Why Touch Latency Is the Ultimate Usability Metric

We obsess over chip names, core counts, and AI TOPS – but the one metric that defines your daily experience is touch‑to‑response latency. It's the gap between your finger landing on the glass and the UI reacting. Below 100ms, it feels instant. Above 120ms, you start to feel like the car is “thinking.” On a bumpy road, with your eyes scanning for traffic, that delay becomes a friction point – and a safety concern.

I tested three vehicles side‑by‑side under identical conditions, using a 120fps slow‑motion camera and a repeatable tap test. The results are clear – but also reveal some surprising variations that have nothing to do with the chip inside.

The Contenders

All vehicles were parked, fully booted, and left to idle for 5 minutes before testing. No charging, no climate load – just the infotainment system running.

Methodology – How I Measured

Equipment: iPhone 15 Pro set to 120fps slow‑motion (each frame = 8.33ms). A tripod mounted the phone so that the screen and the approaching finger were both in frame.

Test procedure:

• A single index finger tap on the native navigation app icon (same position on each home screen).

• Repeated 20 times per vehicle.

• In post‑production (Final Cut Pro), I counted the number of frames from:

  • Frame A – the first frame where the finger visibly touches the glass (deformation or contact shadow appears).

  • Frame B – the first frame where the icon changes colour, highlights, or starts an animation.

• Latency = (Frame B – Frame A) × 8.33ms.

I also ran a second round after 20 minutes of continuous load (navigation active, music streaming, climate on) to test thermal consistency.

The Results – Raw Latency Numbers

Quick takeaways:

• NIO ET5 – genuinely smartphone‑grade response; every tap felt instant.

• Zeekr 001 – consistently smooth; the occasional outlier still stays under the “noticeable” threshold.

• Li L7 – a clear laggard; 30% of taps crossed the 100ms line, and several exceeded 120ms.

Frame‑by‑Frame Visual Comparison (from video)

• Zeekr 001: Tap at frame 0 → UI changes at frame 9‑10 → ~75‑83ms.

• NIO ET5: Tap at frame 0 → UI changes at frame 4‑5 → ~33‑42ms.

• Li L7: Tap at frame 0 → UI changes at frame 11‑12 → ~92‑100ms.

The full side‑by‑side video (annotated with frame counts) is attached to this post – I recommend watching it at 0.25x speed to see the difference for yourself.

Why the Gap? – Let's Break It Down

1. The Silicon Story (Not the Whole Story)

The 8295 in the ET5 is objectively more powerful – roughly 2× CPU and 3× GPU throughput over the 8155. But the gap between the two 8155 vehicles (78ms vs 95ms) shows that software optimisation matters as much as raw hardware.

2. Display Technology

The ET5's AMOLED panel has a faster pixel response time than the LCDs used in the 001 and L7. That shaves about 5‑8ms off the tail end of the pipeline.

3. System Load – The Li L7's Achilles' Heel

The L7 drives two 15.7‑inch screens simultaneously (instrument cluster + centre display) with a heavy 3D UI. The 8155's GPU is running at near‑capacity, introducing queuing delays that the single‑screen 001 doesn't experience. In my thermal stress test, the L7's latency degraded by 24% (from 95ms to 118ms) after 20 minutes of load – the ET5 degraded only 14%, and the 001 degraded 18%.

4. OS Pipeline Optimisation

NIO's Banyan OS appears to have a leaner input handling stack – the 8295 helps, but the software engineering team clearly prioritised low latency. Zeekr's OS is decent, while Li Auto's UI seems to prioritise visual richness over responsiveness.

The Thermal Throttling Effect – Real‑World Consistency

I ran a second test after 20 minutes of heavy use (navigation, music, climate fan at medium). Here's how each held up:

The L7's degradation is the most concerning – it means that on a long drive with multiple apps running, the UI could become noticeably jerky. The ET5's 5nm chip runs cooler and maintains performance better.

Video Evidence – What You'll See

I've uploaded a side‑by‑side comparison video that shows:

1. All three vehicles, same tap action, in slow‑motion (120fps).

2. An overlay showing the frame count from contact to response.

3. A “sanity check” – tapping the same icon repeatedly to ensure consistency.

Link to video: [embed or attach – I'll add the YouTube/Vimeo link in the comments]. Also attached are screenshots of the frame‑by‑frame analysis for each run.

The Verdict – Which One Would I Want?

If you're shopping for a car, test the touchscreen yourself – don't rely on spec sheets. Spend two minutes zooming in and out on the map, swiping through menus, and launching apps. Your fingers will tell you more than any slide deck.

Call for Community Data – Help Build the Leaderboard

I've tested three vehicles. But there are many more out there – and my sample size is only 20 taps per car.

If you own any of these vehicles (or a different model) and can run a similar test, please share:

• Vehicle make, model, year, and trim

• Cockpit SoC and OS version

• Your test method (frame rate, number of runs)

• Your measured mean latency and any thermal observations

I'll compile all submissions into a public latency leaderboard – a living document that ranks cockpits by real‑world responsiveness, across brands, chips, and software versions.

Specific vehicles I'd love to see added:

• Tesla Model 3/Y (AMD Ryzen or Intel Atom)

• Huawei‑powered cars (e.g., AITO, Avatr) with HarmonyOS

• NIO ET5 with the same 8295 but different OS version

• Li L7 after the recent 5.1 OTA – did it improve?

A Final Word – Specs Aren't Everything

The 8155 and 8295 are both capable, but the way an OEM tunes the pipeline, manages thermals, and designs the UI has a massive impact on daily usability. A poorly‑tuned 8295 could lose to a well‑tuned 8155 – and this test shows that the gap between two 8155 implementations (Zeekr vs Li) is almost as wide as the gap between 8155 and 8295.

Don't buy the chip. Buy the experience.