After months of rumors and speculation, Microsoft finally took the lid off its plans for the ninth console generation by providing our most detailed glimpse at the Xbox Series X to date. Microsoft gave us a close look at Xbox Series X hardware, as well as showcasing the console running games. There’s no need to prefix this statement with “ifs” and “buts” anymore: the Xbox Series X is the biggest thing to happen to console gaming in over a decade. Let’s take a closer look at the console, the official specs Microsoft shared, and the implications for ninth-gen video games.
Fab process: Enhanced TSMC 7nm
Microsoft stated that the Xbox Series X’s SoC will be fabbed on an enhanced version of TSMC’s 7nm process. Compared to the 16nm SoC powering the Xbox One X, 7nm allows for over times the transistor density. This means that far more processing and GPU power can be packed into the same die area, without radically increasing thermals or power consumption. This is born out by the fact that the Xbox Series X’s die size, at 360mm2, is actually very slightly smaller than the Xbox One X chip.
CPU: An 8-core, 16-thread Zen 2 processor
According to Microsoft, the Xbox Series X will be utilizing an 8-core, 16-thread processor based on AMD’s Zen 2 architecture. The CPU will be locked to 3.8 GHz when simultaneous multithreading is disabled and will operate at 3.6 GHz when SMT is enabled. In practice, this means CPU power that’s more or less on par with the Ryzen 7 3700X. This is a tremendous leap in processing capabilities compared to the Jaguar-based CPU in the Xbox One X. Microsoft claims a 4x performance uplift, and it’s easy to see where they got this number: clockspeeds have been nearly doubled, from a meagre 2.3 GHz on the Xbox Series X to as high as 3.8 GHz with SMT disabled. Meanwhile, massive gains to IPC and architectural efficiency deliver another boost.
The cumulative impact is a quadrupling of processing capabilities. This, perhaps more than any other Xbox Series X improvement, will play a major role in determining the direction that games take over the next 5-10 years. The eighth console generation was defined by its weak processors. GPGPU, or GPU compute, actually came to the fore as developers struggled to find a way in which to offload physics, AI, and other functions from CPUs that were too weak to handle them. This meant that, while in many cases eighth-gen visuals were markedly than seventh-gen games, the actual gameplay experience didn’t evolve much. For example, the combat-exploration loop in Assassin’s Creed Odyssey, and the depth of NPC interactions (barring the new conversation system) aren’t a generational leap over seventh-gen Assassin’s Creed titles the way that Odyssey’s graphics and world scale are. CPU bottlenecking was so bad that it even prevented less intensive titles—like Alien: Isolation—from running at 60 FPS on consoles, even though PCs with weaker GPUs could easily pull this off.
The radical increase in CPU capabilities is at the cornerstone of many of Microsoft’s stated aims with the Series X. With both 60 Hz and 120 Hz gaming on the cards, the only way to keep the Series X’s GPU fed, without massive bottlenecking, is with a faster processor. When it comes to features like Quick Resume that rely on on-the-fly compression and decompression, the CPU again plays a critical role in ensuring that other parts of the Series X hardware configuration, such as the high-speed NVMe SSD and GDDR6 are utilized to their fullest potential.
The other area that will see transformative change thanks to the Xbox Series X’s CPU is game design. As mentioned, eighth-gen games are generally at a similar level of complexity—in terms of world building, interactivity, and scale—as their seventh-gen counterparts. The Xbox Series X’s faster CPU will enable more complex, CPU-intensive mechanics to be implemented. We’ll see games that leverage the GPU and I/O capabilities of the Series X to deliver seamless LOD scaling, sans pop-up, all the way to the horizon. We’ll see game with advanced AI that allow for NPC interactions far beyond anything we’ve seen so far. Instead of hearing the same canned responses ad nauseam, NPCs could leverage voice assistant-like capabilities to dynamically respond to player interactions: towards the end of the generation, we might even see AI-generated voice overs implemented dynamically. In-game physics will routinely exceed the bar. In action titles, micro-destruction—allowing the player to chip away small parts of the environment—will become commonplace and won’t need to be faked via normal maps and parallax occlusion mapping. In-game objects, and the ways in which the player and NPCs interact with them will also become more complex. This will mean that RPG environments no longer feel like set-pieces, but more like real-world spaces.
GPU: Custom 52 CU RDNA2 with hardware ray-tracing
Microsoft already offered us a glimpse of the Xbox Series X’s GPU capabilities, both through the Hellblade 2 reveal, and by casually throwing around the 12 TFLOP figure. Because of the increase in compute capabilities and the move to a new architecture, there was no real doubt that the Series X’s GPU would offer a substantial improvement over the Xbox One X. What’s remarkable is just how Microsoft got it there. The Series X’s custom GPU is the closest we’ve gotten to AMD’s elusive “Big Navi” in the flesh. As a 56 CU part running at a locked 1800 MHz, this is a generational replacement, not for the RX 5700 XT, but for the big, hard-hitters like Vega 56 and the Radeon Fury. Flagship hardware invariably costs a lot. However, Microsoft has rationalized costs somewhat by using a salvage part with 4 CUs disabled in the Series X.
