Extended reality (XR) is not a rival to augmented reality. XR is the umbrella term, and AR, virtual reality (VR) and mixed reality (MR) all sit underneath it. ETSI lists the hardware targets for immersion: motion-to-photon latency under about 20 ms, displays refreshing at 90 Hz or more, and no visible pixel structure.
Is XR better than AR, or is AR one part of extended reality?
AR is one part of XR, so “better” is the wrong comparison. ETSI defines XR as all real-and-virtual combined environments and the human-machine interactions generated by computers and wearables, including AR, MR, VR and the areas in between. ETSI’s technology leaflet goes further and calls XR the umbrella for VR, AR, MR and “all future immersive technologies yet to be developed”.
The three forms differ in how much of the real world they keep:
- VR replaces the user’s field of view with a rendered scene, usually through a head-mounted display (HMD) with head and motion tracking. IEEE places it at the fully immersive end of the XR continuum.
- AR overlays additional information or generated content on the user’s current environment, either seen directly or relayed through sensors. Unlike VR, it does not block out the surroundings.
- MR is described by ETSI’s 2021 technical report as an advanced form of AR, where virtual elements are inserted so that they appear to belong to the real scene.
Scoping a real project against these categories (which one fits training, try-on or field service, and which headset class follows) is covered in our piece on mixed reality and how VR, AR and XR fit together, and the shorter question of what the difference between AR and VR is has its own explainer.
Immersion and presence are not the same thing
ETSI separates two related terms. Immersion is the sense of being surrounded by the virtual environment; presence is the feeling of being physically and spatially located in it. The distinction matters because ETSI ties presence directly to minimum performance requirements for tracking, latency, persistence, resolution and optics. Presence applies to AR too: for truly immersive AR and MR, users should not be able to tell virtual objects from real ones.
A headset can surround you with imagery and still fail to make you feel there. The gap between the two is where the hardware numbers below come in.
The rendering and GPU demands of immersion in XR
Immersive XR asks the GPU to render stereoscopic views in real time, against performance minimums set by the sense of presence. ETSI’s 2021 report lists the headline targets: less than 20 ms motion-to-photon latency (from head movement to the display changing), a display refresh rate of 90 Hz and beyond to eliminate visible flicker, and typically 100 to 110 degrees of field of view. IEEE gives the same latency threshold for VR displays and says polygon rendering pipelines for stereoscopic output need careful level-of-detail management to hold frame rates above 90 Hz.
Tracking runs faster still. ETSI says XR applications need highly accurate, low-latency tracking of the device at about 1 kHz sampling to keep the virtual and real worlds registered. Displays also need low persistence, switching pixels on and off every 2 to 3 ms so the image does not smear as the head turns.
Worked example: what a frame budget actually leaves you
Android’s XR developer guidance recommends 72 fps as a minimum and 90 fps as the ideal target, which it equates to an 11 ms budget per frame. The arithmetic is simple division of one second by the frame rate:
| Target frame rate | Time per frame (1000 ms ÷ rate) | Source of the target |
|---|---|---|
| 72 fps | about 13.9 ms | Android XR minimum |
| 90 fps | about 11.1 ms | Android XR ideal; ETSI and IEEE display targets |
| 120 fps | about 8.3 ms | Frame rate used as the VR headset requirement by Singh et al., IEEE VR 2023 |
Rendering technique affects how hard that budget is to hit. ETSI notes that ray-traced pipelines are typically considered more computationally expensive and less suited to real-time work, though real-time ray tracing has improved with hardware support. Texture formats matter as well: GPU-native compressed formats such as BC1, ETC2 and ASTC avoid decoding images on the CPU and copying them across.
Foveated rendering on headset versus host hardware
Foveated rendering is one lever for buying back GPU time, and its payoff depends on the workload and on the GPU it runs on. It lowers rendering quality away from the centre of gaze, exploiting the fact that peripheral vision has lower acuity.
Meta’s developer documentation explains why fixed foveated rendering (FFR) helps on its headsets: lens distortion means more pixels are rendered at the edge of the field of view than the centre, and the lenses blur those edge pixels anyway. Meta says FFR can give a 25% performance gain in pixel-intensive applications, while applications with simple shaders that are not fill-bound will likely see little improvement. In one ALU-bound scene Meta reports gains of 6.5%, 11.5% and 21% for its low, medium and high settings.
Tracked foveated rendering (TFR) follows the user’s gaze, so it can use a smaller high-quality region. Singh et al. (UIUC and NVIDIA, IEEE VR 2023) found TFR gave 1.5× to more than 3× speedup over non-foveated rendering, but its benefit over FFR was limited to two scenes and was at most 1.5×. The catch is the gaze tracker: it must finish in under 15 ms, and while its inference runs fast on desktop GPUs it is slow on mobile GPUs and can become the bottleneck. With the default input size, even at its highest pruning level, the authors measured their tracker at roughly 40 ms, too slow, before applying approximation techniques that cut gaze-tracking energy by 20×.
The measured difference is in the gaze tracker’s inference, fast on desktop GPUs and slow on mobile ones, so the same foveation technique costs differently depending on the GPU that runs it. On mobile GPUs the tracker’s inference is slow and can become the bottleneck; on desktop GPUs it runs fast. When a team is missing its frame budget, TechnoLynx’s GPU performance engineering work profiles where the milliseconds go.
