We sort the challenges of virtual reality into two groups, each needing a different decision. The first group is engineering: motion-to-photon latency, frame-time budgets, and the GPU and decoding load of drawing two high-resolution views fast enough that the brain accepts them. The second is commercial: build cost, upkeep, headset friction and, in many uses, added value that nobody has pinned down. The engineering problems are measurable on the device. The commercial ones decide whether the project should exist at all, so a business should settle feasibility and return on investment before it pays for rendering work.
That split also answers the question in the title. VR solves real problems where immersion and true scale change a decision or replace an expensive physical exercise. It struggles where it is a nicer version of something a web page already does.
Which challenges of virtual reality come down to GPU rendering performance?
Michael Abrash, speaking for Valve at GDC 2013, named three hard problems: tracking, latency, and stimulating the human visual system well enough to pass for the real world. Latency and display realism both run through rendering and the display pipeline. Tracking depends on sensor accuracy instead.
NVIDIA researchers wrote in 2017 that the best VR displays of the time had 1080×1200 pixels per eye, about 5 arc-minutes per pixel, against the roughly 0.5 arc-minute the human eye can resolve. Against the eye’s 0.5 arc-minute, 5 arc-minutes per pixel is a tenfold gap in angular resolution. They also noted a field of view of about 100° horizontally, half of human binocular vision, and that stereo depth doubles rendering cost.
Streamed 360° video raises the requirements further. A 2022 arXiv paper on virtual human teleportation cites a recommended minimum of 12K spatial resolution at 100 frames per second, which maps to a data rate of several Gbps. The same authors argue that the computing power available falls a factor of ten short of desktop machines, too little for the decoding and rendering those resolutions and frame rates need. They also argue that for 360° video the hardest task on the headset is decoding the content at the target frame rate, not rendering the viewpoint. Their conclusion is that these limits restrict present VR applications to offline operation, low-fidelity graphics, tethered high-end hardware, and predominantly gaming and entertainment.
Wireless standalone headsets swap the tripping hazard of a cable for lower content quality, frame rate and interactive latency than tethered systems. For a business, the question is not “can the GPU do it” in the abstract. It is whether the specific headset the users will wear can hold the frame budget with the specific content.
Some latency sources sit outside the GPU. Display lag is the sum of sensing, processing, data smoothing, transmission, rendering and frame-rate delays. And because ordinary operating systems give no real-time guarantees, latency spikes can appear without warning even in a system whose average latency was fine.
Why does VR make people sick, and how much do frame rate and latency help?
VR sickness comes from a mismatch: the eyes report motion that the inner ear does not feel, or feels late, and the brain reads the disagreement as something wrong. In a headset, whenever the user moves their head, the virtual head trails its true position because of display lag. A 2020 paper in Frontiers in Virtual Reality argues that the time-varying difference between virtual and physical head pose is what triggers cybersickness. Across their experiments, those authors found sickness severity rose steadily as constant display lag increased, whatever the scene or the axis of head rotation.
Frame rate matters too. When the refresh rate is lower than what the brain expects, users see glitches, and graphics that cannot keep pace with head movement can bring on motion sickness.
Those Frontiers authors report that more than 80% of participants using modern headsets with commercial games had some cybersickness after only 10 to 15 minutes. They also cite a study in which 11 of 20 participants were too sick to finish a head-mounted simulation, while none dropped out of the same simulation on a desktop.
Latency thresholds worth designing to
Quoted latency thresholds come from several sources:
- John Carmack (2013), as cited in a 2020 Frontiers review: under 50 ms to feel responsive, with under 20 ms recommended.
- Abrash (GDC 2013): between 1 and 20 ms for the whole pipeline, from head motion through tracking, rendering and display, when a typical game of the time ran at 35 ms or more.
- Users are sensitive to small changes in display lag, below 20 ms.
- Motion Systems, a simulator vendor, gives about 20 ms as the guidepost for seated VR and about 60 ms as the point where simulator sickness rises above baseline. It calls these guideposts, not hard rules, because sensitivity varies between people.
