Following reports that Samsung throttled performance for several apps and games on its flagship phones (and even throttled recent tablets like the S8 series), other manufacturers have come under scrutiny for their own performance-adjusting behaviors. Testing Xiaomi’s latest lineup with the popular Geekbench benchmarking application, we’ve measured a noticeable performance difference between various “spoofed” versions of the app that were recognized by the system as known apps or games. According to Geekbench, this behavior will result in affected phones being delisted from its benchmarking charts later this week.
We tested several modified versions of the Geekbench application that were meant to trigger any app-identifying performance profiles on the Xiaomi 12 Pro and Xiaomi 12X. We’re still investigating the mechanism behind how Xiaomi’s performance tuning works (and we’ll update if and when we learn more about it), but our preliminary numbers indicate that Xiaomi is giving apps it recognizes as games or benchmarks a better performance profile and throttling performance for other apps. This bears out with differences in benchmark scores for modified versions of the Geekbench app:
|Play Store Geekbench
Specific results of our testing varied between models, but generally indicated that the phone was using multiple performance profiles across apps. Checking two phones from the recent 12 series, the Snapdragon 8 Gen 1-powered 12 Pro is more affected by the performance throttling than the Snapdragon 888-equipped 12X.
Versions of the Geekbench app the system recognizes as either Geekbench or Genshin Impact saw benchmark results with up to 50% better single-core performance on the 12 Pro compared to versions of the app that the system believed were Netflix or Chrome — a substantial performance penalty. Even the 12X was affected to a certain degree, with a similar but smaller difference in results that was nonetheless larger than any expected margin of error.
Interestingly, performance even simply varied when the app identified correctly as Geekbench. The Play Store-distributed version of the app observed higher results than a slightly different corporate version of the app, which was throttled. The Xiaomi 12X also saw a bigger difference between game and benchmark performance, which didn’t appear to be present on the 12 Pro. This could indicate a more complex throttling system, or different performance profiles depending on base hardware or phone model, as chipsets vary between the two devices.
Again, we haven’t been able to dive into the inner technical workings of how Xiaomi’s performance profiling/throttling might work yet, but based on these results, it appears that Xiaomi is simply loading different profiles based on app-identifying details like package names, with games and benchmarking applications getting full performance, while other potentially less demanding apps get a throttled performance profile.
Xiaomi has engaged in similar “optimizations” in the past, with the Xiaomi 11T Pro using different performance profiles for different apps, in that case, adopting a faster-throttling performance profile for apps it believed to be games. Although we did not test “sustained” performance on our devices (we don’t have a set standard for “sustained” testing at Android Police, and we aren’t sure what Anandtech’s standard references might be), the 12-series could have similar behavior.
We have reached out to Xiaomi for more information, but the company did not immediately respond to a list of questions. Geekbench founder and president John Poole provided us with the following statement:
It’s disappointing to see another device manufacturer mislead consumers by reducing application performance but not benchmark performance. We are investigating which Xiaomi handsets are affected and expect to start delisting Xiaomi handsets from the Android Benchmark chart later this week.
A precise schedule wasn’t set, and we don’t know how many of Xiaomi’s phones might have this sort of selective throttling behavior — which Geekbench has previously told us it considers a form of benchmark manipulation — but phones that have been demonstrated to be affected will be removed from the company’s Android Benchmark charts later this week. Previously, Geekbench delisted four generations of Samsung phones for similar behavior. Although a recent update may have adjusted that behavior, Geekbench’s policy to date has been that once a device is delisted, it remains delisted (though that could change). Android Police