How frame rate and latency affect virtual reality sickness
Missed frames and high latency are linked to discomfort and sickness. Android’s XR guidance states plainly that missing frame-rate targets can cause discomfort and motion sickness. A 2020 review by Stauffert, Niebling and Latoschik in Frontiers in Virtual Reality collects studies that manipulated latency as the independent variable and measured cybersickness, and finds that increased latency consistently leads to increased sickness.
Average latency is not the whole story. The same review reports that occasional latency spikes, not only constant latency, provoke cybersickness, and that jitter and periodic latency patterns contribute too. Without a real-time operating system, unforeseeable spikes can harm an experience even when latency was previously acceptable. Latency also varies with device, software configuration and input method, and the authors conclude that spikes can often only be found by measurement.
For hardware choice, this shifts the question from “can it reach 90 fps” to “does it hold 90 fps every frame under the real scene load”. Missed frames and occasional latency spikes both provoke discomfort. Sustained frame time and the worst-case tail therefore matter more than headline peak numbers. A draft comfort standard for XR devices names minimal refresh rate, pixels per degree, flicker and field of view among its display requirements.
Which technical limits still hold back immersion in XR headsets?
Two limits stem from optics and panels rather than rendering: the vergence-accommodation conflict and visible pixel structure.
The vergence-accommodation conflict
In normal vision, the eyes rotate to point at an object (vergence) and focus on it (accommodation) at the same distance. Current XR headsets create depth by shifting an object’s position between the two displays, so the eyes converge as they would in the real world, but they must still focus on the display’s single fixed focal plane. When the two distances differ, the vergence-accommodation conflict (VAC) occurs. Magic Leap lists the resulting discomfort as eye strain, blurred vision, headache, eye fatigue, double vision, dizziness and nausea.
The strength of the evidence deserves care. Hoffman, Girshick, Akeley and Banks (Journal of Vision, 2008) pointed out that earlier studies, which found more fatigue with 3D than 2D displays, could not isolate VAC from other differences such as eyewear, cross-talk and perceptual distortion, and that no one had yet shown the decoupling itself causes fatigue. Kramida’s review of head-mounted display designs treats VAC as a major factor in discomfort, especially for near tasks in VR or AR. Our reading is that VAC is a real contributor to discomfort, strongest for close-up content, while the size of its share relative to latency and other display faults is still less settled.
Magic Leap placed the Magic Leap 2’s focal plane so that content from 0.37 m to optical infinity (10 m) stays within a comfortable range. That helps, but it is a placement choice, not a fix for content held at arm’s length or closer.
The screen-door effect and pixels per degree
The screen-door effect is the visible grid of unlit gaps between pixels, the result of a display with a low fill factor. It is a different fault from mura (uneven colour between pixels) and aliasing (jagged diagonal edges), though higher pixel density reduces aliasing too.
ETSI’s 2021 report frames the target as no visible pixel structure, and notes that the fovea would need about 120 pixels per degree to match reality. It records how the bar has moved: in 2014 about 1k by 1k pixels per eye was thought sufficient, by 2019 2k by 2k per eye was commonly accepted as acceptable, and Palmer Luckey has said pixelation will not fully disappear until around 8K per eye. Meta notes that FFR lets applications raise eye-texture resolution while maintaining performance and power levels.
Checklist: what to verify when choosing XR hardware for immersion
| Check | Target or concern | Why it matters |
|---|---|---|
| Sustained frame rate under the real scene | 72 fps minimum, 90 fps ideal | Missed frames cause discomfort and sickness |
| Motion-to-photon latency, including spikes | under about 20 ms | Spikes and jitter also provoke cybersickness |
| Where foveation runs | gaze tracking in under 15 ms | Tracker inference can bottleneck mobile GPUs |
| Pixels per degree and fill factor | no visible pixel structure | Low fill factor produces the screen-door effect |
| Field of view | typically 100 to 110 degrees | Sets the extent of the observable world at any moment |
| Focal plane and near-content range | e.g. 0.37 m to 10 m on Magic Leap 2 | Close content worsens VAC discomfort |
Where XR immersion is already proving useful in education and life-sciences training
The education evidence shows a small overall gain, larger in some settings than others. Wu, Yu and Gu’s 2020 meta-analysis in the British Journal of Educational Technology pooled 35 randomised or quasi-experimental studies published between 2013 and 2019 and found immersive VR with HMDs more effective than less immersive approaches, with a small effect size of 0.24. The effect was greater for K-12 learners, in science education and specific abilities development, for simulations and virtual worlds, and when compared with lectures or real-world practice, and it held over time.
A small pooled effect is not a reason to dismiss XR in teaching; it is a reason to pick the use cases where moderators point to larger gains rather than headset-everything rollouts.
In health professions, one surgical training study is Seymour et al. (Annals of Surgery), which set out to demonstrate that VR training transfers technical skills to the operating room. One detail is telling for hardware buyers: the MIST VR simulator in that study ran on a 400 MHz desktop PC with a 17-inch CRT monitor at roughly 15 frames per second. A systematic review, Kyaw et al. (JMIR, 2019), assessed VR in pre- and post-registration health professions education using Cochrane methodology across randomised trials from 1990 to August 2017.
The summary for a life-sciences or education team is this: Wu et al.’s pooled evidence shows a small gain, larger for K-12 learners, science education and simulations, and it does not test whether a given headset holds its frame rate in your content. That last part is measured on your own scene, against the frame and latency budgets, before anyone is asked to wear the device for an hour.
