Spec Analysis: Can Project Scarlett truly deliver Xbox's biggest generational leap?
What if raw performance isn't the game-changer this time?
After a pitch-perfect, well choreographed introduction to Xbox One X way back in 2016, hopes were high that Microsoft could repeat the trick for the crucial reveal of Project Scarlett at this year's E3. New details were indeed unveiled, major claims were made - but Microsoft muddied the waters somewhat with messaging that still leaves us unclear about what the new box is actually about, how powerful it is and what the vision is that separates it from Sony's upcoming PlayStation 5, built from the same technological building blocks.
Let's confirm what is definitely on the table - spec points that with one exception are point-for-point the same as Sony's recent next-gen announcement. For starters, both boxes use AMD's new Zen 2 CPU core - the basis of the Ryzen 3000 products arriving for PC users next month. Meanwhile, Microsoft also revealed that AMD's next-gen RDNA-powered Navi technology delivers the graphics muscle for Scarlett - just as it does for the next PlayStation. The platform holder also announced ray tracing support for its new machine, a feature that's also found in PS5 - though Sony has been reticent to explicitly confirm actual hardware acceleration for the job (our money is on the affirmative though).
Then there's the SSD - another hardware point that Sony revealed first, with Microsoft following suit. Another aspect that ties together both platform holders' next-gen announcements is a lack of information on the graphics core. On top of that, there are some curious factoids in Microsoft's reveal that do need some clarification. Scarlett was defined as the biggest generational leap in console technology that the firm has delivered - but I do think it hard to see the computational leap from OG Xbox to the Xbox 360 being bettered. Meanwhile, the 16x increase in RAM allocation seen moving from Xbox 360 to Xbox One is highly unlikely to be surpassed. Then there's the notion of Scarlett delivering a 4x leap in 'processing performance' over Xbox One X. On the CPU side, this does seem likely but the idea of the machine delivering an equivalent of 24 teraflops of GPU compute is unlikely.
Based on our own information, along with teasing reveals within the Scarlett announcement trailer, here are my thoughts on the set-up of the box - well, one of them at least. Curiously, Microsoft is using very strange PR-speak to avoid the question of the leaked lower-end box, codenamed Lockhart. Another interesting aspect is that while the Xbox One X reveal effectively told us the RAM allocation, SoC size, teraflop count and even clued us in on the cooler, Microsoft is being a lot more coy this time around and there may even be some red herrings in the assets this time around.
CPU: AMD Zen 2 architecture
AMD's Zen 2 technology has been confirmed for both Scarlett and PlayStation 5, but not a whole lot more. What we do know is that Sony has confirmed eight cores - a design foundation for the architecture - and thus it's likely that Microsoft will follow suit. The efficiency of the core has seen some fundamental improvements from its first-gen revision, but comparisons with the desktop PC versions seen so far can only go so far. It's unlikely that we'll get the 35MB of L3 cache announced for the new Ryzens, for example. The use of AMD's new scalable 'chiplet' architecture (where groups of eight cores can be clustered together for 16, or even 32 cores in total) also seems unlikely.
Our understanding from sources is that Microsoft is also turning its hand to a fair amount of custom CPU design for Scarlett. The move from Jaguar to Zen 2 represents an actual generational leap, one which Microsoft seem intent on exploiting. There's talk of 120fps gaming, which would certainly soak up the CPU cycles but on the flipside, it also doubles the graphics requirement - one of the areas where we should expect the next-gen consoles to fall a little short.
CPU-wise, there is one further takeaway. Some rumours had discussed a separate CPU and GPU within the new box, but Microsoft explicitly showed an integrated system-on-chip in its reveal trailer. Meanwhile, in a press briefing the day before the MS E3 briefing, AMD CEO Lisa Su confirmed that PlayStation 5 is also using a SoC, so the idea of discrete CPU and GPU components in either of the new consoles now seems highly unlikely. We do have some basic idea of area occupied by the Zen 2 cores within the 7nm processor - it's between 70 to 80mm2 - and this is useful information to have when considering the silicon budgets available to the platform holders in constructing their new processors.
GPU: AMD Navi and the RDNA architecture
Microsoft marketed Xbox One X based on its six teraflop GPU, with peak compute also used to differentiate the enhanced consoles from their 2013 base equivalents - but I genuinely believe that the teraflop was is over now. The GPU's potency in terms of gaming performance can't be measured by its peak compute output in a world where the new Navi-based 9.75 teraflop RX 5700 XT is - according to AMD - 14 per cent more capable than the 13.7 teraflop RX Vega 64. And remember, Vega itself is a significantly different beast to the earlier GCN architectures on which the current-gen consoles are based, with its own performance improvements.
With Navi, AMD takes the existing GCN compute unit and morphs it into the 'dual compute unit' - doubling up on ALU power while reducing latency and boosting cache. AMD is talking about a 25 per cent increase in IPC for Navi, which is equivalent to the boost delivered by a notional 80 CUs compared to 64. There is room to scale up further too, and AMD is expected to deliver a bigger, more powerful Navi GPU next year. All of which is to say: once compute figures do emerge, expect them to be lower than you might think, but rest assured that Sony and Microsoft can both still deliver great boxes.
