Haldi's Benchmark Thread for Governors&Overclocking&Testing the HTC 10

[Haldi4803]

Introduction:

Hello..... people.... i'd love to say friends but i'm new in the HTC section on XDA.
I've been lurking around in the Sony subforums on XDA for a while and owned 3 Sony devices, the X10i, the Z and the Z2. Sadly my Z2 broke down due to waterdmg last month and i bought a cheap replacement phone. Worked pretty well..... until i saw the HTC 10 exhibited in Taipei. Playing around with it for a while made me want to buy a new phone, after more indepth research about the LG G5 and the SGS7 my patience waiting for news about the Zuk 2 or OnePlus 3 was gone and i just bought a HTC 10 Luckily prices are pretty cheap in Taiwan. Worst case i can sell it back home for same value.
So yeah, about Me.... I'm not a developer or anything. My coding skills are horrible, more like scripting to fulfill my needs.
I'm a tech Enthusiast that loves gadgets and everything that's new. I spend money on Kickstarter to often just because i want to see tomorrows technology today. While there is the love for the technology there is also the love for knowledge. Just having something is no fun, knowing how it works is what makes it interesting. Same goes for smartphones. Android is awesome in this case, you can tamper around with so many settings and change so many things.
In the last few years i had so much entertainment with my phones..... looking around the Forums here, there seem to be some competent people around, I'm sure I'm gonna have a great time here

About:
This thread is for gathering informations about Governors, their efficiency, their reaction in daily world usage and not in Antutu benchmark. Or at least that was the original intention. I also did some Benchmarks on Overclocked and Undervolted devices to check if you really gain any benefit. Batterylife was a big concern so i've measured absolut maximum possible and absolut minimum screen on time. After the Z2 Quickcharge hack was made (Japanese Z2 has it, European not) i've started doing more specific Charging time Benchmarks, and i dare say you won't find many as specific charging benchmarks around the web.
Well yeah.... in short whatever i take an interest in i'll benchmark it and document it troughly.
You can see my old threads for the Xperia Z2 or the Xperia Z to get an idea about what i'm talking.

Benchmarking:
Powerdrain: Find out how much the Phone is using under specific circumstances.
Governor: Find out how a governor acts on a specific benchmark course.
Charging: How fast does the Device charge from 1% to 100%


Benchmarking Apps:
Qualcomm Trepn Profiler: CPU and GPU frequency and load, also Powerdrain
Repetitouch: To create custom Benchmarks that simulate normal everyday usage.
BatteryLog: For Charging Benchmark, Voltage and Battery% per Minute
GameBench: For FPS in games. (Make sure you deactivate Screenshot function for HTC10)

Benchmarking Hardware:
YZX Powermonitor: USB 3.0 QC3.0 4-13V 0-3A USB A Male-> USB A Female Powermonitor that measures Voltage and Current passing trough.

Benchmarking Process and Display of Data:
Can all be read in Post NR 80
GPU Logs in Post NR48

More......
I guess that's it for now.
I'll add more if there is more to add about benchmarking.

 

[Haldi4803]

Index:

• Powermeasuring with Trepn Profiler & Wallcharger Post NR3
• Idle Powerdrain with Minimum and Maximum Screenbrightness Post NR4
• Thermal Throttling on CPU and GPU Post NR5
• Charging Stock Charger Post NR6
• Charging QC2.0 Aukey PU-A28 Post NR7
• Charging Compare QC3.0 vs QC 2.0 Post NR8
• Charging Efficiency Post NR9
• Camerabug in old Firmware, max 0.7sec Shutter Post NR10 (now Fixed in 1.80)
• Camera High speed test Post NR11
• Charging in Cold environement Post NR18
• Testing Ghostpepper v8 profile Post NR21
• Testing Ghostpepper v9 profile Post NR22
• RAW performance of SD820 in DMIPS, Singlecore, 2 Cores, 3 cores, all cores Post NR27
• Energy Efficiency, small cores vs Big cores Post NR30
• Gaming and Throttling on Stock Kernel (1.80.709.5) Post NR35
• Gaming with PnP&Thermals tweaks Post NR36
• More Tests about Efficiency of Small and Big Cores Post NR38
• Charging on 1.80.709.1 Firmware Post NR39
• Charging while using the device Post NR40
• Video Analyse Frameskipping? Post NR41
• Guess what you will find in PostNR42
• Screen Batterydrain in Idle Post NR43
• PnP Tweaks compared PostNR46
• Elemental X Kernel v19 with GPU Transitions Post NR49
• Battery Discharge Linearity Post NR50
• Elemental X Kernel v19 + Ivicask's PnP Tweaks v17 UC/OC/Powersafe Post NR51
• Charging Beanstalk CM13 Post NR59
• GPU Benchmarks stop at 58FPS.... Post NR71
• Google account during Doze time, how much is caused by Google Services? Post NR73
• Power_profile.xml how accurate is it? Post NR74
• What about the Power_profile.xml from other Devices? Post NR75
• 6 Years of Video recording Post NR76
• Charging on 1.96.709.5 Post NR77
• Thermal Throttling in a Longterm test Post NR78
• How I Benchmark / Create those Graphs Post NR80
• Audio Testing.... or something like that... Post NR84
• Idle Screen Only Powerdrain Post NR85
• Charging 2.41.709.71 and Battery Deterioration Post NR86
• Charging 3.16.709.3 Stock Oreo Post NR87
• Oreo: Standby Drain in Flightmode Post NR 89
• Oreo: Standby Drain in Flightmode Safemode Post NR 90
• Comparison of the both above Post NR 91
• Comparison in Daily Usage Scenario Post NR 96
• Charging LOS 15.1 with Nebula Alpha Post NR 97
• Per Core Powerdrain, try nr 2 Post NR 98
• Smartpixels Feature on IPS not usefull... Post NR 99

