, the documentation for both processors is very different and can't be directly compared. A lot of AMD's validation testing for the PS5 'Oberon' processor is in the leak, whereas the Series X data is best described as somewhat patchy by comparison. If the PS5 specs are to be taken with a pinch of salt, have an armful of the stuff ready when looking at the table directly above with mooted Series X specs.
With caveats in place, the sparse data in the leak includes mention of 3584 shaders, which would translate to a frankly ginormous processor with 56 active compute units. There is no sign of clock-speeds in the data, but if we assume that 12 teraflops is the target, 1680MHz gives you 12TF on the nose, with a round 1700MHz delivering 12.2TF. If Microsoft was aiming lower than 12TF, the shader count would be much lower and the silicon itself significantly cheaper to produce - in fact, it would likely look closer to the PS5 configuration. The AMD leak does seem to confirm memory bandwidth for the Arden chip at 560GB/s. It's a curious figure, especially if we return to the Project Scarlett reveal teaser at E3 which seemed to show both 1GB and 2GB GDDR6 in play. Maybe we're looking at a hybrid memory interface with some modules using a 256-bit interface with others at 64-bit. This is something of a mystery that will hopefully be resolved when the official spec is shared by Microsoft.
On the face of it, the AMD leak confirms a number of spec points and infers several more. To begin with, it's clear that Microsoft designed Series X to push the limits of console design beyond the established norms in the pursuit of maximum performance. If the 56 compute unit makeup of the processor is confirmed, the cost implication is eye-opening to say the least. Based on what we've seen from the RX 5700 series, adding 50 per cent to the shader count and adding in an eight-core Zen 2 CPU cluster on top of that (along with other 'uncore' elements like the display controller and media engine) suggests a processor far larger than expected. This leak paints the Series X chip as even more of a beast than I expected, but with that in mind, just how expensive will it be?
Meanwhile, the PlayStation 5 spec outlined in the leak points toward a device with more of a balance between price and performance. Assuming we're looking at a not implausible 16GB of GDDR6 and a 1TB SSD, this is still an expensive-looking device - but stacked up against the monstrous Series X, it obviously stands more of a chance of hitting the magic $399 launch price-point that served PlayStation 4 and PS4 Pro so well. On the face of it, Microsoft has the more powerful machine, but some might say that in the console space, the price-point is of paramount importance.
However, I'm acutely aware that specs demand context. I'm still mindful of the conversation I had with PS4 system architect Mark Cerny prior to the launch of the PlayStation 4 Pro. He stressed the importance of customisation in processor design and while there are indications and givens we can take from existing PC Navi hardware in assessing this leak, Cerny and his team at SIE would be buoyed by the successes of the Pro's more modest and efficient design and mindful of its weaknesses.
And straight comparisons with existing Navi PC hardware can only go so far. The fact that hardware-accelerated ray tracing is confirmed for PS5 also demonstrates clearly that Sony has had the same access to future AMD roadmap features that it had with PS4 Pro. None of these features are the focus for testing in the leaked documents but that doesn't suggest that they are not there - particularly when Mark Cerny is saying that they are.
It's also worth re-emphasising that the leak is sparse on concrete details in terms of the next-gen Xbox - there are hints about the two processor designs for the Lockhart and Anaconda boxes, but nothing massively conclusive about what separates them. In fact, I understand that there are absolutely no details whatsoever for the cheaper Lockhart box beyond a potential codename for the processor (Sparkman - again, this is unconfirmed). In fact, what little we can glean about Series X from this leak suggests a top-end, premium-priced console that practically demands some kind of cheaper, mass market stablemate - and with that in mind, it's curious that nothing substantial about that chip has leaked.
Overall, this latest next-gen leak is fascinating and potentially fills in some blanks - but it's definitely limited in scope. The Xbox data raises more questions than it answers in several respects and while the PS5 data is more comprehensive, the bottom line is that all that we really have is more of a block diagram of one component of the processor bolstered by assumptions from PC-based Navi parts. And let's not forget that developers have achieved results from current-gen consoles that simply haven't been matched by Radeon-equivalent hardware. What we do know for sure is that both machines are targeting holiday 2020 for their release and if we look back at the timescales in how PlayStation 4 and Xbox One were revealed back in 2013, we shouldn't have too long to wait for more concrete information.
Will you support the Digital Foundry team?
Digital Foundry specialises in technical analysis of gaming hardware and software, using state-of-the-art capture systems and bespoke software to show you how well games and hardware run, visualising precisely what they're capable of. In order to show you what 4K gaming actually looks like we needed to build our own platform to supply high quality 4K video for offline viewing. So we did.
Our videos are multi-gigabyte files and we've chosen a high quality provider to ensure fast downloads. However, that bandwidth isn't free and so we charge a small monthly subscription fee of $5. We think it's a small price to pay for unlimited access to top-tier quality encodes of our content. Thank you.Support Digital Foundry