Variable refresh rate

Variable refresh rate (VRR) is a display technique in which the display's refresh rate follows the rate at which the GPU or other video source produces frames, instead of the display redrawing at a fixed interval. A 2020 invited paper in the SID Symposium Digest by NVIDIA's Gerrit Slavenburg and co-authors describes VRR displays as displaying an image "as soon as the GPU renders it and then holding it till next image is available", which removes stutter and tearing and minimizes latency through the system. The main commercial implementations are NVIDIA G-SYNC (announced in 2013), VESA DisplayPort Adaptive-Sync (2014), AMD FreeSync (launched 2015) and the VRR feature of HDMI 2.1 (2017).

VRR is common on gaming monitors, televisions and game consoles, but VR headsets have taken a different path. Their displays have generally run at a fixed rate chosen by the system or the application, and runtimes use reprojection to cover missed frames rather than slowing the display down. Low-persistence VR panels pulse their light once per frame, so changing the frame duration on the fly changes perceived brightness and can produce flicker; a 2019 Valve patent application, for example, describes this problem and a way to compensate for it. Current headsets such as the Meta Quest 3 and Apple Vision Pro instead switch between a set of discrete refresh rates.

Reviewed 6 October 2026. All 29 cited sources opened and checked: G-SYNC, FreeSync, Adaptive-Sync, HDMI 2.1, G-SYNC Compatible, PS5 and Pulsar dates and figures; Valve patent text; Meta, Apple and Valve headset refresh rates; and the authors, venues, dates and quoted findings of the seven cited papers. About review dates.

How it works

A conventional display refreshes at a fixed rate, for example 60 Hz, while a GPU renders frames at a rate that varies with scene complexity. In its 2013 review of the first G-SYNC monitor, AnandTech described the two classic failure modes. If a new frame is sent to the display in the middle of a refresh, the screen shows parts of two frames at once, separated by a visible tear line (screen tearing). If vertical synchronization (vsync) is enabled, the GPU waits for the next refresh before sending a frame, and when a frame is not ready in time the previous frame is shown again, which appears as stutter.

VRR removes the fixed schedule. In the G-SYNC implementation reviewed by AnandTech, the module in the monitor manipulated the vertical blanking interval (VBLANK), the idle period between drawing the last line of one frame and the first line of the next, so that the display held the current image until the GPU delivered a new one. The first module could hold a frame for at most 33.3 ms (30 Hz) before it redrew the previous image, and the upper bound of 144 Hz was set by the panel rather than by G-SYNC. AMD's first FreeSync demonstration in January 2014 also relied on variable VBLANK control, which according to AMD was already defined in a VESA standard and supported by the display engines of its recent GPUs and APUs.

Every VRR display has a supported range. When the frame rate falls below the minimum, the source has to repeat frames. AMD calls its version of this low framerate compensation (LFC): frames are duplicated so that the effective rate falls back inside the display's range. AMD's example is a 60 to 144 Hz display showing a 40 fps game, where each frame is doubled and the display runs at 80 Hz. AMD states that all displays in its FreeSync Premium and FreeSync Premium Pro tiers are certified to meet mandatory LFC requirements, and its current tier table also lists LFC for the base FreeSync monitor tier.

Because a VRR display can also slow down when nothing on screen changes, the technique saves power. VESA's 2014 announcement said Adaptive-Sync "significantly reduces power consumption" for static desktop content and low frame rate video, in addition to giving "tear-free images for gaming and judder-free video playback".

Flicker

Variable frame timing has a visual cost. On many panels, luminance differs slightly depending on how long each frame is held, so a changing refresh rate produces low-contrast flicker. Cai, Bozorgian, Ashraf, Wanat and Mantiuk explain that in VRR mode "minor differences in display luminance at various refresh rates create low-frequency components in the Fourier domain, leading to visible flicker", a type of flicker that critical flicker frequency models do not predict. They note that a common mitigation is to narrow the operating range: a monitor that supports 24 to 144 Hz may be restricted to 40 to 144 Hz in VRR mode and rely on LFC below that. A 2025 study in the Journal of the Society for Information Display likewise reports that VRR can cause flicker "when the refresh rate is lowered below 60 Hz or when the refresh rate is changed even above 60 Hz".

History

The underlying capability predates gaming monitors. VESA states that Adaptive-Sync had been a standard component of its Embedded DisplayPort (eDP) specification, used for internal display connections, since eDP's initial rollout in 2009. AMD told AnandTech in 2014 that variable refresh panels had been pushed for mobile devices for some time in order to reduce power consumption.

