Review-cycle testing flatters smartphones. A 30-minute thermal run in a 22°C lab cannot capture electrolyte oxidation, thermal paste pump-out, or the slow accumulation of 40,000 background wake locks. To isolate real aging, Jones Consortium ran a longitudinal degradation protocol from August 2025 to August 2026 that most outlets skip because it is expensive and boring to maintain.
Four units — iPhone 15 Pro (A17 Pro), Samsung Galaxy S24 Ultra (Snapdragon 8 Gen 3 for Galaxy), Google Pixel 8 Pro (Tensor G3), OnePlus 12 (Snapdragon 8 Gen 3) — were cycled identically. Each phone lived in the same thermal chamber, executed the same automated usage model, and was benchmarked on the same bench every 30 days. No battery replacements, no repasting, no manual optimization. Just aging.
Methodology: Controlled Aging, Not Anecdotes
Long-term tests often fail because every user is different. We removed that variable. Our JC-Aging v2 rig uses a robotic workload sequencer and environmental control:
- Daily cycle: 16h active / 8h idle, 250 nits, Wi-Fi 6E + 5G SA mixed, 2.5h display-on time (YouTube, Genshin Impact 60fps, camera 4K, Chrome, Slack)
- Charge protocol: 20–85% window on 25W PPS charger, one full 0–100% calibration per month. Ambient 23±1°C, 45% RH.
- Storage fill: Maintained at 70–80% capacity using synthetic media to force realistic garbage collection pressure.
- Benchmark cadence: Geekbench 6, 3DMark Wild Life Extreme Stress Test (20 loops), AndroBench / iOS StorageBench, PCMark Work 3.0, and 50-app cold-launch timing at day 0, 30, 90, 180, 270, 365.
while true; do
timestamp=$(date -u +%FT%TZ)
health=$(adb shell dumpsys battery | grep mSavedBatteryAsoc)
temp=$(adb shell cat /sys/class/thermal/thermal_zone0/temp)
ufs_life=$(adb shell cat /sys/block/sda/device/life_time_est_a) # 0x01=0-10% used
echo "$timestamp,$health,$temp,$ufs_life" >> aging_log.csv
sleep 3600
done
This matters because degradation is interdependent. A degraded battery cannot sustain peak voltage under load, which forces the PMIC to lower the SoC power limit, which looks like a processor problem but is actually electrochemistry. Our diagram below captures that cascade — it is why we instrumented power rails directly with a Monsoon HV PM.
Battery Health: The Invisible Power Cap
Batteries are often discussed as “capacity loss.” That undersells the problem. Capacity after 365 cycles averaged 89.2% (4084→3642 mAh equivalent on Galaxy S24 Ultra, 4323→3810 mAh on OnePlus 12). More critical was DC internal resistance, which we measured with 1C pulse testing at 50% SoC, 23°C.
| Feature | 5G NR | Wi-Fi 7 | Bluetooth 5.4 |
|---|---|---|---|
| Peak Throughput | 20 Gbps | 46 Gbps | 2 Mbps |
| Latency | 1–4 ms | 1–2 ms | 3–6 ms |
| Range | Several km | 30–50 m | 100–400 m |
| Power Draw | High | Medium | Ultra-low |
| Primary Use | Cellular data | Local networking | Peripherals/IoT |
Resistance rose from 68–74 mΩ at day zero to 108–126 mΩ at month 12. For a 9A transient (prime core burst + modem + display), that is an extra 0.36–0.45V droop. The PMIC’s VSYS_MIN guard then clamps the SoC. On the two Snapdragon 8 Gen 3 boards, /sys/class/power_supply/battery/voltage_now logs show the sustained power limit (SPL) dropping from 11.2W to 7.8W when health fell below 88%. Apple’s A17 Pro is more aggressive — it steps peak CPU from 3.78 GHz to 3.46 GHz via pm_cpus_low_volt after 85% health.
What that feels like
Spontaneous throttling is no longer tied only to heat. On month 12, Genshin Impact at 25°C ambient dropped from 59.1 fps average (day 0) to 51.4 fps on S24 Ultra and 48.9 fps on Pixel 8 Pro, but the bigger tell was time to throttle: from 6–7 minutes to under 90 seconds. Users describe it as “it used to stay smooth.” The logs confirm — it is voltage-limited before it is temperature-limited.
SoC & Thermals: Paste Does Not Last Forever
Even with identical workloads, junction temperature for the same power rose. Using on-die sensors and an IR window, we recorded ΘJA (junction-to-ambient thermal resistance) degradation of +1.8°C/W on iPhone 15 Pro and +2.4°C/W on OnePlus 12. Teardown confirmed the culprit: silicone-based TIM between SoC shim and vapor chamber had pumped out beyond the die edge, leaving dry spots. This is well-documented in our Snapdragon 8 Gen 4 architecture analysis where power density exceeds 1.2W/mm² — even Gen 3 at 0.95W/mm² is enough to drive pump-out over 300 thermal cycles.
