System on a chip

A system on a chip (SoC) is an integrated circuit that combines most of the components of a computer or other electronic system onto a single chip. A typical SoC integrates a central processing unit (CPU), a graphics processing unit (GPU), memory interfaces, input and output ports, and specialized accelerators such as an image signal processor or a neural processing unit, rather than spreading those functions across separate chips on a motherboard. Placing these blocks on one die shortens the electrical paths between them, which lowers power consumption, reduces latency, and shrinks the physical footprint compared with a design built from discrete parts.

SoCs are the standard processing platform for smartphones and other battery-powered mobile devices, and the same characteristics, low power draw, small size, and tight integration, make them central to standalone virtual reality and augmented reality headsets. A standalone headset has no external computer to fall back on, so a single SoC inside the device must render the scene, run tracking, drive the displays, and manage sensors within a strict power and heat budget. Most current standalone headsets, including the Meta Quest 3, Pico 4, and Samsung Galaxy XR, run on Qualcomm's Snapdragon XR family of SoCs, while the Apple Vision Pro uses Apple silicon paired with a dedicated sensor coprocessor.

Architecture

An SoC is built around one or more processor cores and a set of supporting blocks connected by an on-chip interconnect or bus. The CPU handles general-purpose computation and runs the operating system; the GPU performs the parallel arithmetic needed for graphics and increasingly for machine learning. Around these sit fixed-function units that do specific jobs more efficiently than a general processor could: an image signal processor (ISP) cleans up and combines camera feeds, a digital signal processor (DSP) handles audio and sensor math, a neural processing unit (NPU) accelerates machine-learning inference, and dedicated video encode and decode blocks compress and decompress streams. Memory controllers, display controllers, and wireless modems are also commonly placed on the same die.

The trade-off in SoC design is that all of these blocks share a single power and thermal budget. A larger GPU or a higher CPU clock improves performance but raises heat output and shortens battery life, so designers balance the blocks against each other for the device's target use. Because the components are fabricated together, an SoC cannot be upgraded piecemeal the way a desktop PC's CPU and graphics card can; the whole chip is replaced when a new generation ships.

Why SoCs matter for standalone XR

PC-tethered virtual reality offloads the heaviest work, scene rendering, to a separate desktop graphics card that can draw several hundred watts and is cooled by large heatsinks and fans. A standalone headset has none of that. The processor sits on the user's face, runs from a small battery, and can only be cooled passively or with a small fan, so the SoC must deliver enough performance for VR inside a tight thermal envelope. In a 2025 feasibility study of the Quest 3, Laghari and colleagues describe the Snapdragon XR2 Gen 2's thermal envelope as roughly 4 to 6 watts in typical use, peaking near 10 watts, low enough to allow passive cooling in a standalone headset.

Within that budget the SoC has to do several jobs at once. It renders two high-resolution images, one per eye, at refresh rates of 72 Hz, 90 Hz, or higher; it runs inside-out tracking to compute the headset's position from the onboard cameras; it processes camera feeds for passthrough mixed reality; and it drives hand tracking and eye tracking where supported. Doing all of this on one low-power chip is the reason standalone XR SoCs lean heavily on fixed-function blocks: features such as foveated rendering (rendering at full detail only where the eye is looking) and motion extrapolation reduce the load that would otherwise fall on the GPU. The same study notes that sustained mixed-reality compositing on the Quest 3 can begin thermal throttling after roughly 10 to 15 minutes near full load, which illustrates how close these devices run to their thermal limit.

Qualcomm Snapdragon XR

Qualcomm dominates the standalone headset market with its Snapdragon XR line of SoCs. The company announced the Snapdragon XR1 on 29 May 2018, describing it as the first chip platform built specifically for extended reality. The XR1 combined Arm-based Kryo CPU cores, an Adreno GPU, a Hexagon DSP, a Spectra ISP, and the Qualcomm AI Engine, and it supported video output up to 4K at 60 frames per second across one or two displays.

The Snapdragon XR2 Gen 1, released in 2020 on a 7 nm process, became the workhorse of the standalone era. It powered the Meta Quest 2, the Pico 4, and the HTC Vive XR Elite, among others, while a higher-clocked variant, the Snapdragon XR2+ Gen 1, was used in the Meta Quest Pro.

The Meta Quest 3 was the first headset to ship with the Snapdragon XR2 Gen 2, announced in 2023. Built on a 4 nm process and based on the Snapdragon 8 Gen 2 smartphone chip, it offers up to 2.5 times the peak GPU performance of the XR2 Gen 1 with a 50 percent improvement in power efficiency, a CPU more than 33 percent faster, and up to 4 times the peak AI performance (8 times the AI performance per watt for INT8 workloads). The chip supports displays up to 3K by 3K per eye, can process data from up to ten cameras, and includes hardware acceleration that lowers passthrough latency to about 12 milliseconds. A further variant, the Snapdragon XR2+ Gen 2, powers the Samsung Galaxy XR, the first headset to ship with Android XR; Qualcomm states it runs at a 20 percent higher CPU frequency and 15 percent higher GPU frequency than the XR2 Gen 2 in the Quest 3.

The table below summarizes the main Snapdragon XR generations used in shipping headsets.

SoC	Announced	Process	Notable headsets
Snapdragon XR1	2018	10 nm class	Early 3DoF/6DoF standalone and AR headsets
Snapdragon XR2 Gen 1	2020	7 nm	Meta Quest 2, Pico 4, HTC Vive XR Elite
Snapdragon XR2+ Gen 1	2022	7 nm	Meta Quest Pro
Snapdragon XR2 Gen 2	2023	4 nm	Meta Quest 3, Meta Quest 3S
Snapdragon XR2+ Gen 2	2024	4 nm	Samsung Galaxy XR

Apple Vision Pro: dual-chip design

The Apple Vision Pro uses a different arrangement: two chips rather than one. At launch in February 2024, the headset paired the Apple M2 SoC, with an 8-core CPU and a 10-core GPU built on a 5 nm process, with a second custom chip called the Apple R1. The M2 runs the operating system and applications, while the R1 is a sensor coprocessor dedicated to processing input from the headset's 12 cameras, five sensors, and six microphones. Apple states the R1 streams new images from the cameras to the displays within 12 milliseconds, which it describes as eight times faster than the blink of an eye, to reduce perceptible lag.

In October 2025 Apple refreshed the headset, replacing the M2 with the Apple M5 SoC while keeping the same R1 coprocessor. The M5 model renders about 10 percent more pixels on the micro-OLED displays and raises the maximum refresh rate to 120 Hz, and Apple again describes the R1 as processing the same 12 cameras, five sensors, and six microphones with a 12-millisecond image-to-display time. Splitting general computation and sensor handling across two chips lets the sensor pipeline run on its own dedicated silicon, which keeps the main SoC free for rendering and the operating system.

See also

• Qualcomm Snapdragon XR2
• Vision processing unit
• Standalone VR
• Inside-out tracking
• Foveated rendering