 

[Haldi4803]

Powermeasuring:
How do we know how much power our device is using right now?
the Ampere App shows a rough estimate, Qualcomm Trepn profiler shows a better BatteryPower value.
The best way would be to dismantle the phone and use hardware on the wire between battery and phone.
This does seem a little bit extreme so there is another way, plug the phone to the wallcharger and measure how much the charger draws.

So.... are all these measurements accurate?

I've done the comparison: While being plugged to the wall what does the Hardware say, what does Trepn Profiler say.
One thing needs to be said here. On SD810 and onwards Qualcomm decided to limit the "Direct BatteryPower" to once every 30 seconds. If you want a faster update you need "estimated BatteryPower*", thats also what is used when the phone is connected to a charger.
So i did one Benchmark (Stability Test v2.7 CPU Bench) while connected to the Wallcharger at 100% and transfered the info via bluetooth to my phone, at the same time run Trepn Profiler in estimated* mode.
To make sure being connected to the charger does not make a difference i've unplugged the phone and did another run in estimated* mode.
Then i made the 3rd run unplugged while using direct batterypower mode.

Here is the Graph of all 3 runs consolidated into one:

As you can see the Estimated* graph is far off from the direct and wallcharger. (EDIT: Seems like the early version of Trepn had issues with Big cores Clock and other stuff so estimated usage was totally off)
The problem with the direct mode is that it only gets updated once every 30 seconds, so to get accurate results you need to have stable conditions for 30s or more.

So yeah.... thats the situation under heavy load. But what about Idle?
Maybe it works better when idle? Nope.
I've set manual Brightness to the minimum. Activated Flightmode, disconnected WiFi (bluetooth still active for measuring.....) Screen timeout to 1h and left the phone alone for 10 minutes on the Homescreen.

Same problem here. The Wallcharger shows way more than direct which is also way over estimated mode.
Why is the charger so much more? No idea. I've made sure the phone was on 100% battery, and the Powermeter shows 0.032A and 5.93V so about 0.19W when the screen is off.

Here are the Screenshots that show CPU load and Frequencies.
Estimated run while plugged:
https://drive.google.com/file/d/15YLBTjgcK69qNvL14Ubn67qlmuzEn68HLQ/view?usp=sharing
https://drive.google.com/file/d/1nVwAFMrJDhcYhor0Bq9oDLDgg4alF8vzWg/view?usp=sharing

Direct Power run while unplugged:
https://drive.google.com/file/d/1wgrCGWa1afsjjXBXmqxPkxauL_zbS9YsRg/view?usp=sharing
https://drive.google.com/file/d/1mHo6uV0BuAgj7HDHPgurRVZAxeuifPBMJw/view?usp=sharing


I've contacted the Dev's from Qualcomm and asked what they have to say about this, because their last statement was "estimated mode should be fairly accurate"
(EDIT: Seems like the early version of Trepn had issues with Big cores Clock and other stuff so estimated usage was totally off)

 

[Haldi4803]

Screen Brightness

So,
I've been doing these Idle Benchmarks with minimum brightness, so why not just do another with maximum brightness. All in Manual Screen Brightness so it's not the brightest it can get but the brightest you can set.

Average usage in Watt:
Min Wallcharger: 1.23W
Max Wallcharger: 1.66W
Difference: 0.43W

Min Estimated: 0.14W
Max Estimated: 0.63W
Difference: 0.49W


I might be doing another test in direct mode when i have to much time on my hands

 

[Haldi4803]

Performance and Throttling:

CPU:
Originally the SD820 has 2 Cores at 2150,4mhz and 2 cores at 1593,6mhz
Because this can use almost 10W it's impossible to cool passive. So after a short while the CPU starts throttling to prevent overheating.
I currently do not have Root so i don't see my CPU temperature in the status bar but have to open CPU-Z which is way to much hassle to find out a what temperature the SoC starts throttling.
After a longer time stressed out the CPU (and CPU Only) rests at 1324,8mhz on the 2 fast cores and if that's still not enough all 4 cores will drop down to 1324,8mhz

in this Stabilitytes v2.7 Benchmark i did with a cold phone (Room temperature 24°C) the fast cores went down to 1324,8mhz after 30 Seconds and the slow cores reduced their speed from 1593,6mhz to 1324,8mhz at the 200 seconds mark. This frequency (1324,8mhz on all cores) persisted during the 10 Minute Benchmark.

