Xbox Series X GPU may be more powerful than NVIDIA RTX 2080 Super (update)

Xbox Series X Reveal
Xbox Series X Reveal (Image credit: Microsoft)

What you need to know

  • Microsoft revealed that the Xbox Series X GPU features 12 teraflops of power.
  • This puts it ahead of cards like the NVIDIA GeForce RTX 2080 Super.
  • It uses AMD's RDNA 2.0 technology which should give it an even greater advantage.
  • Console games are usually more optimized than PC ports so that should also be considered.

Updated March 16, 2020: Digital Foundry confirmed that early benchmarks are similar to NVIDIA GeForce RTX 2080.

Yesterday, Microsoft confirmed that the Xbox Series X featured 12 teraflops (tflops) of graphics performance on the dot. The rumors of it being 10.8 or 11.3, or some other tflops, proved to be inaccurate. Naturally, when a high figure like 12 tflops is announced, everyone wants to compare it to existing chips on the market, but that comparison may not be that simple.

The Xbox Series X GPU uses AMD's RDNA 2.0 technology, which isn't available in any PC graphics card at the moment. This gives it the potential to surpass other NVIDIA cards in real-life use. However, even if we analyze the 12 tflops figure, it puts the Xbox Series X over the GeForce RTX 2080 Super which stands at 11.3 tflops. It's only beaten by the GeForce RTX 2080 Ti at 13.45 tflops and the Radeon RX Vega 64, at 12.66 tflops. However, the Radeon RX Vega 64 is AMD's last-generation chip so the advantages of Navi should outweigh the 0.66 tflops difference.

We'll only be able to compare the performance once the Xbox Series X lands, but it definitely seems promising. You also have to remember that generally, console games are more optimized than PC ports, so that also leads to some skewed results. Hopefully, Sony will reveal more details about the PlayStation 5 so everyone can make an informed decision come Holiday 2020.

Asher Madan handles gaming news for Windows Central. Before joining Windows Central in 2017, Asher worked for a number of different gaming outlets. He has a background in medical science and is passionate about all forms of entertainment, cooking, and antiquing.

  • TFLops is not the end-all, be-all.
    You can shoe-horn a fully-blown Hemi-supercharged V8 into a Volkswagon Beetle (I've seen one) but it will drive poorly on the track.
    Same thing with the Silicon. It may have the raw TFlops, but if the drivers and architecture it is put into are not tuned for it, it will be slower than older designs.
    Fortunately, MS owns the IP for the new APU, and they can tune the hell out of the drivers and sub-systems to support that kind of power.
    My only question is; How much of the heavy-lifting will be done in the Silicon, and how much will be done in the CPU via software? You can do Software Ray-Tracing right now on any x64 CPU, but it sucks.
    So, that is where "the rubber hits the road". In the software stack for the drivers for that particular hardware abstraction layer, and in the VM drivers for the actual XBseX VM.
    I trust MS to get it right, maybe not the first iteration of it, but they will in later software updates.
    Nvidia does a lot in their Tensor Cores with things like DLSS and defered-rendering and I am not sure this AMD architecture will have equivalent functions in hardware since the DirectRTX interface supports it in software too.
    We shall see............
  • Yeah, it'd be good to know how many Ray Tracing Cores this thing has, and the memory bandwidth. But I'm more excited about the HUGE jump in CPU performance (the CPU bump is far and away greater than the GPU bump), that's going to make a massive difference.
  • I think I realized last night that swapping multiple games around instantly is something I am most looking forward to. I have 3 games I am playing short bursts of right now and I would love to go back and forth without wasting minutes of booting and loading to do so.
  • I am expecting the "instant game start" to be some form of virtual "container" a la Kubernetes (but more light-weight.)
    They can spool up a separate one for each "Game" and keep them in a quiescent state for "instant access" under the XBx's Hyper-V function.
    Of course there will be a limit of how many you can keep "live" in memory at a time (they are much smaller than the actual game code as they are just a "saved state") and you can spool the rest out to 2nd tier on disk, and reload pretty quickly, especially with a NVme SSD.
    I have seen this done in Azure and it's pretty neat how fast you can jump between "instances" of applications this way without impacting your memory usage.
  • Yeah, agreed completely, I always run multiple games at once so knowing that I can avoid waiting for loading screens and the like will mean switching between games much easier. And especially good when people come over for some couch multiplayer, get a bunch of titles loaded up beforehand and just switch as required.