Technology · 08 Processors & Devices
Phones, tablets and laptops changed more in 2025 and 2026 than in the five years before, mostly because of the neural processors now inside them. Here is what that means for the software we build for you, and how we decide what to support.
Updated September 2026 You own everything Reply within one business day
Good morning
Your account at a glance
Balance
12,480
OlDevs, a full-stack studio in Vancouver, designs every app around the devices its users actually own rather than the newest flagship.We keep a device test matrix that spans current Apple, Qualcomm, Google, MediaTek, Samsung, Intel and AMD hardware, set a minimum-OS policy per project (typically iOS 17 and Android 10 or newer), and hold each release to a written performance budget before it ships.
Key facts
08 · Processors & Devices
Each card names the hardware, when it should shape your decisions, and what supporting it will cost you over time.
Right when your audience is consumer or professional and iPhone share is high, as it is in Canada. The A17 Pro and newer run Apple Intelligence and the on-device Foundation Models, so features can use them where present. Later cost: an iOS release every September to test against.
Right when reach matters more than a premium tier. Snapdragon 8 Elite and 8 Elite Gen 5, Tensor G4 and G5, Dimensity 9400 and 9500 and Exynos 2500 all carry capable NPUs; the mid-range phones most people buy do not. Later cost: more devices to test, and features that must degrade gracefully.
Right when you need a desktop or tablet app, or your web app will live on iPad all day. The M4 and M5 offer the memory bandwidth local models need and long battery life. Later cost: macOS and iPadOS release cycles, plus App Store or notarisation upkeep if you distribute natively.
Right when you build Windows software for offices refreshing their laptops. Intel Core Ultra 200V and series 3, AMD Ryzen AI 300 and Ryzen AI Max, and Snapdragon X Elite, X Plus and X2 Elite all pass the 40+ TOPS NPU bar. Later cost: two CPU architectures, x64 and Arm, to test and package for.
Right for nearly every project, because this is where most of your users are. Our default floor of iOS 17 and Android 10 keeps them in. These devices have weak or no NPU, so on-device AI must be optional. Later cost: a yearly, written decision on when to raise the floor.
Right when your audience is broad, the product changes often, or install friction would lose people. The browser is the one target that spans every chip above. Later cost: Core Web Vitals, INP in particular, and a JavaScript budget sized for the cheapest phone you still support.
Stack
Concrete tools, current at the time of writing; confirmed for your project at scoping.
Apple
Android
Windows and PC
On-device model runtimes
Testing and profiling
What changed
What moved in this area and what it means for your build.
Sep 2025
Apple's A19 and A19 Pro shipped with iOS 26, which added the Foundation Models framework so apps can call Apple's on-device model without sending data off the phone. For a client this means private AI features on newer iPhones, with a fallback for older ones.
Aug 2025
Google's Tensor G5 put more weight on on-device Gemini Nano features. It matters less as one phone and more as a signal: Android's own AI APIs (ML Kit GenAI, AICore) are the sanctioned path, so we prefer them over shipping our own model.
Sep 2025
Qualcomm and MediaTek refreshed their flagship phone chips in the same month, and Qualcomm announced Snapdragon X2 Elite for laptops. Flagship Android is now a fairly uniform NPU class while the gap to mid-range widened, so we design for the mid-range.
Nov 2025
New apps and updates on Google Play that target Android 15 or later must support 16 KB memory pages. Pure Kotlin apps were mostly unaffected; anything with native libraries needed rebuilding. If your app has not shipped an update in a while, we check this first.
Early 2026
Intel's Panther Lake chips and Qualcomm's X2 Elite reached shipping laptops, putting Copilot+ class NPUs into mainstream business fleets. Windows ML and ONNX Runtime can now target a real installed base, provided x64 and Arm builds are both tested.
2026
iOS 26 dropped the iPhone XS and XR, and Google Play requires new apps and updates to target a recent Android version each year. Our default floor is iOS 17 and Android 10 at the time of writing, revisited at each annual release using your analytics.
