For the past several months both Microsoft and Sony have been slowly but steadily trickling out additional details about their forthcoming gaming consoles. And now this morning we’re getting our next bit of information from Microsoft, who has released a few more nuggets of information on their forthcoming Xbox Series X console.

When the console was first formally announced at the end of 2019, the company revealed that it would be using AMD’s Zen 2 CPU cores, but they were a bit cagier about the GPU specifications. Now the company has opened the door just a bit more on those, giving us some performance and feature information – and by and large confirming earlier theories about what the hardware would entail.

First and foremost, Microsoft is now confirming that the console’s APU is using AMD’s RDNA 2 architecture for the integrated GPU. Information about this architecture is still limited, but AMD previously disclosed that RDNA 2 would include hardware ray tracing functionality – something not present in RDNA (1) – and Microsoft in turn will be tapping this for their next game console. Microsoft, of course, already has significant experience with hardware ray tracing thanks to DirectX’s own ray tracing functionality (DXR), so the company will be able to hit the ground running here, albeit with AMD hardware for the first time.

Microsoft’s announcement also confirms for the first time that we’re getting Variable Rate Shading (VRS) support. This is another feature that has been supported in DirectX for a bit now (and in rivals Intel & NVIDIA’s GPUs), but isn’t currently available in AMD’s RDNA (1) lineup. A sampling optimization of sorts, variable rate shading allows for the shading rate for an area of pixels to be increased or decreased from the normal 1:1 ratio. The net impact is that an area can be oversampled to produce finer details, or undersampled to conserve resources. As the former is more of a niche use case for VR, we’re far more likely to see undersampling in day-to-day usage. Especially with complex pixel shaders, when used correctly VRS is intended to give developers a way to improve the performance of their games for little-to-no perceptible impact on image quality.


VRS: Visually Represented (Image Courtesy NVIDIA)

Finally, as far as overall GPU performance is concerned, Microsoft’s latest revelation finally gives us a performance estimate: 12 TFLOPs. While the company doesn’t break this down into clockspeed versus compute units, this is none the less twice the GPU performance of the Xbox One X. Or for a more generational comparison, more than 9x the GPU performance of the original Xbox One.

Even at just 2x the performance of the Xbox One X, this is by all objective measures quite a bit of GPU horsepower. To put things in perspective, AMD’s current fastest RDNA-based video card, the Radeon RX 5700 XT, only offers 10 TFLOPs of GPU performance. So the Xbox Series X, a device with an integrated GPU, is slated to offer more graphics performance than AMD’s current flagship video card. Which, to be sure, doesn’t mean the Xbox Series X is going to be more powerful than a PC (there’s no getting around the fact that AMD has been trailing NVIDIA here), but it’s clear that Microsoft has great ambitions for the console’s graphics performance.

Tangential to this, Microsoft has also finally confirmed that the console will support HDMI 2.1. This has long since been a given, as the time frame and Microsoft’s own resolution goals pretty much require HDMI 2.1 to begin with, but none the less we finally have confirmation. The company’s announcement also confirms that along with variable refresh rate support (first introduced on the Xbox One X), the console’s HDMI connection will also support HDMI’s auto low latency mode, which tells a display to switch to low latency mode. Depending on just how fine-grained Microsoft’s implementation is, there’s room here for nuance; for example only engaging low latency mode for gaming, but leaving it off when watching videos so that a display can apply extra processing.

Finally, while Microsoft had previously disclosed that the console would use a “next generation” SSD, it’s interesting to note that the company is now calling it a “custom built” SSD. Absent more details, I’d hesitate to read too much into this, but at a minimum it means Microsoft is not using an entirely off-the-shelf SSD. Whether that means they’re using commercial silicon with different firmware, or ordering their own silicon entirely, remains to be seen. And perhaps the bigger question is whether this is an all-flash setup, or if the console will be running some kind of tiered storage with an SSD and an HDD? Given that even when SSD prices were at their historical lows, a large enough SSD to hold several AAA games could easily run for $100 or more, a pure SSD setup stands to be an expensive venture.

