FORUMS

[OP7PRO] ULTIMATE ROM/KERNEL BENCHMARK Comparison & Discussion - OOS vs AOSP

367 posts
Thanks Meter: 110
 
By rejectedjs, Senior Member on 25th March 2020, 03:33 AM
Post Reply Email Thread

So, I'm sure many of us have the burning question of which ROM/kernel any of us should use when it comes to pure performance. With so many ROMs out there, ones based on OOS, Havoc, Paranoid Android, crDroid, OmniROM, MIUI, the list just absolutely goes on. I want to use this thread as a way to inform people of any potential performance gains/loss when it comes to choosing a ROM, as well as just a thread to compare/talk about benchmarks in general.

So I've taken the liberty to run five different setups, (I can take some more as requests, if necessary.) and ran A LOT of benchmarks between all of them to see which ROM provides the best performance. These tests include four various games, some synthetic benchmarks, and storage tests. At the end of it all, I ran all the tests with some pretty heavy kernel based performance modifications that gave a surprising amount of performance gains in some scenarios.


The ROM setups I am using are:

Completely Stock (OOS Beta 11 with zero modifications)
Stock OOS Beta 11 but Magisk / xXx NoLimits xXx 10.1 / Data Formatted as F2FS
HavocOS 3.3 Build 3/17 w/ F2FS
Paranoid Android Quartz Beta 5 w/ F2FS
& Paranoid Android Quartz Beta 5 w/ F2FS AND SmurfKernel 3.5.1 rc17 slmk w/ Overclocks Enabled



All benchmarks were ran under the scrcpy ADB screen mirroring software, and under the settings I've used, it's shown to not provide any amount of performance loss. The command/options I've ran are "scrcpy --render-expired-frames -b 2M -m 768". That makes scrcpy not drop any lost frames (which will increase delay but make frame rate recording much more accurate), have a bitrate of 2Mbit/s, and have a screen height of only 768, which is easy for the phone to do while not incurring any performance loss. Game performance was recorded using MSI Afterburner over the scrcpy window, which isn't THE MOST accurate, but is the only option available due to certain apps not being able to be run under OOS (i.e. KFMARK actually offers these features in app, but crashes on OOS.)

So, lets get started, first, with the gaming benchmarks. I've chosen these games because they are the only ones that are intensive enough on the phone's hardware that can run without hitting the game's frame rate cap. The overclocking kernel of choice for these tests are @pappaschlumpf's SmurfKernel, all my settings can be found here.


GAMES:
Game 1: Click image for larger version

Name:	smcbench.png
Views:	1250
Size:	64.1 KB
ID:	4978351
Game 2: Click image for larger version

Name:	nimianbench.png
Views:	1202
Size:	52.3 KB
ID:	4978349
Game 3: Click image for larger version

Name:	gofbench.png
Views:	1166
Size:	50.2 KB
ID:	4978347
Game 4: Click image for larger version

Name:	xenobench.png
Views:	976
Size:	54.7 KB
ID:	4978357


ANTUTU:
Click image for larger version

Name:	antutubench.png
Views:	1234
Size:	54.5 KB
ID:	4978333


3DMARK:
Test 1:Click image for larger version

Name:	3dmarkSS.png
Views:	1210
Size:	50.5 KB
ID:	4978239
Test 2:Click image for larger version

Name:	3dmarkSSE3.1.png
Views:	1160
Size:	53.8 KB
ID:	4978321
Test 3: Click image for larger version

Name:	3dmarkSSEU.png
Views:	870
Size:	50.8 KB
ID:	4978323
Test 4: Click image for larger version

Name:	3dmarkSSEV.png
Views:	766
Size:	53.1 KB
ID:	4978325
Test 5: Click image for larger version

Name:	3dmarkSSU.png
Views:	752
Size:	51.7 KB
ID:	4978327


GFXBench 5.0:
Set 1:Click image for larger version

Name:	gfxbench1.png
Views:	782
Size:	62.5 KB
ID:	4978339
Set 2:Click image for larger version

Name:	gfxbench2.png
Views:	754
Size:	59.9 KB
ID:	4978341
Set 3:Click image for larger version

Name:	gfxbench3.png
Views:	720
Size:	62.3 KB
ID:	4978343
Set 4:Click image for larger version

Name:	gfxbench4.png
Views:	658
Size:	60.2 KB
ID:	4978345


Geekbench 5
Click image for larger version

Name:	geekbench.png
Views:	808
Size:	47.2 KB
ID:	4978337


AndroBench (Storage):
Set 1:Click image for larger version

Name:	androbench1.png
Views:	960
Size:	59.6 KB
ID:	4978329
Set 2:Click image for larger version

Name:	androbench2.png
Views:	954
Size:	52.5 KB
ID:	4978331




So, what is the takeaway from all this? Which is the winner of the best performing ROM? I think the short answer is, well, no one. Long answer? It really depends. Currently, I don't think ROMs alone can offer any amount of increase in CPU/GPU, giving any extra gaming performance. That seems to be solely up to the kernel.

At a glance, it may look like Havoc offers an immense increase in gaming performance from looking at all of the benchmarks. However, after checking Kernel Tuner, I actually noticed its kernel overclocks the GPU to 675 MHz, (up from 585).

Strangely, OOS seems to offer better SQLite performance than both other ROMs until some serious kernel tweaking is introduced. It may look like xXx NoLimits xXx gives higher storage based scores, but the gains were due to /data being formatted to F2FS. Thus, it seems like NoLimits provides zero recordable performance difference. Not sure what they mean by Speed/RAM optimized. Maybe it's just purely a debloating and keep-more-apps-in-Ram tool.

