Database Users

Benchmarking

Ever since I've started using android, ive always been fascinated with the Linpack and Quadrant Standard benchmarking apps.
I currently have a G2 with cyanogen and overclock kernel to 1.9ghz. My best linpack score ive received thus far is around 63 or so.
My question is, is if you look for example, the top scores of the G2, You'll see scores of 69, 70's, 80's, even 90's..and when you view the device data, they're clock speed is only at...lets say..1.4ghz.
Why are their scores so high compared to mine?
Of course, benchmarking programs test more then clock speeds, but still..I feel like with a higher clock speed, I would have results on par with, or better then what they're seeing.
Thoughts?

What I have been told is those guys getting those numbers are running with everything they can shut off.
Sent from my T-Mobile G2 using XDA App

Make sure you've got nothing running in the background or sitting in memory, and make sure to set your minimum clock speed in SetCPU to your max clock speed. It gave me a couple hundred point boost in Quadrant, if I remember correctly.

EfemaN said:
Make sure you've got nothing running in the background or sitting in memory, and make sure to set your minimum clock speed in SetCPU to your max clock speed. It gave me a couple hundred point boost in Quadrant, if I remember correctly.
Click to expand...
Click to collapse
Im aware, and gave tried both of these in the past.
still nowhere near these results :S

Low Stock Quadrant Scores

I'm getting pretty frustrated with my nexus 7 because I'm getting horribly low Quadrant scores. I tried messing around and putting custom kernels on it but when I compared the scores to the thread I was reading, I couldn't even get the stock scores mentioned in the thread with a overclocked custom kernel.
I'm getting a total of 3412 on my Nexus 7 on my stock kernel on the stock rooted/unlocked bootloader 4.2.1. I wiped the device, reinstalled the 4.2.1 stock backup, and then installed quadrant and ran it so there shouldn't be background processes problems. This pisses me off because on the Jay's Buttery Tweaks, he got 4875 on his stock kernel/stock ROM. Somethings wrong there. What would cause such a horrible score?

The first thing you need to know about quadrant scores is................they are decieving, and usually bloated. If your N7 preforms well during use, then your good. Also, if they were that important, all rom and kernel developers would post them with their releases. They dont for a reason.
Sent from my Nexus 7 using xda premium

snowman4839 said:
I'm getting pretty frustrated with my nexus 7 because I'm getting horribly low Quadrant scores. I tried messing around and putting custom kernels on it but when I compared the scores to the thread I was reading, I couldn't even get the stock scores mentioned in the thread with a overclocked custom kernel.
I'm getting a total of 3412 on my Nexus 7 on my stock kernel on the stock rooted/unlocked bootloader 4.2.1. I wiped the device, reinstalled the 4.2.1 stock backup, and then installed quadrant and ran it so there shouldn't be background processes problems. This pisses me off because on the Jay's Buttery Tweaks, he got 4875 on his stock kernel/stock ROM. Somethings wrong there. What would cause such a horrible score?
Click to expand...
Click to collapse
Honestly, who cares? Is your nexus laggy? Are you having trouble doing what you want to do? If not then screw it, you're fine. Why is everyone obsessed with idiotic scores and benchmarks? The girls won't sleep with you cuz your Quadrant was over 9000!

Quadrant has never given the Tegra3 Chip really good scores for some reason? Almost all other benchmarking apps like Antutu, CFBench, Vellemo, Smart-bench, and many others will give 3 phones that I have, that are very close in specs a much closer comparison. I don't know why it is but Quadrant has never benched any of my Tegra3 devices high.
And was stated above go by how it feels in real life usage, is it snappy and smooth don't go specifically by benchmarks all the time they can sometimes give you an idea, but they are not tell you real world usage. For example my moms Samsung Transform Ultra with Android 2.3 and a 1GHz single-core Snapdragon is still really snappy. Its hard to believe that it is almost as fast as some of my quad core newer devices.

