Samsung boosts graphics in Exynos Galaxy S4 to manipulate benchmark scores
What’s more - synthetic benchmarks are often taken at face value without further looking into the results. That’s what benchmarks are about after all, a quick and easy way to tell how powerful a device is compared to others.
Now however digging deeper into the Galaxy S4 reveals some very interesting things happening under the surface.
This applies to the Exynos 5 Octa based international version of the Samsung Galaxy S4 and its graphical performance in apps like games.
It turns out that while generally the PowerVR SGX 544MP3 graphics unit runs at 480MHz, whenever you start benchmarking applications like AnTuTu, Quadrant and GLBenchmark 2.5.1 Samsung unlocks an even higher clock speed of around 533MHz.
This gives the device an artificial boost that only applies to benchmarks, but does not otherwise appear in daily use and for games. That difference is a solid 10% or in some cases even more. So next time when you fire up those benchmarks, take the results with some grain of salt. Hit the source link below for a detailed study of how the Galaxy S4 manipulates benchmarks.
11. Nathan_ingx (Posts: 3401; Member since: 07 Mar 2012)
Let us see what the fanboys have to say about this now...
"The GPU was wrong"!?!
Or maybe "Anandtech couldn't be trusted".
14. mobi_user (Posts: 144; Member since: 18 Jun 2013)
I believe if it was Apple or Nokia, there would be 120 comments by now, instead of only 12
60. sgtdisturbed47 (Posts: 311; Member since: 02 Feb 2012)
In a Smartphone world that boasts specs above reliability, this is no surprise.
74. tedkord (Posts: 8323; Member since: 17 Jun 2009)
And sixty of them would be making excuses for Apple, or calling Samsung plastic or saying that Google sells your bank account password.
29. rd_nest (Posts: 1391; Member since: 06 Jun 2010)
Anandtech didn't discover this. Anandtech simply is taking the material from another source. Stop hyping anandtech as if they are some sort of demi-god.
Brian Klug was informed by a well-known Exynos developer AndreiLux about this. This has been in the forums since a long time.
As usual PA has to find this long only after Anandtech "DISCOVERED" this.
42. rd_nest (Posts: 1391; Member since: 06 Jun 2010)
Because uneducated people think Anandtech is the "ONLY" tech site worth reading. The hype around AT is just too much. In reality they are just another review site. These guys are also mere journalists.
BTW, AT's bias towards Intel/Apple is well-known. There was a time when only Apple products used to get full-blown review.
47. jove39 (Posts: 1588; Member since: 18 Oct 2011)
It certainly is not only review site...but it is one of very best...their reviews are heavily detailed...and I haven't seen any bias lately...take it easy and have fun!
36. Nathan_ingx (Posts: 3401; Member since: 07 Mar 2012)
You missed my point...well, the plane flew away. Will come back to you soon as you understand what i meant.
Meanwhile, be patient.
46. rd_nest (Posts: 1391; Member since: 06 Jun 2010)
It doesn't take me long to understand what it meant. Maybe it might be a news for you, but I am a long time member of b3d. I don't need AT to tell me something "new".
If your knowledge on this subject is limited to what Anandtech told you, then you are not worth discussing.
You may find Anandtech as "cool", but for me I find it just average/above-average.
For someone who makes a living on enterprise (read data center) systems performance analysis, I do a lot of benchmarking than most folks around here. And those benchmarks are not for small-petty mobiles, but for multi-PB systems.
50. Nathan_ingx (Posts: 3401; Member since: 07 Mar 2012)
Where did i even say Anandtech is the best or i worshipped it?
It was just a reference since it was a part of the article...and what fanboys usually do.
Spew your bitter-hate somewhere else man! Sheesh.
57. rd_nest (Posts: 1391; Member since: 06 Jun 2010)
I always knew your knowledge is always limited to what someone else (AT in this present case) says.
Waste of time for me to debate with you.
62. rd_nest (Posts: 1391; Member since: 06 Jun 2010)
Not my fault if you didn't receive proper education. You should ask yourself.
