A VR headset’s field of view (FOV) is the extent of the virtual scene visible through it, and many current headsets list about 100-110 degrees. Albert et al. (2017) note that current displays are limited to about 100° horizontally, only half of the human binocular horizontal field. Rakkolainen and colleagues (2017) put the natural human field at about 210° wide by 135° tall, against roughly 100°×110° for the HTC Vive and 80°×90° for the original Oculus Rift. A 2020 survey of wireless VR requirements puts the commercial ceiling at 150°, reached by the Pimax 8K. So a “good” field of view in VR depends on what you are trading it against: a wider periphery costs sharpness, panel size, headset bulk and GPU time. The rest of this article covers how those costs arise and how rendering techniques claw them back.
What is a good field of view for VR, and how does it compare with human vision?
Human vision varies across the visual field. The fovea, the region of high acuity, covers only a few degrees, and colour vision, binocular vision and motion sensitivity all change from the centre out to the edges. NVIDIA researchers (Albert et al., 2017) describe the fovea as about 1.5° across, and note that by 20° from the centre the eye can no longer resolve gratings finer than 7.5 arc-minutes. UploadVR, in its 2022 Quest Pro analysis, gives the foveal area as around 3° wide. Either way, the sharp part of vision is tiny, and the periphery is where optical flow is primarily detected.
The comparison below uses figures from the sources. One caution before you read across rows: sources quote horizontal, vertical or diagonal figures, so the HTC Vive appears below both as “100°×110°” and as “110°”.
| Display or system | FOV as published | Convention | Source |
|---|---|---|---|
| Human visual field | ~210° × 135° | horizontal × vertical | Rakkolainen et al., 2017 |
| Human visual field (requirement) | 210° × 150° (about 200° diagonal) | as published | Cellular-connected VR survey, 2020 |
| HTC Vive | ~100° × 110° | as published | Rakkolainen et al., 2017 |
| HTC Vive, Oculus Rift, Vive Pro Eye, Odyssey | 110° | diagonal | Cellular-connected VR survey, 2020 |
| PlayStation VR | 100° | diagonal | Cellular-connected VR survey, 2020 |
| Pimax 8K | 150° | diagonal | Cellular-connected VR survey, 2020 |
| Microsoft HoloLens 2 | 52° | diagonal | Cellular-connected VR survey, 2020 |
| ThinVR research prototype (Intel) | 180° | horizontal | Ratcliff et al., 2020 |
The same 2020 survey breaks the human field into zones: a central vision zone of about 60°, a less sensitive peripheral zone of about 30° around it, and a monocular zone beyond that, with resolution requirements falling from the centre outwards.
How is a VR headset’s field of view measured: horizontal, vertical or diagonal?
All three are reported, and you need to know which one a number refers to. In 3D rendering, FOV began as the horizontal angular range of the view, a convention that first-person shooters such as Quake made familiar. Renderers only draw rectangular frames, so if you know the angle in one direction and the aspect ratio, the other angles follow from the geometry.
Worked example: one frustum, three numbers
One VR rendering article gives a clean illustration. A single view rendered at 16:9 with a 120.0° horizontal FOV has a vertical FOV of 88.5° and a diagonal FOV of 126.6°. The horizontal value and the aspect ratio were chosen and the other two were calculated. The same frustum can therefore be quoted as “120°” or “127°” depending on which axis the marketing team prefers.
Headsets add two complications. First, the per-eye views overlap only partly, so the HMD Geometry Database reports a total (stereo) horizontal FOV, meaning the range seen by at least one eye, separately from the stereo overlap seen by both. Second, round lenses meet rectangular render targets, so headsets render the full rectangle and then apply a hidden area mask (HAM) over the parts the user can never see. Because of that mask, the diagonal FOV, which you would expect to be the largest number, is sometimes smaller than the horizontal one.
Rendered FOV versus the FOV you actually see
Rendered FOV is the technical condition the headset and application work to, and it is reproducible: the hmdq command-line tool queries the headset through the OpenVR API, reads the geometry it advertises and computes the rendered FOV from it. The user cannot see more than is rendered, but they can see less, depending on face shape, eye depth, facepad thickness, lens FOV and interpupillary distance.
Valve makes the same point from the hardware side. Its Index design notes say it was common for users to get less, sometimes much less, than a headset’s theoretical maximum FOV because of fit and facial geometry. In a lens-limited design aiming above 90°, a single millimetre of excess eye relief costs about 3° of FOV, according to Valve. Valve also notes that the FOV the GPU renders for the Index is similar to that of a Vive or Vive Pro; the gain comes from delivering more of that rendered image to most users through eye-relief adjustment and a 50 mm lens.
For anyone comparing headsets, the practical consequence is to treat a spec-sheet FOV as an upper bound and the rendered FOV from a tool such as hmdq as the comparable number, then expect the visible FOV to vary by person.
Why do most VR headsets stop at around 100-110 degrees of field of view?