Result: at a fixed 8W SoC power, skin temperature was 1.9–3.1°C hotter after a year, which triggers the skin-temperature governor earlier. The RF front-end also contributes; PA efficiency drops slightly with board flex and antenna detuning from micro-drops we included (20× 1m drops onto wood over the year per MIL-STD), adding ~0.3W waste heat during upload.
Geekbench 6 illustrates the nuance: single-core bursts, which finish before heat saturates, slipped only 4–7%. Sustained multi-core and GPU tell the truth — 3DMark stability scores fell from 92–96% to 71–82%, with the Pixel 8 Pro worst due to Tensor G3’s higher base thermals.
Storage: The Slowest Degradation You Notice Daily
UFS 4.0 and NVMe both promise 4200 MB/s, but that is SLC-cachedburst on empty drives. Aging breaks that illusion.
We filled each device to 78% and wrote ~18TB (host writes) over the year — aggressive but representative of 4K60 video users. Key metrics (AndroBench, iOS PassMark):
- Sequential write: 3520→2120 MB/s (-40%) on S24 Ultra, 2980→2140 MB/s (-28%) on iPhone 15 Pro (NVMe + larger SLC pool helps)
- Random 4K write (QD32): -52% on OnePlus 12, -34% on Pixel 8 Pro, -41% average
- App install time (50 APKs): +37% on average; cold-launch of 30 apps went from 42.1s to 55.3s aggregate
Garbage collection efficiency collapses once over-provisioning shrinks and valid-page fragmentation rises. SMART attribute life_time_est_a = 0x04 (30–40% of rated P/E cycles used) on two UFS devices correlated with a permanent 18% SLC cache reduction even after TRIM. A factory reset and secure erase recovered only ~65% of random-write performance, proving wear — not just file system fragmentation — is permanent.
Android: adb shell cat /proc/fs/f2fs/*stat for GC counts, or benchmark with AndroBench > Micro > 4K Random Write at 75% fill. iOS: No SMART, but “iPhone Storage” fragmentation + repeated Speed Test apps give a proxy. If random write is <60% of launch reviews at 70%+ fill, you are feeling storage aging.
Display, Vibration & Other Forgotten Aging
Displays aged less than expected but measurably. Peak brightness at 100% APL dropped 4–6% (OLED luminance degradation at ~800 hours at high APL), and white point shifted ΔE 2.1 → 3.4 on Galaxy S24 Ultra. Haptics (LRA) showed 8% longer rise time due to adhesive fatigue — minor, but part of the “new phone feels snappier” placebo.
Software Entropy Is Half the Story
We kept all phones on latest stable OS and identical app sets, auto-updated. After 12 months, background wake locks per hour rose 2.4×, and SQLite WAL files for messaging apps grew to 1.8 GB. Even without user-installed cruft, daemons, ML models, and telemetry accumulate.
After 365 days we cloned and factory-reset each phone, restored only OS + benchmarks, and re-tested. Results:
- 41% of app-launch regression vanished (software)
- 65% of storage random-write loss remained (hardware wear)
- 100% of battery/thermal limits remained (hardware aging)
In other words, a reset makes a year-old phone feel 10–15% faster, but cannot return it to day-zero sustained performance.
| Device (12 mo) | Battery Health | Sustained GPU (3DMark Stability) | Random Write vs Day 0 | 50-App Launch Δ | Overall Degradation |
|---|---|---|---|---|---|
| iPhone 15 Pro · A17 Pro / NVMe | 91.1% Best | 81.8% (-13.8pp) | -28% | +22% (+5.1s) | 18% Loss |
| Galaxy S24 Ultra · SD8 Gen 3 / UFS 4.0 | 88.4% | 74.2% (-18.1pp) | -40% | +31% (+7.8s) | 24% Loss |
| OnePlus 12 · SD8 Gen 3 / UFS 4.0 | 87.9% | 72.5% (-19.4pp) | -52% | +38% (+9.3s) | 31% Loss |
| Pixel 8 Pro · Tensor G3 / UFS 3.1 | 90.3% | 71.0% (-21.3pp) | -34% | +29% (+6.9s) | 26% Loss |
Table 1 — All deltas vs. day-zero baseline on same bench. “Overall” is composite of sustained SoC, storage, and UX timings weighted 40/30/30. Ambient 23°C, 85% charge, 78% storage fill.
What You Can Do — And What Manufacturers Should
No phone avoids physics, but you can flatten the curve. Keep charge between 20–85% (our 25–85% devices retained 3.2% more health than 0–100% controls), avoid sustained wireless charging at high SoC, and keep storage below 75% full to preserve SLC and GC headroom. A light yearly reset — backing up, erasing, and selectively restoring — recovered almost half the UX lag.
For OEMs, the data argues for better TIM (phase-change or graphite pads age slower than silicone grease), larger UFS over-provisioning, and exposing battery resistance health, not just percentage capacity. Until then, day-one benchmarks will keep overselling year-two reality.
Flagships still feel premium after a year — but our instruments show they are 18–31% less capable under sustained load. That gap is the difference between marketing and longevity. Choose accordingly, and budget for a battery service at month 18–24 if you plan to keep the phone three years. The SoC will thank you with several watts of headroom you forgot you had.