I got 2883 on my bone stock while playing the walking dead...
Sent from my Nexus 7 using Tapatalk 2

Why do you care about benchmark scores? They really really don't matter. What matters is how it performs during usage. Nothing else. You can have awesome scores, yet have the device lag during normal use.
From a Nexus S bathing in Jelly Beans

Alright well that was with the motley kernel and I'm not sure what was up but it felt pretty smooth with the motley kernel but gave me those terrible scores. Performance in GTA3 was a decent amount better than stock.
Then I restored the stock rom and installed M kernel and trickster Mod and after that because the benchmarks were just pissing me off. Overclocked to 1.6Ghz and did some SD modifications and jesus christ everything is buttery smooth, fast loading, gta3 runs awesome, and I can get a like 6200 score in quadrant and 15397 on AnTuTu. Probably the fastest anrdoid device I've ever used.
I just confuses the crap out of me because I get pretty crappy graphics scores and changing the GPU clock and raising the GPU voltage doesn't change the GPU score on either benchmark.
Can someone tell me what the LP core is and why my GPU scores won't change?

I disagree, respectfully of course. Benchmarks like the Quadrant are useful standard tools to measure the performance of these devices. While they are not perfect, they point out potential weaknesses in our phones and tablets and whatnot.
Compared to stock Nexus 7, the iPad mini runs much more smoothly and yet is underpowered spec-wise. Still the mini puts up a better benchmark score than the superior N7.
Sent from my Nexus 7 using xda premium

quadrant doesn't matter, but it can be useful for developers or just for the fun of it. first off, most people don't bench correctly. you want the CPU to be at the highest CPU speed at all times, quadrant doesn't put enough stress on the device to allow that, so the CPU will scale. if it scales, you don't know the actual CPU speed that its testing. you want to set the CPU speed at the highest speed, and you want to set the lowest CPU speed as the highest speed also. also, for higher scores, you'd need to disable fsync, and turn off tegra hot plug(and enable all four cores). sdcard tweaks wont help you, neither will jays buttery tweaks. also, overclocking the GPU wont help since we are limited in fps.
---------- Post added at 03:11 PM ---------- Previous post was at 03:10 PM ----------
jamesc760 said:
I disagree, respectfully of course. Benchmarks like the Quadrant are useful standard tools to measure the performance of these devices. While they are not perfect, they point out potential weaknesses in our phones and tablets and whatnot.
Compared to stock Nexus 7, the iPad mini runs much more smoothly and yet is underpowered spec-wise. Still the mini puts up a better benchmark score than the superior N7.
Sent from my Nexus 7 using xda premium
Click to expand...
Click to collapse
where does the iPad mini show better benchmarks?

Well I've tried like 4 or 5 different kernels and m-kernel is the best. Only one that can score 15,000 on Antutu.
Can anyone riddle me why no matter what I change the gpu max frequency or voltages to, my 2d and 3d graphics are always the same?

snowman4839 said:
Well I've tried like 4 or 5 different kernels and m-kernel is the best. Only one that can score 15,000 on Antutu.
Can anyone riddle me why no matter what I change the gpu max frequency or voltages to, my 2d and 3d graphics are always the same?
Click to expand...
Click to collapse
those benchmarks arent gpu benchmarks, especially quadrant. we are limited to 59/60 fps no matter what gpu you use. for gpu try basemark es2.0 taiji, gpu mark, or the windmill benchmark.
trinity kernel quadrant and antutu..

Jesus christ. Those are great results. What settings are you using? I installed trinity on a fresh restore of 4.2.1 and used Trickster Mod and System Tuner to set it to 1.7Ghz, performance, deadline, 1024 cache, zRAM enabled, fsync disabled, with stock voltages but I got barely over 5000.
I'm getting 6135 on the beefed up M-kernel. You get a decent amount higher in CPU, memory, and I/O, but your 2D score triples mine. 3D is the same.