If you activate the Powersafe mode in settings, your CPU clockspeed will also be Limited to 1324,8mhz.



GPU:
The Adreno530 is by far the fastest current Mobile GPU, but thanks to the 2560x1440pixel screen we really need this performance. If you don't have enough FPS in your games, activate Boost+ GameBoost. It reduces screenresolution for this game to 1920x1080.
To give you an impression on how fast.... if we take the old Epic Citadel Benchmark and keep it running in UltraHigh Quality.... we have about 70% GPU usage ^^

But.... it's not a sustainable performance. The top clock of 625mhz cannot be kept for very long.

(Source: Anandtech)

More detailed Benchmarks will come when i've found a great game to test.....
recommendations are welcome

 

[Haldi4803]

So.... here we go. First full charging Benchmark on my HTC 10 completely logged with an USB Powermeter.
From 1% to 100% takes 95 Minutes.

Statistics from BatteryLog app

And the Voltage and Current recorded via external Powermeter.

Funny How the charge control completely kills the charging to switch Voltage in Big steps.... as seen on Minute 10 and 48.

Another fun fact is that we see the Battery reach 100% After 95 Minutes.
But does it stop charging? No!
It continues until 150 Minutes..... yes no joke. 2.5 hours! That's 1 hour longer than it needed to reach 100%

And one more Longtime Charge.
What happens if you leave your charger connected over night?

(Nope, not the same date as the 3 above, was on another charge)


(If the pictures are to small, blame Tapatalk resize algorithm. I will Upload full size when I'm back at my PC, which might be somewhen between 2 and 6 months.)

 

[Haldi4803]

And QuickCharge 2.0 with my Aukey PA-U28

 

[Haldi4803]

Comparison QC2.0 with QC3.0:

Internal Stats:
Quick Charge 3.0

Quick Charge 2.0

Wallcharger Stats:
Voltage and Current

Watt (calculated by U*I=P)

 

[Haldi4803]

Charging Efficiency:
Using the USB Powermonitor we have a decent Information about how much power flows out of the Wallcharger. U*I=P and we have the Watt flowing out.
On the phone side this seems a little more complicated. To get an accurate value it would be nice to tear down the phone and connect all important points to a Hardware measurement Tool. But yeah. No one is gonna break down his private phone for that.
So we have to take a look at Software.
Battery Monitor Widget does Report the current flowing into the Battery.... same as Ampere.
If we assume the Battery is getting charged at 4.4V we can again calculate the Watt inside the phone.
Comparing the Outside and Inside Watt against each other we have the efficiency. Or at least something that does look like that ....

Quick Charge 3.0

Quick Charge 2.0

I tried to synchronize the graphs but yeah... seems like the logging is a bit unexact. Not sure which one tough.

 

[Haldi4803]

Camera Testing:
EXIF Info says our Camera does not take 2sec pictures but 1 second only...
How true is that?

I tried different Tests but found out that the blur Buster UFO Test at 480pixel/sec gives the best results.
Link: http://www.testufo.com

To compare vs reality I took an EOS 600D and a Macbook with Chrome.

0.5 seconds.

1 second

2 seconds

Then I took my HTC 10 with 1.30.709.1 and HTC Camera App 8.10.748210 Pro Mode JPG iso 100 and set Shutter to:

0.5 seconds

1 second

2 seconds

As you can see from 0.5 to 1 second is nearly no difference and 2 seconds surely doesnt look like 2 seconds from the DSLR at all! What are you doing HTC?


Original files:
https://drive.google.com/folderview?id=0BwqG3liwGXQHVlIyOGZVSWpqOG8

Edit: 17.07.2016
In the latest 1.80 Update this Bug has been fixed

0.6 Seconds:
https://drive.google.com/file/d/0BwqG3liwGXQHbVhkVHBSLWVZd0E/view

1 Second:
https://drive.google.com/file/d/0BwqG3liwGXQHYjRibXFWMkkxTVE/view

2 Seconds:
https://drive.google.com/file/d/0BwqG3liwGXQHU2t4LVJIcEZ5eTQ/view

 

[Haldi4803]

Well....We did the slow Tests. So now come the fast ones.
How do you do that? You need a fast object. 60hz screen? Nope...
Well what did I have around? A fan xD