64. Nathan_ingx (Posts: 3401; Member since: 07 Mar 2012)
What the hell is your problem?
If you're oh so high up there, would you mind EDUCATING us 'low life'?
48. medicci37 (Posts: 1184; Member since: 19 Nov 2011)
You sound mad. Lol the article isn't about Anandtech. Was a fanboy
49. rd_nest (Posts: 1391; Member since: 06 Jun 2010)
Yea, I feel surrounded by uneducated people like yourself. Best part is that I can't even have a technical discussion with you. You are just too ignorant to discuss anything. Only thing you can do is troll.
Maybe if the population was bit wiser, it would have been a much better discussion.
Why don't you do some studies in meanwhile?
53. AppleHateBoy (unregistered)
It's funny that PA didn't mention anything about the CPU boost that is applied to benchmarks on both the Snapdragon and Exynos versions. Also the thermal limit too is lifted by about 10 °C to improve performance.
But I guess Samsung is not alone. Nvidia's so-called "Tegra Optimized Games" have been doing something like this for quite a while. These games only apply the max. setting if they see Tegra GPU. But it only takes root access and some file modification to get the details to the max.
I guess now I understand why Nexuses score so low in benchmarks and yet their real life performance is great.
18. rf1975 (Posts: 254; Member since: 01 Aug 2011)
This called Shamesung Innovation. They thing that they can fool all people by add and paid reviewers.
33. jdoee100 (Posts: 334; Member since: 04 Jun 2013)
Samsung shouldn't be doing this, but they weren't the first ones to do it and won't be the last one. Moreover, it's interesting to see that when intel "cheated" with their AnTuTu test, nobody called them cheaters. Bias much?
Plus, the epitome of sports car makers, Ferrari, cheats as well. Ferrari do not let car magazines test their cars without optimizing their cars(changing suspension settings,etc) on specific test surface to be tested.
37. jdoee100 (Posts: 334; Member since: 04 Jun 2013)
So, don't forget to call Ferrari, Intel, Nvidia, etc,,,,cheats as well. Samsung is in good company.
38. akki20892 (Posts: 3853; Member since: 04 Feb 2013)
oh god how many gs4 are there and how many different type of chipset for that, oh god i'm so confuse......!!!
2. ajac09 (Posts: 1477; Member since: 30 Sep 2009)
your just now realizing benchmarks dont matter?
9. _Bone_ (Posts: 2151; Member since: 29 Oct 2012)
The benchmarks that simulate real-world performance absolutely do matter, question is: do you know how to read them?
17. Nathan_ingx (Posts: 3401; Member since: 07 Mar 2012)
The real question is: Do you know how to manipulate them? :P
3. jove39 (Posts: 1588; Member since: 18 Oct 2011)
Well...this is shameful samsung...people reported this earlier...
And now anand and brian have confirmed !!!
63. giani (Posts: 269; Member since: 13 Nov 2012)
Mate, 'Sammy plastic fan club' is hurt. I mean how much can a plastic fan take these days ?! I used to think that isheep's are the worst. But then Nokia fanboys came along....and just when u thought u seen them all..these new breed of fanboyism..I mean I think they are by far the most pathetic human beings. And their beloved Sammy just makes it so hard for them..to belive in the plastic cause.
exynos lagging in benchmarks - statement " no benchmarks are not everything real world performance is everything"
Exynos gets better " yeah number one the most powerful machine benchmarks are everything"
Htc pulls the pr stunt with htc one outside s4 lunch " Ohhh how could they totally uncool and unacceptable"
Sammy hires students to bitch about competition on forums " Aaa yeah brilliant marketing"
Sony Htc unveal quite a large number of phones during 2012 " ahhh the confusion..bla bla how can they bankruptcy "
Sammy releases 20 models starting with s4 " yeah we love it Sammy gives us choice"
Htc does the + cycle 2 years in a row " ohh bankrupt..how could they the betrayal"
Sammy does the same this year " Aaa nothing wrong here move along"
And many others. Bottom of line ?! The most double faced brain washed (if there even was one to wash) pathetic fan boys out there. The Sammy boys.