snowman4839 said:
Jesus christ. Those are great results. What settings are you using? I installed trinity on a fresh restore of 4.2.1 and used Trickster Mod and System Tuner to set it to 1.7Ghz, performance, deadline, 1024 cache, zRAM enabled, fsync disabled, with stock voltages but I got barely over 5000.
I'm getting 6135 on the beefed up M-kernel. You get a decent amount higher in CPU, memory, and I/O, but your 2D score triples mine. 3D is the same.
Click to expand...
Click to collapse
bench at 1700mhz high/1700mhz low. turn off tegra hotplug and enable all 4 cores(i dont know if trickster has that option, trinity kernel toolbox does), turn off fsync. trinity doent include zram, so you couldnt of used it. zram has nothing to do for performance anyways and doesnt help you with anything else. zram gives a little more ram for devices that are low ram to begin with. the n7 isnt low ram. plus, with zram, itll get slower after a few hours after bootup. i use ondemand/deadline btw. and, go into the devices main settings, developer options, and enable force gpu rendering. that raise your 2d score. i also bet that you are getting caught in the thermal throttle, which lowers your cpu speed when your device gets hot.

alright cool thanks for the info. I'll add in those settings minus the hotplug fix andsee what that does when I get.back. I know thermal throttling lowers scores. I've been doing tests recently on a bag of ice. the cpu stays plenty cool

snowman4839 said:
I'm getting pretty frustrated with my nexus 7 because I'm getting horribly low Quadrant scores. I tried messing around and putting custom kernels on it but when I compared the scores to the thread I was reading, I couldn't even get the stock scores mentioned in the thread with a overclocked custom kernel.
I'm getting a total of 3412 on my Nexus 7 on my stock kernel on the stock rooted/unlocked bootloader 4.2.1. I wiped the device, reinstalled the 4.2.1 stock backup, and then installed quadrant and ran it so there shouldn't be background processes problems. This pisses me off because on the Jay's Buttery Tweaks, he got 4875 on his stock kernel/stock ROM. Somethings wrong there. What would cause such a horrible score?
Click to expand...
Click to collapse
WOW 3412 QS... Then if I show you this you will be pissed.
Full untouched Stock FL04 JB 4.1.2 SGSII E4GT. QS 4348
Quadrant Score are not important. If it performs well who cares about a QS.
Sent from my SPH-D710 using Tapatalk 2

Omar04 said:
WOW 3412 QS... Then if I show you this you will be pissed.
Full untouched Stock FL04 JB 4.1.2 SGSII E4GT. QS 4348
Quadrant Score are not important. If it performs well who cares about a QS.
Sent from my SPH-D710 using Tapatalk 2
Click to expand...
Click to collapse
Edited to delete comment as I was an idiot and did not read the above post fully before I opened my trap.
Sent from my Nexus 7 using xda premium

jamesc760 said:
I disagree, respectfully of course. Benchmarks like the Quadrant are useful standard tools to measure the performance of these devices. While they are not perfect, they point out potential weaknesses in our phones and tablets and whatnot.
Compared to stock Nexus 7, the iPad mini runs much more smoothly and yet is underpowered spec-wise. Still the mini puts up a better benchmark score than the superior N7.
Sent from my Nexus 7 using xda premium
Click to expand...
Click to collapse
I disagree with that. Just by clearing up RAM and turning off Processes that were not being used just sitting in RAM, maybe 30 or 40MB, waiting to be used my score went up over 200 points. If the benchmark was accurate then that would not have actually made any difference as it would have cleared out any unused processes itself before running.
---------- Post added at 01:44 PM ---------- Previous post was at 01:34 PM ----------
snowman4839 said:
Jesus christ. Those are great results. What settings are you using? I installed trinity on a fresh restore of 4.2.1 and used Trickster Mod and System Tuner to set it to 1.7Ghz, performance, deadline, 1024 cache, zRAM enabled, fsync disabled, with stock voltages but I got barely over 5000.
I'm getting 6135 on the beefed up M-kernel. You get a decent amount higher in CPU, memory, and I/O, but your 2D score triples mine. 3D is the same.
Click to expand...
Click to collapse
Wow. Really? Ive gotten 5900 with just 1.4GHz, GPU 2d rendering and clearing out some RAM, fsync disabled running M-Kernel a23.