Wide FOV and high resolution pull against each other, because a fixed number of pixels spread over a wider angle gives fewer pixels per degree. Angular resolution, measured in pixels per degree (PPD), is set by display resolution and FOV together, and Valve describes it as a critical trade-off in HMD design. The 2020 survey table lists 110° headsets at 11-14 PPD and the 150° Pimax 8K at 14 PPD, while its human-perception target is 60 PPD. Albert et al. (2017) state the gap in another unit: a typical display gives about 5 arc-minutes per pixel, while the eye can resolve gratings as fine as 0.5 arc-minute.
Optics and panels add their own limits. With parallel (flat-facing) displays, the panel width needed grows steeply as the horizontal angle widens, which is why some headsets cant their displays outwards: more horizontal coverage from smaller panels, at the cost of less stereo overlap. Intel’s ThinVR team (2020) notes that today’s VR displays are bulky, typically limit FOV to around 90°, and become bulkier still when pushed towards 180°. Even a perfect display cannot beat the face: nose, cheeks and eyebrows block the most extreme peripheral views.
The research prototypes suggest wider FOV is feasible with different optics. Rakkolainen et al. built Fresnel-lens prototypes covering the full human FOV, with optics costing under USD 1 per eye, and concluded that curved optics and flexible displays may make wide-FOV HMDs feasible. They also argue that the periphery needs less precision, since the eye itself cannot see fine detail there, which eases demands on peripheral optics and rendering. ThinVR reaches 180° horizontally with curved displays and custom heterogeneous lenslet arrays.
How a wider field of view raises GPU rendering cost
Rendering cost follows pixels shaded, and a wider FOV at constant sharpness means more of them. Lens distortion makes it worse. Meta’s documentation explains that producing the post-distortion image needs more pixels at the edge of the FOV than at the centre, while the lens blurs those same edges, so the GPU spends a lot of time on pixels that cannot be seen clearly.
Meanwhile, in wide-FOV VR almost all pixels fall where visual acuity is low. Albert et al. (2017) cite a figure that 96% of pixels in wide-FOV VR displays lie outside the central 20°, the region beyond which acuity has already fallen sharply. The 2022 survey by Wang, Shi and Liu reports a 35× drop in spatial sensitivity from the centre of vision to the periphery.
Researchers on gaze-tracked foveated rendering drew the scaling consequence. Their analysis predicts that foveated rendering turns the growth in rendering cost with display FOV into a roughly linear one, rather than quadratic or worse, for planar displays at constant sharpness. Put the other way round: without foveation, widening the field of view at the same sharpness pushes cost up faster than the angle itself. The bandwidth side scales too: the 2020 survey estimates around 50.39 Gbps of downlink for 360° VR at 60 PPD and 120 Hz under 20:1 compression.
How foveated and lens-matched rendering offset the cost of wide FOV
The offsets all rely on the same fact: the image does not need equal quality everywhere. They differ in what they reduce (rendered resolution or shading rate) and in whether they know where the user is looking.
Lens-matched and multi-resolution shading
NVIDIA introduced Multi-Resolution Shading (MRS) with Maxwell and Lens-Matched Shading (LMS) with Pascal so developers could split the render surface into subregions and match the sampling rate to the lens in each, instead of overshading everywhere to satisfy the most demanding spot. NVIDIA’s own figures, reported by HotHardware, show what that buys: the final image in the headset is 1.1 megapixels per eye, the first pass without LMS is 2.1 megapixels, and with LMS it is 1.4 megapixels, which NVIDIA describes as a 50% increase in throughput available for pixel shading. As a check on the arithmetic, 2.1 ÷ 1.4 = 1.5, so the GPU shades two-thirds of the pixels it otherwise would. LMS does this without lowering peripheral fidelity, unlike MRS, according to the same report. Everest VR developer Sólfar reported a 15% performance uplift over Multi-Res Shading; that is one developer’s figure for one title, not a general benchmark.
Variable rate shading
Variable Rate Shading (VRS), introduced with NVIDIA’s Turing architecture, lets each 16×16 pixel region of the screen have its own shading rate, with seven options such as one shading result for a 2×2 or 4×4 block, or non-square footprints like 1×2 and 2×4. VRS changes the shading rate without changing the visibility rate, so geometry edges stay sharp while shading work drops. NVIDIA notes that VRS, MRS and LMS run on separate hardware paths and can be combined. It also ships VRS Wrapper APIs for foveated rendering in DX11 applications, and Variable Rate Supersampling, which uses the saved budget to supersample the foveated region up to 8x, capped by the MSAA sample count.
