News - Samsung's Galaxy S4 features a "BenchmarkBooster"

[watercooled]

No, they can be "money grabbing cheaters" as much as they like or don't like. But it's important we know the score so we can take their other claims with the scepticism they now deserve.

Why do you think those benchmark programs are on the shelf in the first place? The point of a benchmark is a standardised test that also saves the reviewers work. If everyone used different benchmarks all the time then you wouldn't be able to compare the results. That makes then a tempting target for cheating.

Exactly, as much as you can't rely on a single synthetic benchmark when comparing devices, benchmarks are ultimately how we compare performance, which can have an impact on buying decisions; whether something is a worthwhile upgrade for example.

Nvidia and amd have done this for years. Who cares?

As mentioned in the Anandtech article, they've actually done the *opposite*, at least for extreme load programs like Furmark, to avoid damage to the card but lowering scores in the process.

I was going to laugh at you for suggesting this, but then I actually thought about it and you might have a point. How would it be if Samsung's next firmware had - hidden in the developer options of course - a set of overclocking settings. Ideally, what would be nice would be some kind of cpu control on a per app basis - so you could upclock for those FPS that you've got installed but conversely downclock for mail apps etc. After all if they can do that for benchmark programs, then surely it's not rocket science to make that feature more useful for those of us who aren't into numerical willy-waving?

Such an app, like a lot of the 'optimisation' or 'battery saving' apps already available, would be mostly pointless. A lot of work already goes into managing SoC clock speeds, core plugging, power gating, etc, depending on load. Plugging and fully clocking all cores would just waste power and heat the SoC unnecessarily, and limiting clock speed would just slow things down unnecessarily, and perhaps also waste power as the cores would have to be powered for longer.

Samsung might only be able to get away with the higher clocks for specific GPU-heavy but CPU-frugal benchmarks because of power/thermal limitations i.e. such a clock may not be feasible during a real game when both CPU and GPU are loaded? Not saying it's right, but it would partly explain the reasoning behind it.

[watercooled]

Update available at Anandtech: http://www.anandtech.com/show/7192/u...ons-galaxy-s-4

Their explanation does make some sense TBH, essentially to prevent overheating in intensive apps like some games. 532MHz might be sustainable for a short GPU-only benchmark, but with CPU cores active in a game, it might end up throttling down even further if those clocks were allowed initially - 480MHz might just be a decent mid-ground.

However, the 'benchmark boost' phrase is still somewhat suspect IMO, as is the whitelist; if all apps have access to the full clocks by default, why is that necessary?

Edit: Just to add, whatever the reasoning, this still ultimately means the affected benchmark results are likely not proportionally representative of game performance. Benchmarks designed to place a realistic load on both CPU and GPU would therefore be more useful for deducing game performance, provided they're not unfairly whitelisted of course.

[KeyboardDemon]

lol. I beat the news, by 7 hours and 12 minutes.

[watercooled]

http://forums.hexus.net/cpus/241925-...ml#post3002845

TBF was just linking Anandtech though.

[zdn1042]

Well, that's disappointing. Waiting for Samsung's response now...

[mtyson]

Samsung's response has been added at the bottom of the story now.

Well, that's disappointing. Waiting for Samsung's response now...

[kingpotnoodle]

Not surprised? Come on who is seriously surprised, this kind of thing has been going on for years in various arenas, it's why artificial benchmarks mostly suck for measuring performance in the real usage patterns of an average user. The only benchmark I could take without a whole pot of salt are scripted sets of actions recorded from user input... and even they would be with a spoonful of mild salt.

Performance in mobile devices is so subjective anyway, how fast a device feels like it runs is far more important, cludgy software can make animations look like 90s Power Rangers special effects on the fastest hardware whilst an artificial benchmark makes it look like aliens passed on technology...

[Saracen]

Well, reading Samsung's explanation and Anandtech's analysis of it, the impression I'm left with is that Samsung are gaming the benchmark process, have a semi-credible excuse/explanation for what they're doing but I'm not buying it, and that the whole thing rather sucks.

That said, as I'm not in the market for a premium smartphone, and unless they start selling for a LOT less than they currently do, I probably never will be, so I don't much care, personally.

I'm typing this on a Samsung Galaxy, but it's a tablet not a phone. I didn't even look at what processor it was before buying, much less benchmark scores, before buying. I played with it in-store, and decided "fast enough", and that was that. I also compared Samsung prices to the iPad prices, noted the £300+ difference, and bought the Samsung.

[watercooled]

Yeah I agree, the excuse is semi-plausible but the whole thing still stinks, they didn't explain the whitelist; their explanation implies they'd need a blacklist to prevent 'intensive' applications from causing problems.

[kingpotnoodle]

Yeah I agree, the excuse is semi-plausible but the whole thing still stinks, they didn't explain the whitelist; their explanation implies they'd need a blacklist to prevent 'intensive' applications from causing problems.

Whitelists are easier to make, generally shorter and easier to maintain with less explosive consequences if something is missed off the list. Running a bit slower is a whole lot less bad than trouser fires.

[watercooled]

But a whitelist ensures an awful lot of applications (probably the vast majority) which could benefit from the higher clocks will not being included, so again the whitelisted benchmarks aren't a fair comparison.

Thermal protection built into the SoC would step in long before anything became dangerous, so that's not a valid reason either.

[mtyson]

Another shady Samsung story has broken. Sounds a bit like more forum/opinion manipulation, you could call it "cash for 'casual and organic' questions". Here it is, over at The Verge: Who's offering developers cash to secretly promote Samsung?

[crossy]

Another shady Samsung story has broken. Sounds a bit like more forum/opinion manipulation, you could call it "cash for 'casual and organic' questions". Here it is, over at The Verge: Who's offering developers cash to secretly promote Samsung?

Hmm, that's what I can't understand - why Samsung feel that they need to resort to "underhand" tricks. Heck, they've already got the lions share of the market and stuff like this - when/if it gets found out - is surely going to create more negative PR than the positive spin they would have got from the operation itself?

Likewise the BenchmarkBooster. Joe Public isn't going to care one whit about AnTuTu scores (if they even know what AnTuTu IS), and the hardcore mobigeek is going to know that something like this is going on, so won't be impressed. Which, to my mind, leaves the techno reporter as the only likely target of this. In which case - like the forum manipulation - this is inevitably going to backfire on them.