Nexus 4 Benchmarks that are all over the place?

Ok so I've been tweaking my phone, doing everything in my knowledge to make it faster. I am using the stock rom with faux123 mako jb enhanced stock 422-r2. So my first benchmark on Antutu i got 18003. re-ran the same benchmark again with the same tweaks and got 14206 and thought it was anomaly so i ran it one more time and got 16142. I went back and re tweaked everything and my scores were every were, even in quadrant. the ranges were between 14200-16800. I haven't been able to get it back to 18003, maybe because i did it on a cold boob. but, the scores shouldn't be all over the place compared to benchmarking I've done on other phones. Any ideas on this and or fix it?

Your phone is throttling. Your first benchmark will normally be your fastest. I that's how it was for me anyways.

I've also had the same experience. It is certainly due to throttling.
Sent from my Nexus 4 using Tapatalk 2

Welcome to thermal throttling. I recommend a ROM with the Bionic patches (there is an otherwise 'stock' AOSP ROM with these in the dev section, I'm not a stock ROM user so I haven't any experience with it) and you should notice AnTuTu scores going up to 20k+ and a noticeable boost to actual usage. If your device will under volt well (most will do at least -100mV AFAIK) you may mitigate the throttling to some extent without compromising stability however I don't recommend disabling the thermal protection as is possible in some kernels due to the red light death issue. Throttling is only really noticeable when you benchmark back-to-back. It's not a massive issue in most real-world use. I suspect it will become more and more noticeable on newer devices as device manufacturers and SoC designers are forever scrambling to provide the best performance within the thermal constraints of a smartphone. As idle and low load consumption goes down with new processes, full load GPU and CPU thermals are forever increasing!

PA tweak improves benchmarks
I have to agree with you on this. The stock gave me an score of 17K (antutu). stability curve was all over the place in stock. switching to PA did not help much either, the scored went down to around 14K. yesterday i got an update from PA and flashing that gave me a boost of 18991. some good tweaks from PA i guess.

MrKickstand said:
Ok so I've been tweaking my phone, doing everything in my knowledge to make it faster. I am using the stock rom with faux123 mako jb enhanced stock 422-r2. So my first benchmark on Antutu i got 18003. re-ran the same benchmark again with the same tweaks and got 14206 and thought it was anomaly so i ran it one more time and got 16142. I went back and re tweaked everything and my scores were every were, even in quadrant. the ranges were between 14200-16800. I haven't been able to get it back to 18003, maybe because i did it on a cold boob. but, the scores shouldn't be all over the place compared to benchmarking I've done on other phones. Any ideas on this and or fix it?
Click to expand...
Click to collapse
Lmao cold boob!
Sent from my Nexus 4 using xda app-developers app

MrKickstand said:
Ok so I've been tweaking my phone, doing everything in my knowledge to make it faster.
Click to expand...
Click to collapse
You are making faster a device which already is one of the FASTEST devices on this planet?
maybe because i did it on a cold boob
Click to expand...
Click to collapse
LOL:laugh:

See if it's a governor problem, set it to performance and see if it levels out. GPU scores on antutu vary by @ 1000 marks, so take that into account too.

[Q] Performance difference from different kernels on different roms

Hi,
I've noticed a huge performance difference between kernels and the roms they're used with.
For example:
I was using AOKP and Franco kernel and got around 20000 antutu points, I've switched to Carbon Rom (because of the build in pie control) and Franco kernel and only get around 13000 points, that's a huge difference.
As a test I've installed Matr1x-kenel on Carbon and get around 21000 points.
I really like Franco-kernel and all the tweaks it offers but don't like the huge drop in benchmarks, I know benchmarks are not a real representation of actual performance but it's still a big difference.
This also occurs in Quadrant and Geekbench.
So my question is why does this happen?
Aren't most roms supposed to be compatible with most kernels?
Thank you in advance.
Best regards.

I can't answer your question as to why that happens (no doubt someone else will) but you seriously should just stop bothering with benchmarks and use your own eyes and experiences as a measure of how good a kernel/ROM is. I doubt you could find a kernel which made the phone visibly slow or that affected usability so I don't see what your concern is tbh.