The average is not the whole story. The 2020 Frontiers review found that latency jitter and periodic latency patterns contribute to cybersickness as well as mean latency does. Motion Systems goes further and says a steady, slightly higher latency can be more comfortable than a lower but jittery one. A high frame rate is necessary but not sufficient: a 90 Hz panel refreshes roughly every 11 ms, which sets a floor under latency without guaranteeing a low end-to-end figure.
There is one more caveat on latency fixes. An effective display lag of 3 to 4 ms is now achievable, yet some participants still report symptoms under those conditions. The Frontiers authors think latency spikes are the most likely cause. They conclude that software techniques which hide display lag, such as asynchronous timewarp, will not fully solve cybersickness. Seating users helps, but according to Wikipedia’s entry on VR sickness it only delays the onset, and much immersive VR depends on standing or walking.
Frame budgets per refresh rate
Every frame has to finish its CPU and GPU work within a fixed number of milliseconds to hit a target frame rate. The budget is 1,000 ms divided by the frame rate.
| Frame rate | Frame budget (1000 ÷ fps) | Where it applies |
|---|---|---|
| 60 fps | 16.7 ms | Meta’s minimum rendering rate since its August 2024 update |
| 72 fps | 13.9 ms | Minimum refresh rate on most Quest devices |
| 90 fps | 11.1 ms | Supported on Quest 3 and 3S; the target Audi’s showroom models were built for |
| 120 fps | 8.3 ms | Supported on Quest 3 and 3S, not on all devices |
Meta’s two documents give different minimums. The optimisation guide states a 72 FPS minimum, while the Virtual Reality Check page records that on 7 August 2024 the rendering minimum dropped to 60 fps, or 30 fps with Application SpaceWarp, and the minimum refresh rate stayed at 72 Hz (60 Hz on Quest 2). We would treat the dated VRC page as current for certification and still budget for the refresh rate, because sickness comes from the frames you miss, not from what a store accepts. Interactive apps must run at one of 72, 80, 90, 96, 100 or 120 Hz.
GPU techniques that buy back frame time
Meta’s optimisation workflow starts with diagnosis. To check whether an app is GPU-bound or CPU-bound, turn the render camera off. If performance barely changes, the app is CPU-bound. If it improves a lot, the app is GPU-bound. Then drop the render scale to something tiny like 0.01. If performance does not change, the app is vertex-bound. If it improves, the app is fragment-bound. Vertex-bound apps usually need fewer draw calls, simpler geometry, level-of-detail systems or batching. Fill-bound apps need cheaper shaders. This split comes before choosing an optimisation, because Meta’s guide ties vertex-bound and fill-bound apps to different fixes.
Draw-call reduction. Road to VR reported that Audi’s engineering models, with every panel, screw and fuse, were far too complex to render at the 90 FPS modern headsets demand, and that reducing them by hand took one to two weeks per vehicle. The fix merged many parts into larger pieces, cutting draw calls while keeping geometric detail. It reduced the manual process to about 20 minutes per vehicle.
Foveated rendering. Foveated rendering draws a high-quality image where the eye is looking and a cheaper one in the periphery. NVIDIA’s 2017 study found that a total eye-to-image latency of roughly 50 to 70 ms could be tolerated before users started to notice artefacts. A paper at IEEE VR 2024 reported that retinotopic foveated rendering cut fragment shading by 27.2% on average with no perceived loss of image quality. Another IEEE VR 2024 paper applied the same idea to fluid simulation and reported a 2.27x speedup.
Multi-resolution shading. Lens-distortion correction squeezes the edges of the rendered image, so pixels shaded there are thrown away. NVIDIA’s Multi-Res Shading splits the image into a 3×3 grid of viewports and shades the outer ones at lower resolution. NVIDIA says it has seen this raise the frame rate of Everest VR by 40%, a vendor figure for one title.