Auto:
1/30 iso 250

Pro Mode jpg.
1/100 iso 756

1/500 iso 3465

Or with Flash
1/500 iso 400

1/1000 iso 400

1/2000 iso 800

1/4000 iso 1600

1/8000 iso 3200

All pictures in original size on gDrive
https://drive.google.com/folderview?id=0BwqG3liwGXQHRlBCUDNIX05BZk0

 

[Haldi4803]

Memo to myself:

&devfreq_cpufreq {
m4m-cpufreq {
cpu-to-dev-map-0 =
<  307200   307200 >,
<  422400   307200 >,
<  480000   307200 >,
<  556800   307200 >,
<  652800   384000 >,
<  729600   460800 >,
<  844800   537600 >,
<  960000   672000 >,
< 1036800   672000 >,
< 1113600   825600 >,
< 1190400   825600 >,
< 1228800   902400 >,
< 1324800  1056000 >,
< 1401600  1132800 >,
< 1478400  1190400 >,
< 1593600  1382400 >,
< 1728000  1382400 >;
cpu-to-dev-map-2 =
<  480000  307200 >,
<  556800  307200 >,
<  652800  307200 >,
<  729600  307200 >,
<  806400  384000 >,
<  883200  460800 >,
<  940800  537600 >,
< 1036800  595200 >,
< 1113600  672000 >,
< 1190400  672000 >,
< 1248000  748800 >,
< 1324800  825600 >,
< 1401600  902400 >,
< 1478400  979200 >,
< 1555200 1056000 >,
< 1632000 1190400 >,
< 1708800 1228800 >,
< 1785600 1305600 >,
< 1824000 1382400 >,
< 1920000 1459200 >,
< 1996800 1593600 >,
< 2073600 1593600 >,
< 2150400 1593600 >,
< 2265600 1593600 >;
};

mincpubw-cpufreq {
cpu-to-dev-map-0 =
< 1728000 1525 >;
cpu-to-dev-map-2 =
< 2073600 1525 >,
< 2150400 5195 >,
< 2265600 5195 >;
};

Small cores bench:
307
556
652
1036
1113
1324
1401
1478
1593

Big cores bench:
307
556
652
1036
1113
1324
1555
1824
1920
1996
2073
2150

 

[sydeu]

Nice topic, looking forward to the undervolting results!

 

[Haldi4803]

Not gonna happen as flar2 stated his kernel will not have Undervolting available.....
Good thing you mention that. I should edit the topic title

Gesendet von meinem LENNY2 mit Tapatalk

 

[punti_z]

Can someone post the FPS (over time) difference while running games/benchmarks with and without boost+ .. like sustained FPS on 1440p and then on 1080p.. i have been hunting around for numbers on this but cant find anything.

 

[hamdir]

Can someone post the FPS (over time) difference while running games/benchmarks with and without boost+ .. like sustained FPS on 1440p and then on 1080p.. i have been hunting around for numbers on this but cant find anything.

37fps vs 60fps in Trex long term test, Haldi posted the car chase test somewhere but it was also around 2x, so using Boost+ is essential unless you want to end with M8 like graphics speed

 

[Haldi4803]

Can someone post the FPS (over time) difference while running games/benchmarks with and without boost+ .. like sustained FPS on 1440p and then on 1080p.. i have been hunting around for numbers on this but cant find anything.

External factors always matter in this Benches.
Whats your room temperature?
Do you have Sunlight on your phone?
What's the screen brightness?
Whats the PhoneCase temperature?

Depending on these values Throttling will start earlier or later.


Gesendet von meinem LENNY2 mit Tapatalk

 

[Haldi4803]

I thought so.....
Chargig the HTC 10 is NOT Sustainable. The phone throttles charging to safe the Battery from Overheating.
To proof this i've put my phone in the fridge while charging. (you don't have to tell me, I'm fully aware that I'm crazy!)

If we compare the charging duration its about the same.
But we can see the BatteryVoltage is waaayyy higher. Probably due the cold Battery.
And. We don't see the charging Votage droping to 4.6V
Its constantly on 7V

So.... here we go. First full charging Benchmark on my HTC 10 completely logged with an USB Powermeter.
From 1% to 100% takes 95 Minutes.

Statistics from BatteryLog app

And the Voltage and Current recorded via external Powermeter.

 

[ChronoReverse]

I wonder if it'll be possible to find the "most efficient" frequencies (perhaps hold a particular frequency and grind out a fixed workload) where energy over work is minimized for the CPU and GPU.

Very interesting information though, thanks for the graphs and tests.

 

[Haldi4803]

Guess what i've been doing all night ? [emoji14]

https://docs.google.com/spreadsheet...TCFUH9v8VsI4t-1vcig4A68/edit?usp=docslist_api

No time for formating.... my Bus for Tokyo will be leaving soon.