The trade-off: less than 10 percent of the GPU’s performance for a substantial reduction in cost makes a lot of sense. It’s almost certainly how Microsoft managed to get a 56 CU part in a console. Exactly how fast is the Xbox Series X’s GPU, though? If we consider RDNA2 to be an iteration of RDNA—meaning no radical increases to IPC, it’s possible to get an idea of capabilities by comparing the Xbox Series X GPU with Navi 10. The RX 5700 XT is a 9.7 TFLOP part with 40 CUs, meaning 2560 shader cores. It’s roughly 10 percent off-pace compared to the 1080 Ti, Nvidia’s previous-gen 4K flagship, and about 10-15 percent slower than the RTX 2070 Super and RTX 2080 respectively. At 12.25 TFLOPs, the Xbox Series X’s GPU delivers 26 percent greater compute performance than the 5700 XT. This would put its somewhere in between the RTX 2070 Super and the RTX 2080 Ti, making it, as of now, the second-fastest GPU available on any platform. With Nvidia’s Ampere cards and AMD’s PC-centric RDNA2 lineup slated to arrive in a couple months, the Xbox Series X’s capabilities vis a vis the best of PC might change a little.
Nevertheless, we’re looking at an extremely powerful GPU here that’s entirely capable of powering 4K experiences with little to no compromise. The real game-changer, though, is support for hardware-accelerated ray-tracing. According to Microsoft, the Series X will be able to tap into the equivalent of 25 TFLOPs of compute power when handling a ray-traced workload thanks to its dedicated ray tracing hardware. This puts the console’s ray-tracing capabilities into perspective: hardware acceleration allows nearly the same ray-traced capabilities as a software implementation that completely utilizes the Series X’s compute horsepower. With ray-tracing incurring such massive performance penalties on PC right now, this is great news: it could mean native 4K experience with ray-tracing, and sans upscaling compromises like DLSS.
While the CPU upgrade is set to play a big role in enhancing the in-game experience, a key highlight of the Xbox Series X’s GPU—its support for faster low-precision integer operations—up to 97 TOPs of 4-bit compute—is an ace up the sleeve when it comes to AI and machine learning workloads. AI and machine learning workloads are unique because, unlike rendering, they don’t need to use high-precision 16-bit and 32-bit math to work. By accelerating low-precision integer op performance by a factor of 8 times, Microsoft has ensured that the Series X will be able to run advanced ML and AI workloads in the years to come on the GPU, with a minimal performance impact.
Storage and Memory
These are areas where we haven’t seen a major console evolution in over a decade. The PlayStation 3 introduced Blu-ray to the console space back in 2006. It was, as a matter of fact, one of the cheapest Blu-ray players on the market for a considerable amount of time. Apart from this innovation in optical storage, there have been no major changes to storage technology since the seventh gen. The Xbox 360, Xbox One, and Xbox One S all use 5400 RPM hard drives by default. The memory pool did see a considerable leap between the seventh and eight generations—from 512 MB of RAM on the Xbox 360 to 8 GB on the Xbox One, and again to 12 GB on the Xbox One X. However, the general situation—slow storage paired with relatively slow memory pools—didn’t change much over the years. This had a profound impact on how games were designed. Data loaded from the hard drive to memory at a snails pace. And SoC access to memory wasn’t exactly fast either. Assets needed to load in time, and in open world games, especially, this meant aggressive LOD transitions, lower resolution textures across the board, and a loading screens everywhere.
The Xbox Series X’s combination of a high-speed 2400 GB/s NVMe SSD and a 16 GB pool of GDDR6 memory delivering over 560 GB/s of bandwidth, is set to have a transformational impact on loading. Microsoft elaborated on its concept of “extended memory” for the Series X. Essentially, its SSD storage is fast enough to be able to act as an extension of RAM—the way page filing works—without a perceptible hit to performance, by dynamically compressing and decompressing data on the fly and loading only what needs to be on the RAM at a given point of time. This means that loading screens, will either be reduced or completely eliminated. Even in games that have loading screens, Microsoft’s Quick Resume functionality—where a copy of a game’s assets on RAM is cached to the SSD—means that startup loading will become a thing of the past. Quick resume means that, from the game’s perspective, the user never quit. This makes it possible to jump back into the action at exactly the point of time a user closes a game.
Together with the vastly improved CPU and GPU capabilities, the Xbox Series X’s storage and memory upgrades mean that open world games will look and play better than ever before. We’ll see games with environments that put most games to shame without a single load screen and with assets that load in so fast that LOD swapping isn’t even perceivable.
Microsoft’s decision to completely take the wraps off the Xbox Series X this early in the year is a bit of a surprise. It’s a very welcome one, however. After an entire generation characterized by underperforming consoles, the sheer power on tap with the Xbox Series X is refreshing. With consoles as powerful as the Xbox Series X, our expectations are sky high right now.