facebook rss twitter

Samsung's Galaxy S4 features a "BenchmarkBooster"

by Mark Tyson on 31 July 2013, 10:30

Tags: Samsung (005935.KS), PC

Quick Link: HEXUS.net/qabzav

Add to My Vault: x

Investigations by tech site Anandtech have revealed that Samsung has implemented a "BenchmarkBooster" power scheme into its Galaxy S4 smartphones. On the Exynos 5 Octa chip equipped S4 both the CPU and GPU were boosted for certain benchmark programs, on a Qualcomm powered S4 the same boost behaviour is exhibited on the CPU side of things. As an example of what is happening; even when running the most demanding games titles the Exynos GPU never goes faster than 480MHz, yet fire up certain popular 'named' benchmarks and the GPU clock hits 533MHz.

GPU boosting

Anandtech reports that AnTuTu, Quadrant and GLBenchmark 2.5.1 benchmark programs all trigger the Exynos 5 Octa powered Samsung Galaxy S4 to use a 532MHz GPU clock which is not available to any other applications. The successor to GLBenchmark 2.5.1, called GFXBench 2.7.0, doesn't trigger this MHz boost, suggesting that Samsung has software on the S4 which watches for certain program strings. When comparing the GPU triangle throughput results on the same smartphone using these two benchmarks Anandtech saw the boosted device (running GLBenchmark 2.5.1) had a 14 per cent performance increase in triangle throughput performance.

CPU boosting

Using the Android System Monitor Anandtech found similar boosting behaviour affecting the Samsung Galaxy S4 CPU. Using the Exynos 5 Octa handset and firing up the GLBenchmark 2.5.1 caused the big.LITTLE configuration Exynos to switch to the more powerful ARM Cortex A15 cores at 1.2GHz and never drop below that. Again, using the newer version of that benchmark, the lower powered A7 cores could run at 500MHz while no large CPU grunt was required.

The CPU boosting is also present on the Qualcomm variant of the Samsung Galaxy S4, but on the Snapdragon the GPU isn't 'fiddled with'.

BenchmarkBooster strings in an Android APK

Digging around in the APKs on the Samsung Galaxy S4 Anandtech found that 'TwDVFSApp.apk' contained references to a BenchmarkBooster listing (picture above) with minimum clock speed settings and strings listing popular Android benchmarking programs as pictured below:

The conclusion is that Samsung is optimising the Galaxy S4 performance to run repeatable optimal results in popular benchmarks. Anandtech suggests that Samsung either allow these boosted settings for all users/apps or remove them and also stop wasting time on optimising for benchmarks and just optimise for user experience. It will be interesting to see any official response from Samsung about this sneaky behaviour.

UPDATE:

Samsung have issued a statement denying shady behaviour. The firm said that "The maximum GPU frequencies for the GALAXY S4 have been varied to provide optimal user experience for our customers, and were not intended to improve certain benchmark results." Apps such as S Browser, Gallery, Camera and Video Player also run at the maximum GPU frequency of 533MHz for best performance, according to the South Korean electronics giant.



HEXUS Forums :: 28 Comments

Login with Forum Account

Don't have an account? Register today!
The conclusion is that Samsung is optimising the Galaxy S4 performance to run repeatable optimal results in popular benchmarks. Anandtech suggests that Samsung either allow these boosted settings for all users/apps or remove them and also stop wasting time on optimising for benchmarks and just optimise for user experience. It will be interesting to see any official response from Samsung about this sneaky behaviour.
Utterly agree with Anandtech - although surely the overclock that they did would be achievable via certain apps? Does anyone - other than hardcore mobile nerds - play any attention to these benchmarks anyway? I'd consider myself fairly technically literate but I've no idea what scores my Galaxy S3 gets, and they played absolutely no part in my buying decision.

Although they're not important, the fact that Samsung apparently saw the need to “game” the benchmarks stinks to me. At this rate my S3 will not be replaced by an S5 - might defect to Nokia or Sony, especially if/when the latter actually starts putting an up to date version of Android on their Xperias.
That's the point though, certain benchmarking apps are explicitly whitelisted by Samsung to use the full clocks, if it's not whitelisted you get the lower clocks.

Like you say, the fact they've spent effort on deceiving people is bad enough by itself - they obviously saw the benchmark scores as an important enough.

Also, it doesn't look like the Snapdragon version actually clocks higher than usual, rather it plugs all cores and holds them at their normal max clocks.

Anyway, hopefully the media kick up enough of a stink about this to prompt a resolution from Samsung, and discourage other MFRs from doing the same. Again though, as we say with the recent skewed Antutu results when comparing between architectures, you can't ultimately trust a couple of synthetic benchmarks!
Shame on you, Samsung.
Tee Hee
crossy
I've no idea what scores my Galaxy S3 gets, and they played absolutely no part in my buying decision.

shhhhhhhh! How can you claim the cost of your handset back from them in a class action if you say that? :mrgreen:

I guess it's the same as most/all big companies - you can't trust them. I'd be interested to see whether they have referred to any of the benchmarks in their marketing. I doubt this kind of thing would creep into trade descriptions etc… by itself (although it's obviously designed to mislead), but it might be taken more seriously if they also advertised the results obtainable, claimed it scored higher than other phones etc…