Spacewarp. Oculus described Asynchronous Spacewarp as a frame-rate smoothing technique that almost halves the CPU and GPU time needed for nearly the same output. It does not scale well below half the display’s refresh rate, and imperfect extrapolation can leave visible artefacts. On Quest, Application SpaceWarp may run parts of an experience at half the refresh rate, but only if the app generates motion vectors to keep artefacts down.
A studio guide from Framesixty makes the related point that a scene running well on a workstation can still make a headset user queasy, so on-device testing is not optional.
Engineering limit or adoption problem: a triage table
| Challenge | Kind | What moves it | Basis |
|---|---|---|---|
| Motion-to-photon latency and jitter | Engineering | End-to-end measurement, prediction, timewarp; spikes need measuring to be found | Peer-reviewed review |
| Frame rate below refresh rate | Engineering (GPU) | Profiling GPU vs CPU and vertex vs fragment; draw-call and shader work | Vendor docs |
| Resolution and field of view | Engineering (GPU, display) | Foveated and multi-resolution shading | Research and vendor docs |
| 360° video streaming | Engineering (decode, network) | Viewport-driven tiling instead of streaming the full panorama | Research paper |
| Build cost and upkeep | Commercial | Format choice, scope, reuse of frameworks | Studio price guides |
| Comfort, hygiene and staffing | Operational | Session length, cleaning routine, trained staff | Expert interviews, studio guide |
| Unclear added value | Commercial | Defining the problem before the build | Expert interviews |
What does it cost to build and run a VR experience?
Most commissioned VR productions cost between $10,000 and $200,000 and take six weeks to nine months, according to Framesixty, a VR studio. That guide argues the format, not the target headset, sets the budget. VR Vision, a training developer, puts CG-based interactive simulations at $100,000 to $200,000 and says 360° video is usually 40% to 60% cheaper for the same runtime. Both guides come from vendors selling this work, so treat them as price signals rather than neutral surveys.
Interactivity drives cost. VR Vision describes cost as scaling with state: every variable a user can change adds engineering work. Multiplayer adds backend, networking and hosting, and VR Vision says it easily costs 1.5 to 2 times a single-player equivalent. Framesixty quotes Suffescom Solutions’ CEO saying that syncing spatial positions within a sub-30 ms window can double the timeline and raise the budget by 100% to 150%. In other words, multiplayer cost is a latency cost in disguise.
Running costs keep coming after launch. Framesixty advises that a $120,000 training module needs another $40,000 to $80,000 over its first two years for headset OS updates and content refreshes. That is between a third and two thirds of the build cost again within two years ($40,000 ÷ $120,000 and $80,000 ÷ $120,000). Integrating with a learning management system typically adds $15,000 to $40,000, per VR Vision.
The return shows up when an expensive exercise repeats. In a cost analysis of hospital disaster training at Wright State University, Farra and colleagues found that once development costs were spread over three years of repeated training, the VR exercise cost $115.43 per participant, while the cost of the live exercise stayed fixed. The same authors note that adoption may be delayed by high up-front costs with unknown returns.
Can VR let shoppers see a product at full scale before they buy?
Yes; the published examples include cars and furniture. Road to VR reported that Audi built a VR car configurator where buyers see detailed, life-size models in chosen colours and configurations, and deployed it to some 1,000 dealerships. Getting there required cutting the models’ complexity, as the draw-call example above shows. Retail experts interviewed for a 2025 Frontiers in Virtual Reality study saw more potential for furniture stores than for groceries or hardware, because customers get a better sense of how a piece looks in three dimensions.
Framesixty puts the pivot plainly: retail VR rarely fails on technology, it fails on throughput and measurement. Its arithmetic is simple. An eight-minute session plus two minutes of cleaning means one headset serves six shoppers an hour at best. The Frontiers experts named the cost of the technology, customer value that has not been recognised, and high implementation expense as the main reasons retailers hold back. They were also sceptical that implementation costs could be justified, and called the profitability of VR shopping unclear. One expert asked what direct added value VR offers over the real thing, and another called it a nice-to-have, not a must-have.
