iPhone 18 Pro La Croix Cooling Exposes a 25% Sustained Performance Gap

Developer Matt Birchler placed a cold La Croix can against an iPhone 18 Pro and raised its sustained benchmark performance by about 25 percent.

The improvised cooling also helped the phone complete his entire transcription workload 14 percent faster. Those are different measurements, but both point toward the same constraint.

Apple’s A20 Pro can process the workload faster than the phone can continuously cool it. Even Apple’s redesigned vapor chamber does not eliminate that thermal ceiling.

The result is less a practical cooling recommendation than an unusually clear demonstration of mobile performance limits. Peak benchmark scores describe how quickly a phone starts, while sustained tests show what remains after heat accumulates.

Birchler’s test used a single phone, one application, and a decidedly uncontrolled cooling method. It is not enough to establish a universal 25 percent improvement across games, video exports, or generative AI models.

Still, the experiment challenges an important assumption about Apple’s newest thermal design. The iPhone 18 Pro thermal throttling he observed was modest, yet enough unused performance remained for a refrigerated drink to expose it.

The La Croix Test Raised Sustained Speed by About 25%

The cold can did not make the A20 Pro faster. It helped the chip retain more of the performance it already had.

Birchler conducted the experiment after testing transcription performance in Quick Subtitles, an application he developed. Its benchmark mode processes a built-in audio file with Apple’s on-device transcription model.

His original benchmark included one warm-up pass followed by 20 consecutive measured runs. That structure was designed to capture both peak speed and the decline after the phone became warm.

The uncooled iPhone 18 Pro reached a reported maximum of 447.9 words per second. Performance then dropped as the repeated workload heated the device.

During later runs, the result settled slightly above 320 words per second. That decline is thermal throttling, which reduces operating speed when temperature, power, or both approach a system limit.

Birchler could also feel the phone becoming hotter. He wrote that the iPhone 18 Pro took longer to warm than its predecessor, but then warmed more quickly.

For the follow-up, he removed a cold can of La Croix from his refrigerator. He started another benchmark and placed the can against the back of the phone.

The cooled test produced the result he expected. Sustained throughput stayed materially higher, while the complete benchmark finished sooner.

A reported benchmark analysis characterized the later-run improvement as roughly 25 percent. Birchler reported that the full job completed 14 percent faster.

Those figures measure different parts of the run. The larger figure describes the sustained performance region, while 14 percent reflects total completion time.

Early passes already benefit from a relatively cool device. External cooling matters more after repeated computation has transferred enough heat into the phone’s structure.

That explains why the improvement in total time was smaller than the later-run performance increase. The can had its greatest effect after the phone would otherwise have reduced speed.

The result also avoids a common benchmark misunderstanding. A cooling accessory does not rewrite the architecture or increase the chip’s theoretical peak.

Instead, it changes how long the processor can operate near its faster states. For continuous transcription, that difference directly affects the time required to finish a batch.

Birchler did not publish laboratory measurements for the can’s temperature, contact pressure, room conditions, or phone surface temperature. His result therefore remains an enthusiast experiment, not a controlled thermal study.

However, the underlying behavior is technically plausible. A cold aluminum can can absorb heat from the phone, while its large surface and liquid contents provide useful temporary thermal capacity.

That extra capacity delays the temperature rise that triggers throttling. Once the drink warms, the benefit should shrink unless another cooling source removes the accumulated heat.

The test provides a visible illustration of a normally hidden process. Performance changed because the phone’s thermal environment changed, not because its software or silicon changed.

Why iPhone 18 Pro Thermal Throttling Still Matters

A phone can lead short benchmarks while delivering a different ranking during workloads that run for several minutes.

Smartphone processors frequently operate in bursts. Opening an application, applying a quick photo effect, or loading a webpage may finish before temperature becomes the main constraint.

Longer work behaves differently. Video encoding, local AI inference, transcription, gaming, and sustained graphics rendering continuously produce heat.

That heat must travel from the A20 Pro through several interfaces before reaching the enclosure and surrounding air. Every layer adds resistance to the path.