Returning to silicon area, the new Navi GPUs from AMD deliver a total of 40 compute units requiring a remarkable 10.3 billion transistors. The nearest last-gen PC equivalent is the 36 CU Polaris chip used in RX 570, 580 and 590 with a total of 4.7 billion transistors. Going back to the first-gen GCN used in the PS4 and Xbox One, 32 compute units required 4.3 billion transistors. Clearly, Navi's compute units are significantly more complex and more capable - and transistor count is another indication of potential performance.
Memory: GDDR6 confirmed - but not much else
Microsoft's trailer presents the new SoC surrounded by GDDR6 memory, specifically 14gbps modules fabricated by Samsung. Beyond that, it's difficult to draw too many conclusions as there does seem to be some very odd properties in the memory set-up.
While the 2016 Xbox One X reveal actually gave away most of the technical make-up of the machine, Microsoft has been more coy this time around, with obscured test motherboard shots, an Xbox One X mobo fake-out (that seems to be the green PCB viewed overhead in the trailer) and lavish use of depth of field photography to blur out a lot of the interesting details on the all-important SoC shot.
However, drawing upon a 4K feed we captured direct at the Microsoft conference, two of the G6 modules come into focus enough for us to catch their precise part numbers, ending in 325BC-HC14 and 325BM-HC14. The 'HC-14' confirms 14gbps speed, but the two part numbers suggest a mixture of both 1GB and 2GB memory modules. Whether this is just Microsoft seeking to throw off people like me looking for clues on the make-up of the system remains to be seen, but this is certainly a strange set-up. Who knows, maybe we're looking at two 128-bit buses addressing 8GB and 4GB respectively? A straight 256-bit memory bus would deliver 448GB/s of bandwidth and if there is a 384-bit bus out of shot there, we'd be looking at a mammoth 672GB/s instead.
The potential power of Project Scarlett
We don't know the full extent of the make-up of Scarlett's memory system, but we do get a pretty decent look at the processor itself, which does invite some speculation. Some cunning ballpark guestimates (hat tip to Proelite on the Beyond3D forum) suggest that the Scarlett SoC is larger than Xbox One X's Scorpio Engine, with estimates ranging from 365 to around 380mm2, almost certainly based on the 7nm production process.
The new 7nm Navi GPU from AMD clocks in at 251mm2 for 40 of its new compute units, while educated guesses on the size of the eight-core Zen 2 CPU cluster come in at 70-80mm2. Historically, console SoCs have clocked in at between 320mm2 - so simply adding the two components as-is together and accommodating interconnectivity fabric and hardware accelerated ray tracing support already gives us a console-sized processor at the lower end of the scale, similar in size to PlayStation 4 Pro's. Realistically we'd be looking at 40 compute units with 36 active - all of the current-gen consoles have disabled components so that more of the chips from the production line are useable. At the upper end of the scale with a Scarlett SoC at ballpark 380mm2, a 48 CU graphics core with 44 active seems viable. This is guesswork though - and if Xbox One X taught us anything, it's that console developers often deliver a surprise or two in their final hardware.
In the case of both new consoles, clock speeds will prove vital - the AMD Gonzalo processor leaked with a mooted 1.8GHz, but to what extent that is actually viable remains unknown: the PC-based Navi RX 5700 XT has a sustained 'game clock' of 1755MHz but a peak clock of 1905MHz. Short of delivering a big box with a high-end cooler, it's challenging to imagine consoles matching - though to be fair, after Xbox One X's superb cooling performance, maybe we are due another big surprise in the upcoming generation.
|RDNA: Navi||GCN 4.0: Polaris||GCN 1.0: Tahiti|
|Process||TSMC 7nm||Global Foundries 14nmFF||TSMC 28nm|
And what should we make of the SSD? Sony has said that its solid-state technology is proprietary, and my understanding is that Scarlett's is also similarly bespoke, with low level access and terrific performance. Just don't go into this expecting standard M.2 slots on the motherboard with upgradeable storage - beating PC performance in a console is going to require a tightly integrated custom design. There's a reason why both Microsoft and Sony are pushing storage hard: it's a genuinely game-changing addition that will see the surrounding hardware pushed in all sorts of new directions. Low level integration may also mitigate the need for a typical 8x next-gen sized increase in RAM for the new consoles - the storage itself effectively becomes a secondary pool of memory, and there is some precedent for this: AMD has released a couple of high-end GPU products with SSD memory, allowing pro users to address vast data sets.
Wrapping up, we are left with the sense that Microsoft's pitch for the future of gaming, based on its hardware design at least, is very close to Sony's: the best that AMD can offer, paired with extremely fast storage which opens up a wealth of opportunity in using the surrounding architecture in groundbreaking ways.
The firm is also strongly advocating its back-compat credentials with support for all Xbox console generations - a promise we know it can deliver based on its impeccable track record. 120Hz support? Variable refresh? It's good to see these namechecked, but 8K? Well there's no reason why the reconstruction techniques of today can't scale up to higher resolutions tomorrow - but in truth, in the age of temporal super-sampling and other 'smart upscaling' technologies, I'm not sure that higher pixel counts are 100 per cent synonymous with a guarantee of better image quality any more.
Put simply, there's a better use for GPU power - especially if onboard solid-state storage can genuinely provide a vast pool of 'virtual memory' capable of delivering more detail, bigger, more contiguous worlds and more immersion. And with both Sony and Microsoft making the same bet on SSDs, it seems that this is almost certainly the way forward in delivering a next generation gaming experience.
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 £4.50. 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