How we choose
If your own developers will take over, we match their platform skills and tooling. If we maintain it, we still document the device floor and test matrix so the knowledge stays with you, along with all code and accounts.
Existing products come with device and OS data. We read it before setting any floor, and place the floor where you keep the users who matter rather than at the newest release.
A warehouse scanner, a nurse's tablet, a commuter's phone and an office laptop have different chips, screens and network conditions. The environment decides the reference device and the performance budget.
On-device processing keeps personal data on the phone, which can simplify privacy obligations in Canada. Where a feature still needs a server, we place it in a region you choose.
Every chip generation and OS release adds cost: testing, rebuilds, store submissions. We estimate that yearly load up front so the quote reflects the life of the product, not just the launch.
WCAG 2.2 AA and bilingual EN/FR support are part of the device plan. Assistive technology, large text settings and older hardware are tested together from the first weekly demo.
Who it's for
Whether the audience is a customer, a member, a citizen, or your own team, the choice has to hold up under real use.
Fleet devices are usually standardised and managed through MDM, which lets us target a narrow, well-known matrix and use NPU features with confidence. We document that matrix so IT can plan refreshes around it.
Members and the public bring whatever they own, including older phones and assistive technology. We set a wide floor, keep AI features optional, and verify WCAG 2.2 AA on real devices in both official languages.
Each location buys its own hardware, so consistency is the real problem. We recommend a short approved-device list, build to the weakest entry on it, and test point-of-sale and tablet flows on exactly those models.
Ship for the devices your first customers hold, not the ones in a keynote. We help you pick one platform to start, set a floor that avoids expensive edge cases, and leave a clear path to on-device AI once the audience has the hardware.
Related technology
Native iOS in Swift and SwiftUI, native Android in Kotlin and Jetpack Compose, or cross-platform with React…
Explore07Most AI projects never need a training cluster. We match each workload to the right silicon, rent before we…
Explore09From the workstations we build on to the kiosks, terminals, scanners and sensors your product ships with,…
ExploreFAQ
No. New chips arrive each September and are a small share of users for months. We test on them so nothing breaks, but the product is designed for the two to four year old phones most of your audience carries. Flagship-only features are added as optional enhancements.
An NPU, or neural processing unit, is the part of a chip built to run machine learning models efficiently, so transcription, image cleanup or summarising can happen on the device without draining the battery. Most apps do not need one. If yours benefits, the feature uses the NPU when present and falls back to a server or a simpler path when it is not.
Those floors keep almost everyone with a phone from 2019 onward, while letting us use modern security, privacy and UI frameworks. They are defaults, not rules: analytics showing a meaningful group on older versions can lower the floor for a defined period, and managed devices can justify raising it.
We keep physical phones, tablets and laptops in the studio covering the main chip families, and use cloud device farms such as Firebase Test Lab, BrowserStack and AWS Device Farm for breadth. Automated tests run on every build; a manual pass on real hardware happens before each weekly demo and release.
A written set of limits the product must meet on a reference device: how quickly the first screen appears, how long key actions take, how much memory and battery it uses, and for web apps the Core Web Vitals including INP. We measure against it on every release so regressions are caught before users notice.
Only if your users have those laptops. Snapdragon X series machines are common in some business fleets and rare in others. Most modern frameworks ship x64 and Arm builds from one codebase, so the extra cost is mainly testing and packaging, which we include when your fleet data justifies it.
Where the device and region support them, yes. Apple exposes its on-device model through the Foundation Models framework on Apple Intelligence-capable hardware running iOS 26; Android offers Gemini Nano through the ML Kit GenAI APIs on supported phones. Both are limited to newer devices and certain languages, so we keep a fallback.
Let’s connect
Tell us what you’re building. We’ll reply within one business day with recommended platforms, structure and a tailored quote — no obligation.