Retail VR cost spans a wide range. Framesixty quotes under $15,000 for a templated WebVR showroom and $400,000 or more for a multiplayer, AI-driven experience, with most first custom projects at $130,000 to $200,000 all-in. A first project takes 18 to 26 weeks. The conversion figure often quoted, products with 3D or AR content converting 94% higher (Shopify data compiled in BrandXR’s 2025 report), is about 3D and AR content. It should not be read as evidence for in-store headsets.
The cheaper routes deserve weight. The Frontiers experts suggested 360° websites, which need no headset, as a way to reduce cost, health and space problems. They also suggested using VR as a marketing and brand-experience tool rather than as a full store. Back-of-house virtual category testing avoids the throughput problem entirely, because it needs one workstation and no headsets facing shoppers. For the customer-facing side, we look at how immersive XR changes customer engagement separately.
Advantages and disadvantages of virtual reality for a business deciding whether to deploy it
The advantages of virtual reality are strongest where the real-world alternative is expensive, unrepeatable or impossible to stage:
- Repeatable training. It replaces a live exercise whose cost does not fall with repetition.
- Exposure therapy. It gives access to stimuli that in-vivo exposure cannot reach, such as thunderstorms, flying or crowds.
- Design review. In a systematic review of building information modelling (BIM), customers on a flat screen cannot see the real sizes of geometry and make wrong design decisions, while VR lets them see the actual dimensions of a building.
- Shared learning without travel. Participants can learn together in an interactive 3D world without being in the same place.
These are also the areas covered in more depth in our pieces on VR in healthcare treatment, VR for education and AR and VR in real estate.
The disadvantages of virtual reality are just as concrete. A 2025 survey of clinicians in Frontiers in Psychiatry grouped barriers into professional, financial, therapeutic and technological ones, including cost, immature technology and cybersickness. Respondents described headsets as uncomfortable and prone to artefacts, and complained about a lack of technical support. Some pointed out that headsets are affordable but the software is scarce or too expensive. Retail experts added that headsets become noticeable within half an hour and that someone trained has to be on hand to help customers. Some problems also need augmented reality rather than VR: on a building site, immersive VR cannot be used because workers must see the real world. The trade-off is set out in our guide on how to tell augmented and virtual reality apart.
Our position, carried over from how we scope AI work, is to establish first whether the expected outcome is plausible at current capability. A repeatable, well-defined task is far more likely to show measurable ROI than an ambitious one. For VR, that means asking whether a specific, recurring exercise gets cheaper or safer, not whether immersion will transform the customer relationship. One of the Frontiers experts made the same point: understand the problem first, then judge whether VR solves it better than the alternatives.
Where VR is worth its frame budget
Put the two groups of challenges together and a pattern appears. VR earns its cost where the decision depends on scale or presence: a car at life size, a building at real dimensions, a hazardous procedure rehearsed safely, a phobia stimulus that cannot be staged. Those are also the cases where it is worth paying for the engineering: profiling on the real headset, holding the frame budget, and measuring jitter rather than trusting an average. As a lower-barrier option, the experts point to 360° websites that need no headset.
So the open question for any deployment has two parts. Will the content hold its frame budget on the headset users will actually wear? And will the use repeat often enough to cover a build that, by the guides above, keeps costing money after launch? Where VR actually earns its cost in future applications and our wider list of virtual reality use cases and examples apply those two questions.
FAQ
What is the major challenge in VR access?
Retail experts in a 2025 Frontiers in Virtual Reality study named headset cost as a major barrier, citing the Apple Vision Pro at 3,499 USD. The experts also said older customers in particular struggle with the complex handling. Browser-based 360° stores lower that barrier because they need no headset.
Is VR healthy for your brain?
Cybersickness symptoms can persist for up to 12 hours after a session in some cases, according to the 2020 Frontiers in Virtual Reality authors. They also show that vomiting is rare: 15 of 1,028 university students in one study.