Thanks for your answer.
Yeah, I read that a lot on XDA, don't trust benchmarks...I understand that but they must have some meaning.
I mean, if not why do they exist or do people bother using them?
To be honest I don't really notice any real performance difference between most kernels I've tested.
Best regards

some roms include many optimizations(like skia/dalvik, krait optimizations, and others), while some dont. its not thekernel thats crapping out on you, its the rom.
---------- Post added at 07:02 PM ---------- Previous post was at 06:58 PM ----------
Nigeldg said:
I can't answer your question as to why that happens (no doubt someone else will) butcomseriously should just stop bothering with benchmarks and use your own eyes and experiences as a measure of how good a kernel/ROM is. I doubt you could find a kernel which made the phone visibly slow or that affected usability so I don't see what your concern is tbh.
Click to expand...
Click to collapse
your eyes can be decieved.. they can see whats happening in the ui for example, but you can not see the complex calculations that are being performed or how your cpu is really performing. you can have a slow device whos ui is quick.

OK, I can understand that not all roms are equal but why does changing the kernel have such a seamingly big impact?
If a rom is bad to begin with it should stay that way no matter what kernel you use with it.
Offcourse what do I know, I'm not a developer so my knowledge on the subject is limited.
I'm just trying to understand what's going on...
Best regards

Pihkal said:
OK, I can understand that not all roms are equal but why does changing the kernel have such a seamingly big impact?
If a rom is bad to begin with it should stay that way no matter what kernel you use with it.
Offcourse what do I know, I'm not a developer so my knowledge on the subject is limited.
I'm just trying to understand what's going on...
Best regards
Click to expand...
Click to collapse
kernels vary too, and they impact greatly because they control just about everything in the phone, kinda like a brain. since the kernels themselves vary, one kernel might be better set up than another to deal with certain code from a certain rom. and then also, every phone reacts differently to each kernel(and roms to a point). thats why its recommended to try out different kernels, combos. only then you can find the perfect combo for you/your device. what works great for somebody, can be lousy for another.

OK, so if i understand correctly it boils down to this:
1. You can do benchmarks but don't base your opinion on just the benchmark scores.
2. Roms can vary greatly in optimizations and efficiency of coding.
3. Kernels can also vary greatly in optimizations and efficiency of coding.
4. There's no such thing as a "best for everyone rom/kernel combo".
5. Not all roms/kernels play equally nice with each other.
6. Play around with as many roms / kernels as possible and decide what works best for ME based on MY experience.
Thanks for the advice.
Best regards.

Its been well over a year since I ran any benchmark of any sort but I tested Franco and carbon because that's what I'm on and you mentioned low scores. I'm on Franco m3 with some tweaked settings and carbon nightly from 7-5. Antutu gave me 20636. I'm using stock CPU and GPU frequencies.

username8611 said:
Its been well over a year since I ran any benchmark of any sort but I tested Franco and carbon because that's what I'm on and you mentioned low scores. I'm on Franco m3 with some tweaked settings and carbon nightly from 7-5. Antutu gave me 20636. I'm using stock CPU and GPU frequencies.
Click to expand...
Click to collapse
Maybe the nightly Carbon rom is more optimized?
I'm on Carbon 1.7 Stable and used Franco nightly 162 to test with.
When I benchmark I try to be as consistent as possible ie same temperature, performance governor, airplane mode etc.
I even cooled my Nexus in the freezer for some minutes to eliminate thermal throttling (yeah I know, watchout for condensation) but still got the same low scores.
Best regards.

