[AT] Samsung overclocking Octa-core to get better benchmark performance

[NTMBK]

http://www.anandtech.com/show/7187/looking-at-cpugpu-benchmark-optimizations-galaxy-s-4

Lots of skullduggery going on here. They force the CPU to stay on the A15s at 1.2GHz the moment you fire up a particular named benchmark, disabling core-hopping functionality of the big.LITTLE setup in order to artificially boost scores.

 

[beginner99]

Damn, I just wanted to create this thread and name it:

"Samsungs unethical benchmark scaming"

You know, to give the thread a tabloid-like touch. Which seems to be a trend around here lately just like the sinking quality of thread created by those users.

Back on topic.

Well even I with a little more interest in electronics than the average joe will pick my phone based on form factor / design / usability and price and then maybe because of the SOC.

I did not buy a RAZR I because it has an Intel inside (I heard of the model because of that) but due to it's small size compared to the screen and according to reviews above average battery life especially talk time.

I doubt this trick actually ends of with a net profit. Just imagine all the meetings to get this in. lol. usually a PITA in large companies like Samsung.

 

[ShintaiDK]

LOL! GPU as well it seems.

 

[joshhedge]

That's certainly more than optimisation, it's deception plain and simple as it affects the end user in no beneficial way other than increasing their E-pen size.

 

[Kenmitch]

That's certainly more than optimisation, it's deception plain and simple as it affects the end user in no beneficial way other than increasing their E-pen size.

<---- Currently owns no Samsung devices other than a memory card

Isn't that the whole purpose of the synthetic benchmarks in the first place.

For the most part they are just unrealistic performance numbers that don't relate to everyday use scenarios.

Seems like Samsung is just saving the nerds the hassle of overclocking the chip.

Not a expert here but....Would it be possible for a person to add certain apps/games to the list of the boosted ones?

 

[Arkaign]

Aha! I could say some rather hilarious things regarding some trolls we have here, but I'll refrain.

Anyway, this is an interesting development, and far worse imho than the supposed controversy with Intel's SDP metric, or AMD's ACP.

Samsung will probably fix this now that the cat is out of the bag. What surprises me is that something so blatant was attempted.

Others have wisely noted that benchmarks don't sell tablets/phones, good word of mouth and hype sells tablets/phones. I think battery life handily trumps benchmark scores in most buyer's hearts and minds given the lowly duties that they perform.

And that note leads to the truth that PC's are very far from dead, sure tons of tablets are sold, but they're overwhelmingly used only for simple media consumption or light communication duty, not for content creation or serious work. Even if many typical home users shift entirely to tablet/phone as they already are, there is a huge market that remains for devices capable of real work.

 

[Arkaign]

Isn't that the whole purpose of the synthetic benchmarks in the first place.

For the most part they are just unrealistic performance numbers that don't relate to everyday use scenarios.

Seems like Samsung is just saving the nerds the hassle of overclocking the chip.

No, synthetic benchmarks are not deceptive unless they are taken in a vacuum or examined without a 'big picture' viewpoint as offered in things like full AT reviews.

Having the unit overclock beyond normal levels for a particular benchmark is a deceptive tactic unless that same overclock is applied in equal regular consumer use scenarios (major 3d game, etc).

While I don't think it's something to get the pitchforks out for, it's a bit shady for sure. Others have made a gigantic mountain out of simply re-examining the nature of measuring power consumption in tablet/phone type devices, this is a completely different league.

 

[jvroig]

Damn, I just wanted to create this thread and name it:

"Samsungs unethical benchmark scaming" You know, to give the thread a tabloid-like touch.

Which seems to be a trend around here lately just like the sinking quality of thread created by those users.

The problem with your proposal is that if you did so, you might be compelled to change it, because the source article DID NOT entitle the article as such. Compare with the obvious thread you seem not to be able to get over, the OP merely followed the forum guidelines to the letter: he used the exact title of the article he is quoting, and added a citation on the title with regards to the source. It was clearly by the book. That you don't like the article quoted and discussed in that thread is not the OP's fault. If it makes you feel better, you could find the nastiest article written about AMD, start a thread about it, quote the article title exactly as is and cite the source, and that would also be perfectly acceptable and would be left open no matter how many AMD fanboys report it to the mods.

Therefore, your misplaced aggression here is unnecessary. Please, get over it. The forums cannot be a place where only your preferred point of view is accepted and allowed. If you dislike threads, leave them alone. If said threads are breaking some of our guidelines on decorum, report them. If you have a personal issue with some members, don't do anything that will only demean you. It is simple. Don't make it a crusade to make everyone follow what you deem is correct and acceptable, and don't get into a hissy-fit just because fanboys of the opposite camp refuse to see things your way. These forums are not meant to be a place where everyone has similar opinions and conclusions, and everybody just pats each other on the back. It is meant to be exactly what it is - a free exchange of ideas and opinions, from different points of views, from all possible parties, even fanboys.

Also, meta commentary such as you did here does not belong on the tech forums. Normally, this would be grounds for a mod reminder or warning, but instead I would rather take this opportunity to publicly reiterate the AT Forums policy on related issues, for the benefit of all: We don't police fanboys from any company. In these forums, it is perfectly all right to be a fanboy through-and-through. We let all of you (Intel, NV, AMD, ARM fanboys) play here as long as you follow the expected decorum. Once you cross the line (personal attacks, thread crapping, profanity, etc), you will be reminded by a mod of the forum guideline(s) that you crossed, no matter whose company you side with. If you can't interact with your forum peers without getting emotional and hostile, then perhaps you should limit your participation until such a time that you can keep your emotions in check enough that you can participate without breaking the AT forum guidelines.

