Actually, if the benchmark is at all properly written and the DVFS as regulated by the operating system (and user settings) is at all reasonable then this is exactly what should happen when you launch a benchmark. If you're running something that pegs the CPU it should be given as much CPU time as possible within the operating limits of the device.
Anand Lal Shimpi - Tuesday, October 01, 2013 - link
Yep, I believe so (need to confirm with Brian since I don't have the device in front of me) - all modern non GPe Samsung devices (as well as those from other OEMs) do the same manual DVFS setting upon benchmark detect unfortunately.
Nothing new, they also did it with the S4.
Using ICC vs generic compiler is not cheating.
Personally, I think this is a positive change and should be the default behavior for all devices.
It's wrong. No other OEM is ignoring the TDP cap in certain applications.
nVidia caps Tegra 4 in Shield to 1,48GHz with 4 threads in every Apps and you can't force an ignore to the TDP limit.
/edit: And Samsung is doing the same with their own SoCs. So it's more a normall Samsung behaviour.
i cant see the cheating bit, all they did was the equivalent of putting your windows laptop in high performance mode so your laptop wont speedstep or c&q down the clocks when idle. the processor is capable, it isnt operating beyond spec nor were there any software hacks to the benchmark itself.So you think it's okay to get the best performance only in benchmarks but not in every normal workload?! :\
SPersonally, I think this is a positive change and should be the default behavior for all devices. In the past, I recall a time when I ran a benchmark on my Note 2 and got vastly different results each run. Like a 500 score, a 1730 score, an 800 score, etc. Not consistent at all. I'm guessing that was because of the cores under-clocking themselves because they were not properly detecting the load the benchmark was trying to provide. It sounds like this is a direct response to that sort of issue. If it results in more consistent benchmark scores, it sounds great.
i cant see the cheating bit, all they did was the equivalent of putting your windows laptop in high performance mode so your laptop wont speedstep or c&q down the clocks when idle. the processor is capable, it isnt operating beyond spec nor were there any software hacks to the benchmark itself.
also where is the evidence that the processor cant operate at these performance levels when under stress after x amount of time.
This isn't "turbo boost." This isn't "throttling." This is cheating.
The only applications that are allowed to use the higher frequencies are the benchmarks. You browser, your game, app launch, etc... anything you would do on your phone to actually use it rather than test "how fast it is" does not get the extra frequency.
This is the equivalent of giving the review sites devices that run at a 20% higher frequency, seeing glorious reviews about how fast the device is, and then selling that device to the masses at the lower frequency.
The frequency boosting that is being done on PCs is generally applicable for anything the CPU is doing, as long as there are no thermal issues. This means your bowser and your game can use it. You can argue about "how long" or "how much" it can use it, but it's still available. Some people tell you that you need to turn it off in order to see how fast your PC "really" is, since it's hard to make it stop kicking in. That's a great thing for the user, because it's showing that it's generally useful.
An optimization (including one that raises the frequency) that is only available for specific use cases that only the OEM can specify (in this case, Samsung and things they have identified as benchmarks) is clearly a cheat and has absolutely zero value to the person buying the device, other than to mislead them about how "much faster" it will be than their previous device.
There is no similarity to ICC here. In the case of ICC, at least that's a compiler that will attempt to apply the optimizations generically. It's possible that the vast majority of the code doesn't get the optimization applied since the compiler can't figure out how to apply it. I don't think anyone has ever accused ICC of having a "white list" of source files that it handles differently, maps to hand coded assembly, etc. ICC is potentially shady, but not cheating.
There are certainly real-world examples of ICC making code faster that solves real world problems, unlike this Samsumg optimization.
Except you can put your windows laptop in "high performance mode" whenever you want. You can't edit their white-list, and the white-list doesn't have anything in it that would make using you device actually faster.
It's like windows only allowing you to turn on "high performance mode" when you're on the screen asking it to calculate the windows experience score. Completely misleading, and completely useless.