Milestones in VRR for PCs, televisions and consoles

Date	Event
18 October 2013	NVIDIA announces G-SYNC, a module built into monitors that synchronizes the monitor to the GPU's output. It required a GeForce GTX 650 Ti Boost or better GPU, and the first module was made for the ASUS VG248QE monitor.
6 January 2014	At CES 2014, AMD demonstrates variable refresh on two unmodified Toshiba Satellite Click laptops with Kabini APUs, under the working name "FreeSync".
12 May 2014	VESA adds Adaptive-Sync to DisplayPort 1.2a for external displays and offers it to VESA members without a license fee.
19 March 2015	AMD releases its first FreeSync driver. FreeSync uses DisplayPort Adaptive-Sync, which bit-tech reported was an optional part of DisplayPort 1.2a.
28 November 2017	The HDMI Forum releases HDMI 2.1, which includes Variable Refresh Rate alongside Quick Media Switching, Quick Frame Transport and Auto Low Latency Mode.
15 January 2019	NVIDIA drivers begin supporting VESA Adaptive-Sync monitors. Of more than 400 Adaptive-Sync displays NVIDIA tested, 12 were initially certified as "G-SYNC Compatible".
25 April 2022	Sony announces VRR support for the PlayStation 5 on HDMI 2.1 VRR-compatible TVs and monitors, delivered by a system update.
2 May 2022	VESA launches the Adaptive-Sync Display Compliance Test Specification, with more than 50 test criteria including flicker and two logo tiers: AdaptiveSync Display for gaming and MediaSync Display for jitter-free media playback.
8 January 2024	NVIDIA unveils G-SYNC Pulsar, which combines VRR with variable-frequency backlight strobing.
7 January 2026	The first G-SYNC Pulsar monitors go on sale: four 27-inch 2560x1440 360 Hz models from Acer, AOC, ASUS and MSI, priced from US$599.

The first G-SYNC module replaced the monitor's scaler with a board carrying an FPGA and 768 MB of DDR3 memory and connected over DisplayPort 1.2 only. FreeSync took the opposite approach: bit-tech noted that scaler makers did not have to pay a license fee to AMD or VESA to support DisplayPort Adaptive-Sync. NVIDIA's later G-SYNC Compatible program set requirements that a certified monitor show no blanking, pulsing, flickering or ghosting during VRR gaming and support a VRR range of at least 2.4:1, such as 60 to 144 Hz. The 2026 Pulsar monitors were the first G-SYNC displays developed with MediaTek, with G-SYNC built into the scaler rather than a separate module.

VRR and strobing

Backlight strobing (low persistence) and VRR were long incompatible on desktop monitors. NVIDIA wrote in 2024 that "strobing the backlight at a frequency that is not fixed causes serious flicker", and that this had until then prevented effective strobing on VRR displays. Its earlier ULMB and ULMB 2 modes worked only at fixed refresh rates. According to NVIDIA, G-SYNC Pulsar modulates overdrive by screen location and refresh rate and tunes the brightness and duration of the backlight pulses in response to the changing render rate. A similar conflict between low persistence and changing frame times arises in VR headsets (see below).

VR and AR headsets

Fixed refresh and reprojection

VR runtimes are built around a steady display rate. Michael Antonov of Oculus explained in 2015 that when rendering takes too long and a frame is skipped, the video adapter scans out the same image a second time, producing judder; asynchronous timewarp (ATW) addresses this by having a separate thread generate "a new timewarped frame from the latest frame completed by the rendering thread" before every vsync. Antonov also wrote that "there is no substitute for hitting the full frame rate when it comes to delivering great VR." Later techniques keep the display rate fixed as well. When an app keeps dropping frames, asynchronous spacewarp (ASW) and SteamVR Motion Smoothing make it render at half the display rate, such as 45 fps on a 90 Hz headset, and the compositor generates every second frame synthetically.

Display persistence is one documented reason. The background section of a Valve patent filed on 30 April 2019 and granted on 1 December 2020, with Jeremy Adam Selan as inventor, states: "Traditional displays for virtual reality (VR) systems, such as those embedded in a VR headset, operate at a fixed refresh rate." The patent explains that on a low-persistence display the light pulses come closer together at higher refresh rates, creating a brightening effect, and farther apart at lower rates, creating a dimming effect, so a varying refresh rate is seen as flicker. Its proposed fix changes the magnitude or duration of each light pulse according to the frame time so that brightness stays constant on "a low-persistence display that supports a variable refresh rate". A patent describes a proposed method; it does not show that a shipping product used it.

Academic work has noted the same gap. Denes and colleagues wrote in 2020 that "VR/AR headsets require constant and sustained refresh rates", so VR rendering engines adjust resolution to fit the frame budget, and that their adaptive-refresh algorithm "could not be tested because of the lack of adaptive-refresh-rate headsets". Cai et al. point out that low-persistence VR displays, which may light the image for about 2 ms of a 7 ms frame, can produce visible, high-contrast flicker, and that manufacturers have set minimum refresh rates empirically. They used their flicker model to predict the minimum flicker-free refresh rate for the Apple Vision Pro, Microsoft HoloLens 2 and Meta Quest 3, a prediction the authors note has not been validated.