If heat enters the system faster than it leaves, internal temperature rises. The phone then lowers frequency, voltage, or power to protect its components and manage surface temperature.

This is why the 447.9 words-per-second peak does not describe the entire experience. It captures what the processor achieved before the thermal system reached equilibrium.

The later figure above 320 words per second is more relevant to a long transcription. It reflects the phone’s ability to balance computation with heat dissipation.

That distinction matters as phones handle more work locally. Apple increasingly assigns transcription, image processing, language features, and other machine-learning operations to on-device hardware.

Local processing can reduce network dependence and keep some data on the device. It also converts electrical energy into heat inside a small, sealed enclosure.

A short AI request may never reveal the limit. Processing a long recording, generating repeated outputs, or running multiple tasks can push the device into its sustained range.

Developers should therefore treat peak throughput as only one part of application performance. Batch size, execution time, ambient temperature, and device state can all change the user’s result.

A transcription application offers a particularly useful example. Someone processing a brief voice memo may experience the initial high speed and never notice a slowdown.

A journalist, researcher, or student processing an hour-long recording will spend more time in the throttled portion. Sustained throughput then becomes the meaningful number.

That does not make the iPhone 18 Pro slow. Birchler described the A20 Pro as delivering excellent transcription performance, including a substantial peak increase over the previous generation.

The issue is the gap between maximum and repeatable performance. The cold-can test suggests that cooling, rather than compute capability alone, defines part of that gap.

Laptop computers provide a useful contrast. Birchler reported that his M4 Pro MacBook Pro remained within a consistent performance range during the same type of testing.

A laptop has more internal volume, a larger heat-spreading structure, and cooling hardware designed for longer workloads. A phone must operate within tighter limits on thickness, weight, noise, and skin temperature.

The comparison is not about which device wins a benchmark. It shows why the same processor family can behave differently inside different physical designs.

For mobile software teams, this changes how performance testing should work. One cold run cannot represent an extended user session.

Teams should test repeated operations, warm-device conditions, charging states, and realistic ambient temperatures. They should also record completion time, not only the fastest observed throughput.

Workload scheduling can help. An application might divide a large job into smaller segments or pause nonessential background work when thermal pressure rises.

Those choices involve tradeoffs. Slower pacing can preserve responsiveness and battery life, but it can also increase the time before results become available.

Users organizing recordings and transcripts can preserve their source material in a searchable knowledge base. That workflow remains useful regardless of which device performs the initial processing.

The broader lesson is straightforward. Mobile AI performance depends on the entire device, including cooling, software scheduling, memory movement, and power management.

A20 Pro sustained performance cannot be inferred from one peak score. The La Croix result makes that limitation unusually difficult to overlook.

iPhone 18 Pro La Croix Cooling Tests Apple’s Thermal Redesign

Apple expanded the cooling system substantially, yet Birchler’s result shows that a larger vapor chamber cannot repeal the physics of a compact phone.

Apple introduced the iPhone 18 Pro with a redesigned thermal architecture around the A20 Pro. The chip uses new packaging that places memory beside the processor instead of stacking it above.

That arrangement removes memory from the primary thermal path. It also lets the processor connect more directly to the phone’s cooling assembly.

Apple says the vapor chamber has three times the surface area of the previous design. The company also claims up to 40 percent better sustained performance than the preceding generation.

Those figures should not be confused with Birchler’s 25 percent result. Apple compared generations under its own testing, while Birchler compared one phone with and without improvised external cooling.

Apple’s claim concerns the redesign’s improvement over an older device. The enthusiast test asks whether the new phone still has additional performance available when cooling improves.

Both propositions can be true. A much better thermal system can raise sustained speed while still leaving the processor thermally constrained during a heavy workload.

A vapor chamber is a sealed heat spreader containing a small amount of working fluid. Liquid evaporates near a hot area and condenses at cooler surfaces.

A wick then returns the liquid through capillary action. This cycle distributes concentrated heat across a larger area that can release it more effectively.

The vapor chamber does not make heat disappear. It moves heat from the processor toward the enclosure, where it must eventually enter the surrounding environment.

Independent teardown evidence supports Apple’s description of the design. An iPhone teardown found the A20 Pro on the outside of the logic-board assembly.