I am not sure where you read about higher freqs but the article clearly shows that during benchmark/high performance mode it doesnt overclock the soc, it just doesnt downclock the soc. It booster doesnt allow any of the cores to go to lower power states.
you also seem to be claiming that samsung doesnt allow other software to have access to the fastest clockspd, which isnt true.
Note that the CPU behavior is different from what we saw on the GPU side however. These CPU frequencies are available for all apps to use, they are simply forced to maximum (and in the case of Snapdragon, all cores are plugged in) in the case of these benchmarks. The 532MHz max GPU frequency on the other hand is only available to these specific benchmarks.
WEI itself is kinda useless but the idea is, in an ideal environment, to get the best score as a comparison point between different machines. so getting your WEI score in a less than ideal condition such as having the cpu throttle during testing would throw off such scores, which would throw off the comparison.
Doesn't that represent the "real" performance of your device and what you'll see day in and day out?
No, not really, because day in and day out I am not running benchmarks.
This is standard practice with PC benchmarking- you turn off all extraneous software, you make sure CPU is fully clocked, turn off power saving functions, etc.
Why is it suddenly cheating when the same practices are used on a phone?
I've heard (but have been unable to verify) that these shenanigans also raise the thermal limits of the processor. So not only will all CPU cores start out at peak frequency when running benchmarks, they'll run at peak frequency for longer (before throttling down due to thermal limits) when running benchmarks.
Apparently browser speed isn't increased by Samsung trick. So browser benchmark scores as found by AT can be trusted and they are good: http://www.anandtech.com/show/7376/samsung-galaxy-note-3-review/4My performance is going to be a function of the network, the temperature in the room, the amount of work my device has been doing (its temperature), the quality of the network drivers, the quality of the DVFS governor, the resolution of the display, etc. Having benchmarks that override ALL of this makes the benchmarks useless.
The reason phones are different than PCs is because they are universally thermally limited, and each form factor is sufficiently different to have different behaviors.
You wouldn't expect a PC to throttle while surfing the web.
Phones, on the other hand, are quite likely to. Once you turn on the 3G/LTE chip, plus GPS for the adds to track where you are, the screen so you can read, there's not always enough budget left over to run the CPU at max frequency.
And when you run benchmarks on PCs, you don't disable turbo, and you don't disable thermal throttling. If you go back to jfpoole's post:
If Intel said "hey, I know I'm running Cinebench" and then disabled thermal throttling and allowed 4 cores to turbo up to the single turbo point, would that be considered cheating within PC benchmarks?
Wasn't "quak3" considered cheating? Of course. This is the same, is it not?
I've heard (but have been unable to verify) that these shenanigans also raise the thermal limits of the processor. So not only will all CPU cores start out at peak frequency when running benchmarks, they'll run at peak frequency for longer (before throttling down due to thermal limits) when running benchmarks.
A PC in sleep mode or with several cores parked will indeed be less responsive at visiting a website than an otherwise identical PC running at full speed.
Are you trying to argue that a PC isn't slowed down by running multiple pieces of software? I beg to differ. I work on computers every day, half of them have multiple browser bars, 5 different programs that auto-start on boot, antivirus, chat programs, drop box and/or the multitude of clones, etc. They run like crap.
On the other hand, if you follow normal bench-marking practices, install a fresh drive with clean OS and none of that crap, those exact same PCs will run significantly better. This has been considered okay for years, why is it suddenly cheating when a phone is the hardware?
If you go to his post, you see some speculation. We don't know for a fact that thermal throttling is disabled.
However, even if it is, this is nothing new!
Here is Intel's take:
Note: Intel Turbo Boost Technology 2.0 allows the processor to operate at a power level that is higher than its rated upper power limit (TDP)
I hope not, because that is exactly what turbo boost does.
In that instances, image quality was sacrificed for a performance gain. In this case, a performance gain is obtained at no cost. Vastly different.