Selectable and content-matched refresh rates

Instead of continuous VRR, headsets offer several fixed rates and switch between them.

Examples of refresh-rate switching in headsets

Headset	Rates	How the rate is chosen
Valve Index	80, 90, 120 and 144 Hz	Chosen by the user in SteamVR. A December 2019 SteamVR update allowed switching in real time without restarting SteamVR, on NVIDIA GPUs with current drivers.
Meta Quest 3	72, 80, 90, 96, 100 and 120 Hz; 144, 180, 200 and 207 Hz for apps built against OpenXR 1.0.63 or later; up to 240 Hz in developer mode with display scaling	Requested by the app through the OpenXR XR_FB_display_refresh_rate extension. The default is 72 Hz, and thermal throttling may drop an app running above 72 Hz to 72 Hz.
Apple Vision Pro (M2)	90, 96 and 100 Hz	Selected by the system according to content.
Apple Vision Pro (M5)	90, 96, 100 and 120 Hz	Selected by the system. Apple says the M5 model can raise the rate to 120 Hz to reduce motion blur when users look at their physical surroundings and for Mac Virtual Display.

Meta's documentation tells developers they can find out when throttling has changed the rate from the predictedDisplayTime field returned by xrWaitFrame. It also warns developers not to assume that future headsets will support rates above 120 Hz.

Apple's choice of rates follows the content. At WWDC 2023 Apple said the Vision Pro display runs at 90 Hz for most content and switches automatically to 96 Hz for 24 fps video. As UploadVR explained, 96 is a multiple of 24, so each film frame is shown for exactly four display frames and the judder of uneven frame pacing at 90 Hz is avoided. Code in a visionOS beta, reported by MacRumors, showed that the 100 Hz mode is used "to compensate for detected 50Hz flicker from artificial lighting", and that travel mode requires 90 Hz at all times. These are switches between fixed modes, not the frame-by-frame timing used by VRR monitors.

Display interfaces

The HDMI 2.1 specification that introduced VRR also introduced Quick Frame Transport, which the HDMI Forum said "reduces latency for smoother no-lag gaming, and real-time interactive virtual reality". Quick Frame Transport is a separate feature from VRR.

Research

Several perception and graphics studies treat the refresh rate as one variable in a trade-off between spatial and temporal resolution.

• Denes, Jindal, Mikhailiuk and Mantiuk (ACM Transactions on Graphics, 2020) built a model of perceived motion quality from judder and blur, measured motion quality between 50 Hz and 165 Hz, and used it in a motion-adaptive algorithm that drove a G-SYNC monitor's refresh rate from the rendering budget and on-screen motion. In validation it performed better than constant-refresh-rate rendering. The authors reported that rapid rate changes combined with the G-SYNC control system occasionally skipped frames, so for unpredictable motion they limited the algorithm to integer divisors of 165 Hz (55, 82.5 and 165 Hz). They suggested that a direct interface for requesting a specific refresh rate from the monitor would reduce this latency.
• Jindal, Wolski, Myszkowski and Mantiuk (ACM Transactions on Graphics, SIGGRAPH Asia 2021) combined refresh-rate selection with variable rate shading in a method called ALSaRR, tested on adaptive-sync monitors. The authors wrote that they hoped it could improve quality "not only for regular adaptive-sync monitors, but also for VR headsets."
• Cai et al. (SIGGRAPH Asia 2024) built what they describe as the first VRR flicker detection dataset, measured on an LG OLED G1 television, and used their elaTCSF model to calculate flicker-free refresh-rate ranges for VRR displays at different luminance levels.
• Tohidypour, Seto and Nasiopoulos (Journal of the Society for Information Display, 2025) compiled a dataset of 160 VRR luminance signals, ranging from 2 to 40 cd/m2, with perceived flicker levels from subjective testing. They reported that JEITA, which they describe as the most widely used flicker metric for VRR displays, correlates with subjective flicker perception at only 71.43 percent.

Related VR work varies temporal resolution inside the image rather than at the panel. Flöter, Geringer, Reina, Weiskopf and Ropinski (ETRA 2025) updated peripheral regions of an HTC Vive Pro Eye image less often than the fovea while the display stayed at 90 Hz, with the slowest regions updating at 18 Hz. In a study with 15 participants, the authors report that pixel rendering costs could be reduced by up to 63.6 percent "without users feeling uncomfortable". This is a form of foveated rendering and does not change the display's refresh rate.

See also

• Refresh rate
• Frame rate
• Screen tearing
• Judder
• Persistence
• Reprojection
• Motion-to-photon latency
• DisplayPort
• HDMI