If anyone has any further clarifications about forum policies, do not continue it in this thread. Do so in Moderator Discussions, so this thread can continue with no more off-topic commentary. In Mod Discs, you can debate the finer points of forum policy with the mods to your heart's content.

For the reference of everyone concerned. Please be guided accordingly.

 

[JimmiG]

You should always disable power saving anyway when benchmarking. This has always been true for desktop computers and it's no different with mobile devices. There's always the chance of the frequency dropping temporarily for a split second during a period of light load, affecting the benchmark result.. Or in the case of the bigLittle, the entire thing jumping to the A7 cluster and back again.

 

[ChronoReverse]

This is cheating plain and simple.


Well, I'm not on the Octa nor am I running Touchwiz so it doesn't affect me directly.

 

[moonbogg]

Call me crazy, but I cannot for any reason imaginable be caught caring about cell phone performance. That being said, if I was paid a huge amount of money to consider caring about this, then I would have to say they cheated. Why is cell phone performance even on the list of things that barely matter? Really?

 

[AnonymouseUser]

So besides cheating in benchmarks, they are also lying about the S4 ACTIVE being waterproof....

 

[monstercameron]

even though this is for certain programs, how is this different from turbo?

 

[SiliconWars]

Read about this a month ago on B3D. Nobody should be surprised, when it comes to antitrust and general consumer shafting Samsung are only marginally less corrupt than Intel. Should be fun watching these two in direct competition.

 

[joshhedge]

even though this is for certain programs, how is this different from turbo?

Because turbo would work on all applications and would improve real-world performance, this works only on certain benchmarks, serving no purpose to the user. But I can see it being turbo for benchmarks as it raises the core clock of the GPUs.

 

[monstercameron]

Because turbo would work on all applications and would improve real-world performance, this works only on certain benchmarks, serving no purpose to the user.

but what purpose do benchmarks have for the average user?

 

[joshhedge]

but what purpose do benchmarks have for the average user?

None, other than for comparisons sake between kids comparing their phones. Although I use them to compare different ROMs and Kernel combinations to aid my choice.

 

[BallaTheFeared]

So besides cheating in benchmarks, they are also lying about the S4 ACTIVE being waterproof....

The comments section is golden.

 

[NTMBK]

Isn't that the whole purpose of the synthetic benchmarks in the first place.

For the most part they are just unrealistic performance numbers that don't relate to everyday use scenarios.

Seems like Samsung is just saving the nerds the hassle of overclocking the chip.

The whole purpose of a (well designed) benchmarksis to give us a reproducible test which gives us a decent indication of how a chip will perform in day to day use, and lets us fairly compare various different parts. If Samsung are artificially boosting performance in these tests, they are making their test results invalid as they no longer indicate how the device will perform day to day.

 

[NTMBK]

but what purpose do benchmarks have for the average user?

They let them approximate the performance of different parts.

Imagine a world without benchmarks. "Why should I choose this i7 desktop over an Atom nettop? The Atom uses way less power." We need some rough number to indicate how fast a thing goes.

 

[NTMBK]

Call me crazy, but I cannot for any reason imaginable be caught caring about cell phone performance. That being said, if I was paid a huge amount of money to consider caring about this, then I would have to say they cheated. Why is cell phone performance even on the list of things that barely matter? Really?

Because slow-ass crappy computers are annoying.

A cell phone is not just a cell phone, it is a portable web browser, game machine, and general use computer. Would you compromise performance on your desktop?

 

[monstercameron]

The whole purpose of a (well designed) benchmarksis to give us a reproducible test which gives us a decent indication of how a chip will perform in day to day use, and lets us fairly compare various different parts. If Samsung are artificially boosting performance in these tests, they are making their test results invalid as they no longer indicate how the device will perform day to day.

they arent modifying the code of the benchmark nor the os, they are just overclocking[turbo?] the chip while the benchmark is running. They could literally patch the os now to make this permanent.

aside: did they find out if the oc was constant while the benchmarks was running or until the chip ran out of thermal headroom?

 

[NTMBK]

You should always disable power saving anyway when benchmarking. This has always been true for desktop computers and it's no different with mobile devices. There's always the chance of the frequency dropping temporarily for a split second during a period of light load, affecting the benchmark result.. Or in the case of the bigLittle, the entire thing jumping to the A7 cluster and back again.

This is a mobile device- you want to test how well it performs in the way that people will actually use it. If it's going to core-hop while trying to render a web page, or run a game, then that should be showing up in the benchmarks.

 

[joshhedge]

they arent modifying the code of the benchmark nor the os, they are just overclocking[turbo?] the chip while the benchmark is running. They could literally patch the os now to make this permanent.

aside: did they find out if the oc was constant while the benchmarks was running or until the chip ran out of thermal headroom?

Looks like they tested it for 5 runs and the results showed no throttling as the average was within the spread of the results with no drop indicating throttling. So that would insinuate there being enough thermal headroom for the benchmark and hence the OC being constant.

 

[NTMBK]

they arent modifying the code of the benchmark nor the os, they are just overclocking[turbo?] the chip while the benchmark is running. They could literally patch the os now to make this permanent.

aside: did they find out if the oc was constant while the benchmarks was running or until the chip ran out of thermal headroom?

They could do it, yes- but they would ruin their battery life, which would annoy their users. (A lot.) There's a reason those A7 cores are there in the first place. What they have done is make the benchmark not represent how the phone actually performs in real world use.