65. WHoyton1 (Posts: 1635; Member since: 21 Feb 2013)
How long did it take for you to write that: 45mins
How many f*cks anyone gives: 0
Get the picture.....
84. giani (Posts: 269; Member since: 13 Nov 2012)
Which part got u annoyed...the one where I said yr pathetic or the one where I said u don't have a brain ?!
86. giani (Posts: 269; Member since: 13 Nov 2012)
If what you said it took me 45 min to write 200 words is based on yr own capabilities..then don't bother answering because I know.
5. darkvadervip (Posts: 365; Member since: 08 Dec 2010)
Now I will say that's innovation of being a cheater. Lol
7. EXkurogane (Posts: 863; Member since: 07 Mar 2013)
Cheap plastic, and now cheap tricks. What's next?
8. alpinejason (Posts: 262; Member since: 06 Sep 2011)
Does not surprise me Samsung has never been innovative. Alwaysa company who follows others
10. XPERIA-KNIGHT (unregistered)
Exactly...I swear I'm jus waiting for Sony and HTC to pick up their marketing game so they can finnally "do away" with this cheap and fraud of a company.
26. PapaSmurf (Posts: 10226; Member since: 14 May 2012)
HTC? The company who produced a high end flagship that's still dying?
And technically, Samsung didn't cheat. In the specs of their phone, they list the GPU clocked at 533MHz.
39. Berzerk000 (Posts: 4239; Member since: 26 Jun 2011)
I don't think you're getting the point. The thing people are getting flustered about is that it's only 533 MHz in benchmarks, otherwise, it's 480 MHz. That means that Samsung intentionally boosted the clock speed in benchmarks to try and seem like they have an edge on the competition in terms of performance.
It makes them look good to anyone who might be looking at those benchmarks, but in reality you don't get any sort of performance boost in real use.
They used irregular techniques (overclocking in benchmarks) to give them an advantage, that's cheating. It's like comparing the Droid DNA to the Xperia Z, but overclocking the Xperia Z to 1.7 GHz during the benchmarks and not telling anyone, which gives Sony the "edge" in performance when (theoretically) they both should be equal.
41. PapaSmurf (Posts: 10226; Member since: 14 May 2012)
When running a benchmark test, it's smart to run said GPU/CPU to the fullest extent to see how powerful the chip is. Therefore, I don't see the issue here.
On a side note, I can't believe the amount of people bashing Samsung. Where were they when Intel had unbelievable results due to Turbo Boost?
44. Berzerk000 (Posts: 4239; Member since: 26 Jun 2011)
But the point of a benchmark is to see how the device will perform under daily use. If the GPU is underclocked outside of benchmarks, that means the benchmarks aren't entirely accurate and consumers are getting slightly lower performance than what the benchmarks indicate, it's misleading consumers. That's what the problem is.
From my understanding, Turbo Boost does overclock the CPU, but it's affects can be used in any task, not just in benchmarks. With Turbo Boost, you get above average performance all around when it is required, but with what Samsung is doing, it's only in benchmarks. Correct me if I'm wrong though.
75. PapaSmurf (Posts: 10226; Member since: 14 May 2012)
If you call me a fanboy then you obviously must be new here.
20. PapaSmurf (Posts: 10226; Member since: 14 May 2012)
Your comment made lol. Tell me again how LG's UI is a terrible knock off of Touchwiz Nature UX?
70. thelegend6657 (unregistered)
Before you forgot you shames**t fanboy that will defend Samsung till the death , LGUI 3.0 got unveiled on the Optimus 4xHD before Samsung unveil the Galaxy crap 3 with its nature UX *inspired by crayons , designed for dogs*
21. MrKoles (Posts: 363; Member since: 20 Jan 2013)
So, you say that Samsung suddenly became not innovative because they boosted the GPU with f*ckin´ 53 MHz?
