Hacker News Reader

Comment by mokus | Hacker News Reader

If the higher clock speeds are truly never used for anything but benchmarks then that's downright dishonest. But is that established to be the case? Perhaps those speeds are available for extremely heavy loads under specific circumstances (wired power, etc) that are rare but can actually occur. If they are, then this does seem like a fairly reasonable thing to do. Still mildly shady, but no more so than the norm when it comes to benchmark optimization.

View on HN

The AnandTech link in the article says they found the string "BenchmarkBooster" in the apk alongside the class names of several well-known benchmarks.

http://anandtech.com/show/7187/looking-at-cpugpu-benchmark-o...

The fact that they found a code path that ensures particular programs get the full hardware speed does not mean that no other code path enables that performance level. I'm asking whether it is established that there is no other way that performance level is activated, as it does not seem clear to me from the article that that is the case.

It's pretty obvious that, at the very least, these specific benchmarking apps are given a hugely unfair advantage, and probably only based on internal app ID strings. Apparently the frequencies were boosted for one of the benchmarking apps even if it was just sitting idly showing a menu or something.

Yea, that much is clear. And the AnandTech article (which probably should be the one on the HN front page if it hasn't popped up there since I last looked) is clearer about some of the things they ran to establish that 480 MHz is the norm. I still wasn't sure from that article, though, whether the games and such were run with the device connected to power, whether there are other user-tunable power saving settings that could affect it, etc.

My only point is that I personally feel that more investigation is needed in that direction before I'll feel very negatively about the story. If it's a rarely used but possible mode, it seems like it's on the same level as disabling power saving mode when running a recognized benchmark. I kinda expect benchmarks to represent the peak attainable performance, even if it's not the typical performance.

I find it extremely shady to read that "glbenchmark 2.7" is unable to exceed 480mhz while "glbenchmark 2.5.1" goes into 532mhz even when idle.

My reading of the AnandTech article was that the CPU was boosted to max speed for the benchmarks, but the GPU was over-clocked past what normal apps had access to.

>The fact that they found a code path that ensures particular programs get the full hardware speed does not mean that no other code path enables that performance level.

When those particular programs are benchmarks and the path is coded specifically to boost performance on a unique set of said benchmarks in contrast to the most common use cases, that's an issue regardless of whether or not you can imagine an outlier real-world scenario that utilizes a code path that tweaks the performance.

You have to bend over backwards to see that as anything other than dishonest.

> You have to bend over backwards to see that as anything other than dishonest.

"Our new hot device has this awesome performance mode but those benchmarks aren't aware for it and it won't be utilized in the device reviews. So let's make sure it's used in these benchmark applications."

If they just flip the switches available to every developer/application it's not necessarily dishonest just tech marketing.

1. If this awesome performance mode is turned on automatically when the app needs it. Then it must have turn on automatically for the benchmark without actually checking the application name, right?

2. Otherwise, if it require the application to be aware to utilize this feature, then why hasn't Samsung announce this "mode" to developer to utilize it yet?

A high-performant mode would be a feature that would be marketed and lauded by the OEM, not something hidden from view and hardcoded only to benchmark testers.

Also, benchmarks are designed to compare apples to apples, so if, as you suggest, the high-performant mode wouldn't be triggered by benchmark use-cases normally, then hard coding the exception is dishonest, as the only explanation is turning your apple into an orange.

The processor/GPUs do absolutely scale by load, however in this case they're scaling for specific apps, presumably because scaling that high for any other use (even where the throughput would be beneficial) would yield too hot of a device/kill battery.

It is incredibly dishonest, and there is simply no rationalizing it -- they are cheating, and everyone involved with this is kind of a scumbag.

You can certainly try to dig your head into the sand.

Reply on news.ycombinator.com