Fixed foveated rendering
Fixed foveated rendering (FFR) renders the edges of the frame at lower resolution than the centre, with the high-resolution region fixed in the middle rather than following the eye. Meta states that FFR can give a 25% performance gain in pixel-intensive applications, and lists the benefits as higher framerate in fill-bound apps, lower power and heat, longer battery life, or headroom to raise eye-texture resolution. Meta also gives the limits plainly: FFR may not help applications with simple shaders, and content that draws the eye to the edges (a belt-mounted interface, for example) will expose the degraded image. FFR levels can be changed per frame, and Meta suggests level transitions, menus and new map areas as natural points to change them.
Eye-tracked foveated rendering
Eye-tracked foveated rendering (ETFR) moves the full-resolution region to wherever the user is looking. On Quest Pro, Meta’s developer options reduce the periphery from 4x fewer pixels at Level 1 to mostly 16x fewer at Level 3. Our overview of VR headset hardware and eye tracking covers eye tracking in more detail.
The measured and claimed savings vary widely, so the table keeps each figure with its source and what kind of figure it is.
| Technique and setting | Saving | Kind of figure | Source |
|---|---|---|---|
| FFR, Quest Pro, default resolution | 26-36% | Meta test app | UploadVR (2022) |
| ETFR, Quest Pro, default resolution | 33-45% | Meta test app | UploadVR (2022) |
| FFR, Quest Pro, 1.5x resolution | 34-43% | Meta test app | UploadVR (2022) |
| ETFR, Quest Pro, 1.5x resolution | 36-52% | Meta test app | UploadVR (2022) |
| FFR, pixel-intensive apps | 25% | Vendor documentation | Meta documentation |
| FFR, ALU-bound scene, low/medium/high | 6.5% / 11.5% / 21% | Vendor documentation | Meta documentation |
| FFR / ETFR, PlayStation VR2 | ~60% / ~72% | Vendor claim, not an independent measurement | Sony, via UploadVR (2022) |
| Gaze-tracked foveation, 1920×1080 desktop | 5-6x faster, 10-15x fewer pixels shaded | Research result | Guenter et al. (2012) |
Two patterns stand out. In Meta’s test app, savings were larger at 1.5x resolution than at default resolution. Shipping headsets also sit far below the research multipliers: UploadVR attributes this on Quest Pro to roughly 50 ms of end-to-end eye-tracking latency and a 90 Hz display ceiling. Latency limits ETFR. Albert et al. (2017) found that foveated rendering tolerates about 50-70 ms of eye-to-image latency, beyond which the degradation is easily seen and the technique loses effectiveness. Added tracker latency of 80-150 ms significantly reduced the acceptable amount of foveation, while 20-40 ms did not. The same study infers that perceptual improvements such as temporal stability also improve latency tolerance.
Profile before deciding how much FOV the GPU can afford
Fixed foveated rendering helps only when the cost it removes is the cost that is actually limiting the frame. Meta’s own notes show this: simple-shader applications are a listed exception for FFR, and in its ALU-bound example 16% of GPU time went to timewarp, which FFR does not affect. A team that turns on aggressive foveation in a vertex-bound or CPU-bound scene gets the visual cost without the frame-time gain.
This is a specific case of a broader position we hold: optimisation advice only makes sense in the context profiling gives it, and vendor-specific features such as LMS or the VRS Wrapper APIs give access to deeper optimisation without being automatically faster than standard paths. Before widening the rendered FOV or tuning foveation levels, we would capture GPU frame timings to see whether the frame is fill-bound, shader-bound or limited elsewhere, then measure each technique against that baseline on the target headset at its real resolution.
How does peripheral vision in VR affect immersion and motion sickness?
A wider field of view increases presence and also increases VR sickness, and the periphery drives the sickness. A comparative analysis of HMD systems associates higher FOV with higher immersion because the user perceives more of the environment. Rakkolainen et al. report that wide-FOV displays improve orientation, situational awareness and object avoidance, are generally preferred by audiences, and affect distance judgement and motion sickness. Valve adds comfort and, depending on the experience, gameplay satisfaction to the benefits.
The sickness side has a known mechanism. Self-motion perception from optical flow is driven most strongly by motion at the periphery of the retina. Adhanom, MacNeilage and Folmer summarise earlier studies as showing that larger FOV increases presence but also increases VR sickness. That is why restricting the FOV during locomotion, by drawing an opaque vignette with a clear centre, is an established way to reduce sickness.
The restriction is noticeable. In their study on an HTC Vive Pro Eye with a field of view of around 110°, the authors restricted the view to a minimum of 55°, close to what experiences such as Google Earth VR apply; that was highly effective against sickness but quite noticeable to participants. A gaze-following (foveated) restrictor produced no significant difference in sickness or noticeability compared with a head-fixed one, but let participants scan a wider area of the scene without turning their heads.
The design implication is that maximum FOV is not always the target. For artificial locomotion, the periphery is where sickness enters, and narrowing it temporarily during movement is a deliberate choice, not a compromise. Rakkolainen et al. also note that the periphery does not need a sharp image: even peripheral lights without any image can improve perception. Low-detail peripheral content is therefore a legitimate design option.