12. rantao333 (Posts: 319; Member since: 21 May 2013)
look at how anand said
"...The risk of doing nothing is that we end up in an arms race between all of the SoC and device makers where non-insignificant amounts of time and engineering effort is spent on gaming the benchmarks rather than improving user experience. Optimizing for user experience is all that’s necessary, good benchmarks benefit indirectly - those that don’t will eventually become irrelevant."
sammy fans still brag about benchmarks nowadays... while the others realize user experienced is more important..
13. yoavst (Posts: 17; Member since: 29 Jul 2013)
it called benchmark optimization - set the CPU and GPU on their Higher clock rate.
it is legal, cuz the device didn't cheat, just run at full power (even not full power with this exynos).
So stop say that samsung cheated, she didn't do something ilegal
27. alterecho (Posts: 1097; Member since: 23 Feb 2012)
"benchmark optimization" or cheating.
It IS cheating if it has an unrealistic advantage using artificial techniques, in benchmarks alone, over the competition, but performs more or less the same as the competition in real usage. Its fooling the users into believing that it performs better than the competition.
30. rf1975 (Posts: 254; Member since: 01 Aug 2011)
Looks like you are one of those trolls payed by Samsung.
15. ianbbaa (Posts: 305; Member since: 20 Mar 2013)
BAD written article dear PA.
...PowerVr 544mp3 graphic unit on S4 is made for 533mhz...it is also stated in specs everywehere...
...it runs antutu and other bench apps on 533 - is there smthng wrong?
..when it runs on 480mhz in normal usage, it doesnt mean that they are cheating - they just downclocked the gpu to save more battery for normal usage.
According specs of the gpu unit, they do not CHEAt anyone, they just downclock everywhere expect antutu...am i right???
24. Nathan_ingx (Posts: 3401; Member since: 07 Mar 2012)
"This seems to be purely an optimization to produce repeatable (and high) results in CPU tests, and deliver the highest possible GPU performance benchmarks."
31. techspace (Posts: 965; Member since: 03 Sep 2012)
downclocking it everywhere else except on antutu and/or the other the other benchmark apps is cheating...because that would make the results inaccurate,people will think that the phone is equally fast with the other apps and games...
this is the reason why benchmarks people should stop looking at the specs and benchmarks...experience is more important.
35. techspace (Posts: 965; Member since: 03 Sep 2012)
"this is the reason why benchmarks are not as useful as they appear to be"(correction)
16. pwnarena (Posts: 1122; Member since: 15 Feb 2013)
Hey Victor, I don't think this statement is accurate:
"The Samsung Galaxy S4 is the best selling Android smartphone of all times"
The S3 is still the best-selling Android smartphone. As of the latest Samsung announcement, the S4 only managed to sell 23 million units. I think the more accurate way to put it is "the FASTEST-SELLING Android smartphone of all time." We still don't know if it beats the S3 record.
22. PapaSmurf (Posts: 10226; Member since: 14 May 2012)
Galaxy S4 sold 20 million in two months. GS3 sold 10. Therefore statement is still valid.
45. neutralguy (Posts: 1152; Member since: 30 Apr 2012)
Actually, saying best all times means it has the best number/most number. As of now, s3 is still superior to s4 in terms of number of sales. But that wont be long.
58. pwnarena (Posts: 1122; Member since: 15 Feb 2013)
Well in my experience as a regular consumer of web and print content written in the English language, be "best-selling" generally means having the most number of sales. Think best-selling album, best-selling record, best-selling video game.
I think you need to know the English vocabulary better. Best-selling does not equate to being the fastest-selling. Until S4 sales exceed the total number of S3 sales, it cannot be considered the best-selling yet.
Well, this is a blog after all. There's no such thing as journalistic standards here, apparently. I yield.
83. Victor.H (Posts: 592; Member since: 27 May 2011)
Absolutely, thanks for the correction. I've now fixed this in the article.