The teardown also identified a conformable thermal interface between the processor and cooling system. That material fills microscopic gaps that would otherwise impede heat transfer.

The chamber extends across much of the phone’s body. Apple’s aluminum unibody then provides another path for spreading and releasing heat.

This design helps explain Birchler’s observations. A more effective internal heat path can move energy away from the A20 Pro quickly, but it may also make the enclosure feel warmer.

A warm exterior does not automatically indicate failed cooling. It can indicate that heat is successfully moving away from the processor.

However, the enclosure still has practical temperature limits. Users must be able to hold the phone, while the battery and other components need protection.

The cold can effectively added another heat sink outside the phone. It absorbed energy from the enclosure and maintained a larger temperature difference across the thermal path.

That difference accelerated heat transfer. The processor could then use more power before reaching the same internal temperature limit.

This is the central reversal in the story. Apple’s improved heat transfer makes the phone better at reaching its enclosure, but the enclosure remains the final bottleneck.

Once heat reaches the outer surface, passive cooling depends heavily on ambient air, hand contact, cases, direct sunlight, and surrounding temperature.

A refrigerated can changes that final stage dramatically. It provides a cooler object with more thermal mass than still air.

The experiment therefore does not invalidate Apple’s redesigned vapor chamber. It shows the chamber working within a system whose last step can still be improved externally.

Gaming phones address this problem more aggressively. Some use larger vapor chambers, airflow systems, clip-on fans, or chassis designs that prioritize sustained performance.

Apple balances other goals, including size, weight, battery capacity, durability, and silent operation. Those priorities limit how much continuous heat an iPhone can reject.

The A20 Pro increases the importance of that balance. Faster compute hardware creates value only when applications can use it within the phone’s power and thermal boundaries.

Apple’s product announcement presents the chip and next-generation vapor chamber as one integrated performance system. Birchler’s test reinforces that framing, although not entirely in Apple’s favor.

The processor’s sustained output is inseparable from its cooling path. A cold beverage revealing another performance tier makes that relationship visible to ordinary users.

What the 25% Figure Does Not Establish

One striking result cannot show that every iPhone 18 Pro gains 25 percent from external cooling, or that users should copy the setup.

Birchler’s benchmark has several strengths. He used the same phone, application, model, and built-in test file across repeated runs.

The workload also ran long enough to expose performance after the device became warm. That makes it more informative than a single short benchmark pass.

However, several uncontrolled variables remain. The published posts do not specify the room temperature, beverage temperature, contact area, case configuration, or exact starting temperature.

The first test occurred shortly after the phone had been configured. Birchler noted that indexing and other setup work might still have been running.

He believed intensive work probably paused those background processes, but that possibility was not independently measured. Even small differences could affect a short comparison.

The two headline percentages also require care. A roughly 25 percent increase in settled throughput does not mean every part of the workload became 25 percent faster.

Birchler’s reported completion improvement was 14 percent. The early uncooled runs had already operated near a higher speed, reducing the difference across the full job.

The benchmark tests Apple’s transcription model inside one application. It does not directly measure GPU-heavy gaming, video rendering, camera processing, or other neural workloads.

Different tasks activate different parts of the system. They also produce distinct patterns of memory traffic, processor utilization, and power consumption.

A game may alternate between CPU and GPU pressure. A video export can stress media engines, while transcription may emphasize neural processing and memory bandwidth.

Cooling improvements could affect each task differently. Some applications may hit software, storage, or memory limits before thermal conditions dominate.

The cold can also introduces practical risks that the benchmark did not evaluate. A refrigerated object can produce condensation when humid air contacts its surface.

Water resistance does not make unnecessary moisture exposure advisable. Condensation near ports, damaged seals, or accessories creates avoidable uncertainty.

The can can also slide or apply uneven pressure to the camera area. Users should not interpret the test as an endorsed cooling procedure.

Dedicated phone coolers offer more predictable contact and mounting. Even those accessories can create condensation if they lower a surface below the surrounding air’s dew point.

Battery temperature deserves particular caution. Batteries have operating ranges that differ from a processor’s preference for lower temperatures.