But whatever the storage architecture is, it sounds like Microsoft is putting it to good use. On top of the previously mentioned loading benefits, the company is touting a feature they’re calling “Quick Resume”, which allows for several games to be suspended at once. Since suspending multiple games in this fashion all but requires evicting them from RAM and sending them to non-volatile storage, Microsoft will need a high performance (and reasonably spacious) SSD to power this feature.

As always, expect to hear a lot more about the Xbox Series X over the next several months, as Microsoft ramps up to launch it for Holiday 2020.

Source: Microsoft

Comments Locked

74 Comments

View All Comments

  • c1979h4life - Monday, February 24, 2020 - link

    That's the rumor. My hope is that they let you run a version of Windows, if your a subscriber to Game Pass Ultimate.
  • FreckledTrout - Monday, February 24, 2020 - link

    The XBOX will have 20% more tflops than the 5700XT? Whoa the power efficiency of RDNA2 must be vastly better. Pretty impressive.
  • DanNeely - Monday, February 24, 2020 - link

    Spending a big chunk of extra transistors on ray tracing most likely.
  • blckgrffn - Monday, February 24, 2020 - link

    They are using a huge chunk of silicon and likely using much lower clockspeeds. Check out the 5600 XT review with the lower clock speed bios. Given the power supply on the Series X is going to be ~300W and we can figure the CPU of for ~50W, with the next rev of TSMC silicon and a decent power budget this all seems reasonable.
  • haukionkannel - Monday, February 24, 2020 - link

    Interesting to see how big feal of that tflop power is from raytrasing hardware. It is quite possible that 5700
  • carcakes - Monday, February 24, 2020 - link

    Another triple socket 14 tb ddr4 vs 24tb optane vs 8 GB hbm2 ddr5 hbm3 -_-!
  • Santoval - Monday, February 24, 2020 - link

    12 TFLOPs of FP32 performance from an *integrated* iGPU? That's frankly insane! No matter how efficient RDNA2 is the die must be quite large, since it will also have plenty of RT (and tensor?) cores for ray-tracing. Thus it will probably get pretty hot. I wonder what TDP they managed to reach.

    They must have downclocked and downvoltaged the 8 Zen 2 cores to keep thermals in non insane levels. They might also get extra margins if this is fabbed at TSMC's 7nm+ node instead of 7nm. As for the "custom SSD" I hope the only part of it that's custom is not going to be a proprietary connector that enforces buying new or larger SSDs only from Microsoft...
  • Billy Tallis - Monday, February 24, 2020 - link

    A proprietary connector on the SSD wouldn't be too bad. That's pretty easy to work around with adapters or third-party SSDs that adopt the same form factor, as happened with Apple's SSDs before they moved the controller onto the T2 chip.

    What would really limit third-party options is if Microsoft's custom SSD requires non-standard firmware providing features that aren't normally found on consumer NVMe drives. Eg. if they're working with Enmotus, or if they're using enterprise features like multiple namespaces, then it would be a lot harder for other companies to start producing aftermarket upgrades.
  • c1979h4life - Monday, February 24, 2020 - link

    I would be okay with a thunderbolt 2-4 port. 2 TB ssds will take another 2-3
    years to drop below 180, due to this shortage and new phones. while 14 tb cost about 200 bucks. These games take over 60+ GBs. If they can find a way to load up what you need into the onboard ssd, ala streaming data in the background. I am good with 14tb usb C drive. A 4TB drive is more than a PS4 and Xbox one S combined.
  • lilkwarrior - Tuesday, February 25, 2020 - link

    You would not want to use mechanical HDs with these next-gen games + the games will finally unapologetically assume most are using SSDs just like how game developers can now unapologetically use WDDM 2.0 & deep-learning (not just for ray-tracing) in next-gen games

Log in

Don't have an account? Sign up now