Pihkal said:
Maybe the nightly Carbon rom is more optimized?
I'm on Carbon 1.7 Stable and used Franco nightly 162 to test with.
When I benchmark I try to be as consistent as possible ie same temperature, performance governor, airplane mode etc.
I even cooled my Nexus in the freezer for some minutes to eliminate thermal throttling (yeah I know, watchout for condensation) but still got the same low scores.
Best regards.
Click to expand...
Click to collapse
It might be, I didn't do anything special. I left all my background apps running, didnt close anything in the recents, didn't cool the phone first. I just downloaded it and hit start. I use the interactive governor tweaked a bit, and I also tweaked the hotplug settings so it more readily onlines all 4 cores instead of waiting for some of the higher loads to trigger it.

username8611 said:
It might be, I didn't do anything special. I left all my background apps running, didnt close anything in the recents, didn't cool the phone first. I just downloaded it and hit start. I use the interactive governor tweaked a bit, and I also tweaked the hotplug settings so it more readily onlines all 4 cores instead of waiting for some of the higher loads to trigger it.
Click to expand...
Click to collapse
Well, I had to try it myself so I did a factory reset, cleared system,data and dalvik, installed latest carbon nightly.
With stock kernel I almost reached 21000 points, with franco I barely get 17000 points.
Very strange...
edit:
I stand corrected, did a second benchmark and am now getting 20880 points...

are you benchmarking with your cpu speed benchmarked set as highest and lowest cpu speed? you should. if you dont put the same cpu speed as highest and lowest then itll scale up and down. if it scales, you dont actually know what speed its testing and it gives you inconsistamt scores. you want the cpu speed to be the same throughout the test.

When I benchmark I set the governor to performance, this should keep the cpu running at maximum speed without scaling unless I'm mistaking...

Pihkal said:
When I benchmark I set the governor to performance, this should keep the cpu running at maximum speed without scaling unless I'm mistaking...
Click to expand...
Click to collapse
meh, performance is the worst for benchmarking. its such a deceiving name. try either ondemand or interactive. set your cpu speed to be the same high and low.

simms22 said:
meh, performance is the worst for benchmarking. its such a deceiving name. try either ondemand or interactive. set your cpu speed to be the same high and low.
Click to expand...
Click to collapse
Woow, that's a great tip, I now get 22003 points with Matr1x-kernel.

Pihkal said:
Woow, that's a great tip, I now get 22003 points with Matr1x-kernel.
Click to expand...
Click to collapse
better :highfive:

Antutu UX I/O performance score indicator for phone "snappiness"?

Hey guys,
would you be so kind and tell me your specific antutu score from the section named above. I had now 3 nexus 6p in my hands who feel more or less snappy (same setup ofc) and it matches with their ux i/o performance score.
With that being said, mine always comes out at around 3k with my girlfriend's phone having nearly 1k more. And like I said, theirs feels much smoother when using it.
Thanks.

Benchmarks are useless for testing performance. They're supposed to be taken with a grain of salt. Run a benchmark, now run another one. And 1 more after that. Each one will lower in score every time since the CPU temp is increasing which kicks thermal throttling into gear. Besides, what's your control group? Stock, unmodified? Because you'll likely have a number of different results from people running any combination of ROMs/kernels. Some may overclock, some may disable hotplugging, using different governor or schedulers, etc. The results would be widely different and wouldn't tell you much. Aside from that, I believe there is a dedicated benchmark thread buried in this section somewhere if you want to see others results or post yours.

RoyJ said:
Benchmarks are useless for testing performance. They're supposed to be taken with a grain of salt. Run a benchmark, now run another one. And 1 more after that. Each one will lower in score every time since the CPU temp is increasing which kicks thermal throttling into gear. Besides, what's your control group? Stock, unmodified? Because you'll likely have a number of different results from people running any combination of ROMs/kernels. Some may overclock, some may disable hotplugging, using different governor or schedulers, etc. The results would be widely different and wouldn't tell you much. Aside from that, I believe there is a dedicated benchmark thread buried in this section somewhere if you want to see others results or post yours.
Click to expand...
Click to collapse
i am not talking of cpu bench results. im well aware of the thermal throttling kicking in.
i/o performance should not decrease and does not decrease over several runs.
also i stated that both mine and my girls phone were running a clean pure nexus nougat build (latest) with ex kernel and same setup.
the scores differed in points to a degree where are i can rule out benchmark tolerance.

some more ux i/o performance scores from users here, perhaps?