An external cooler applied without temperature control might create an uneven thermal environment. The benchmark provides no battery temperature data.

There is also no evidence that La Croix has any special cooling property. Birchler explicitly suggested that another cold beverage would probably work similarly.

The relevant variables are temperature, thermal mass, contact, and material. The brand supplied a memorable object, not a proprietary thermal mechanism.

The result has not been independently reproduced under standardized conditions. A stronger evaluation would use several devices and multiple ambient temperatures.

It would record processor power, surface temperature, internal thermal state, and task duration. It would also compare passive operation with controlled external cooling.

Repeated runs should begin from the same battery level and device temperature. Researchers would need to account for background tasks and software versions.

Such testing could determine whether the 25 percent sustained difference repeats consistently. It could also show how long the improvement lasts as the external heat sink warms.

The most defensible interpretation remains narrow. External cooling materially improved one iPhone 18 Pro transcription benchmark under Birchler’s reported conditions.

That is enough to establish an interesting performance clue. It is not enough to promise a universal gain or recommend balancing drinks on expensive electronics.

Three Signals Will Show Whether Cooling Becomes the A20 Pro’s Defining Limit

The next evidence must show whether Birchler found a repeatable device constraint or only an entertaining edge case.

The first signal is independent sustained testing across multiple applications. Reviewers should compare cold-start results with performance after ten or more minutes.

Transcription should remain part of that work because it produced the original observation. Games, video exports, and continuous local AI tasks should broaden the picture.

If several workloads show large gains from controlled external cooling, the case becomes stronger. Cooling would then represent a general A20 Pro sustained performance constraint.

If only Quick Subtitles shows the effect, application behavior or its specific model may explain more of the result. The broader conclusion would weaken.

Reviewers should also separate the iPhone 18 Pro from the larger Pro Max. Different enclosure area and battery capacity can change how each model handles sustained heat.

The second signal is performance across realistic environmental conditions. A cool office, a warm car, and direct outdoor heat create very different thermal budgets.

Passive cooling works best when the surrounding air remains much cooler than the device. As ambient temperature rises, the phone loses that advantage.

If sustained speed drops sharply in warmer environments, Apple’s improved internal cooling still faces a strong external limit. That would reinforce Birchler’s central finding.

If the redesigned system maintains similar throughput across temperatures, the cold-can result becomes less representative of normal use.

Charging conditions should be included because charging adds heat. A user might transcribe a long recording or play a game while connected to power.

Cases also deserve testing. A case can protect the phone but change surface airflow and heat transfer, depending on its materials and construction.

The third signal is software behavior during prolonged local AI work. Apple and developers can tune scheduling, model execution, and power management without changing hardware.

An operating-system update might reduce peak power slightly to preserve consistent output. That could lower the first benchmark score while improving total completion time.

Developers might also expose thermal-aware modes. One mode could prioritize maximum immediate speed, while another favors predictable performance and battery efficiency.

If software updates narrow the decline between early and late runs, the cold-can advantage should decrease. That would weaken cooling as the dominant explanation.

If the same gap persists after software matures, hardware heat rejection becomes the clearer boundary. Accessories designed for controlled cooling would then have a stronger performance case.

Users should watch completion time rather than chase the highest instantaneous number. The faster device is the one that finishes the real task reliably.

That principle applies beyond transcription. A phone used for long video projects, local models, or extended gaming must sustain useful performance after the first minutes.

The iPhone 18 Pro La Croix cooling test turns that technical distinction into an image anyone can understand. A cold can unlocked performance that the processor already possessed.

Yet the responsible conclusion is not to refrigerate a drink for every demanding task. It is to evaluate mobile hardware under the conditions where the work actually happens.

For developers, that means testing warm devices and repeated runs. For buyers, it means reading sustained results alongside peak benchmark charts.

For Apple, the question is whether future software can narrow the gap without sacrificing responsiveness. The alternative is a larger thermal budget in later hardware.

Watch the next controlled A20 Pro sustained performance tests closely. If external cooling keeps producing double-digit gains, thermal design will remain central to the iPhone’s mobile AI story.