CPU scores between all ROMs is all within margin of error. I noticed higher MEM/Storage benchmark scores on AOSPA, as well as slightly faster app install times. However I imagine that comes from the fact that it uses the @arter97 kernel, not due to the ROM itself.

So, with all this said, I think the best ROM to choose is whichever one you feel like has the best features / ability to be a daily driver, not what you think will be better performance. I am personally sticking with Paranoid Android because some others have one or two annoying bugs that they haven't squashed and apps like Reachability Cursor and KFMark work seamlessly on it (unlike OOS).
The Following 25 Users Say Thank You to rejectedjs For This Useful Post: [ View ] Gift rejectedjs Ad-Free
25th March 2020, 09:04 AM |#2  
Junior Member
Thanks Meter: 1
 
More
remarkable
25th March 2020, 10:39 AM |#3  
Senior Member
Thanks Meter: 56
 
More
I miss crDroid, AOSPA and TreskMOD in that comparison. Good work anyway !
25th March 2020, 02:22 PM |#4  
OP Senior Member
Thanks Meter: 110
 
More
Quote:
Originally Posted by Mystenes

I miss crDroid, AOSPA and TreskMOD in that comparison. Good work anyway !

How on earth did you miss AOSPA?
25th March 2020, 02:26 PM |#5  
Senior Member
Thanks Meter: 56
 
More
Quote:
Originally Posted by rejectedjs

How on earth did you miss AOSPA?

Should be AOSiP
26th March 2020, 05:25 PM |#6  
OP Senior Member
Thanks Meter: 110
 
More
Quote:
Originally Posted by Mystenes

Should be AOSiP

AOSiP is based on pure AOSP and doesn't have any official Android 10 releases, it's still running Pie, so I don't want to waste time getting into it. TreskMod is just another mod of OmniROM, which HavocOS is already largely based from. crDroid looks interesting, so I'm currently running it and trying it out for myself and will update with benchmarks eventually.
26th March 2020, 07:13 PM |#7  
Senior Member
Thanks Meter: 213
 
Donate to Me
More
for a reliable test, you must flash the same kernel with same settings in all roms, smurf kernel for example works in stock and customs.
You have smurf kernel with forced 90 hz, the same setup must be in all customs and is misconfigured, if you have surfaceflinger boost on, you must disable frame commit boost and viceversa.
26th March 2020, 07:31 PM |#8  
Member
Thanks Meter: 11
 
More
how about call of duty mobile game?did you test it out?thanksyou
26th March 2020, 09:14 PM |#9  
OP Senior Member
Thanks Meter: 110
 
More
Quote:
Originally Posted by Toni Moon

for a reliable test, you must flash the same kernel with same settings in all roms, smurf kernel for example works in stock and customs.
You have smurf kernel with forced 90 hz, the same setup must be in all customs and is misconfigured, if you have surfaceflinger boost on, you must disable frame commit boost and viceversa.

I know SmurfKernel works on OOS / AOSP, the overclocking benchmarks weren't there to show "Look! SmurfKernel + AOSPA is so much better than others!", it's to show people how much of a difference overclocking can make, for them to make the decision of whether they think SmurfKernel's overclocking is worth flashing for or not. I didn't flash the same kernel to all ROMs because that's not the way the developers intended it to be. If the developers wanted their rom to be best experienced under SmurfKernel, they would have either explicitly stated it or packaged it in with the ROM. For example, the way AOSPA does with arter97's kernel. Flashing Smurf to everything would have defeated the purpose of comparing the ROMs.

Quote:
Originally Posted by xNovaLeader

how about call of duty mobile game?did you test it out?thanksyou

Call of Duty Mobile already runs at it's frame rate cap of 60FPS on our phones, there would be zero difference between ROMs.
The Following User Says Thank You to rejectedjs For This Useful Post: [ View ] Gift rejectedjs Ad-Free
26th March 2020, 09:16 PM |#10  
Member
Thanks Meter: 11
 
More
Quote:
Originally Posted by rejectedjs

I know SmurfKernel works on OOS / AOSP, the overclocking benchmarks weren't there to show "Look! SmurfKernel + AOSPA is so much better than others!", it's to show people how much of a difference overclocking can make, for them to make the decision of whether they think SmurfKernel's overclocking is worth flashing for or not. I didn't flash the same kernel to all ROMs because that's not the way the developers intended it to be, if the developers wanted their rom to be best experienced under SmurfKernel, they would have either explicitly stated it, or packaged it in with the ROM, like the way AOSPA does with arter97's kernel. Flashing Smurf to everything would have defeated the purpose of comparing the ROMs.



Call of Duty Mobile already runs at it's frame rate cap of 60FPS on our phones, there would be zero difference between ROMs.

isee..thanks for the info..its suck when its come to flagship phone with higher refreshrate..
27th March 2020, 01:00 AM |#11  
Member
Thanks Meter: 8
 
More
Wow, that's great! Now it seems almost strange why this wasn't done so far What about battery? It would be interesting to benchmark idle / active drain (maybe in 2 scenarios like chrome browsing and other in gaming?), just a though
Post Reply Subscribe to Thread

Tags
benchmark, gaming, kernel, op7, roms

Guest Quick Reply (no urls or BBcode)
Message:
Previous Thread Next Thread
Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes