Hi there! Any devs here?
Maybe anyone knows why in few games Adreno CANT BE on 100% usage?
For example: my phone is Pocophone F1. SD845, Adreno 630. Usage in any benchmark at 100%. Minecraft: 414 Mhz (instead of 710 Mhz - max for MIUI kernel), 80%, 50% usage in total.
At the same moment Mali at 100% usage with a LOT bigger FPS. Same with GTA:SA, possibly Modern Combat 5. What's happening? Any fix?
Maybe game devs should do something like PC game devs do?: (enable NVidia Optimus, bad example btw) vOptimusEnablement = 0x00000001 (just an example).
P.S. Oops, forgot. Throttling disabled, 5 minutes in CPU Throttle Test shows no difference (5min enough for this because the game started lag at the whole start).
All programming instructions ( whether they are graphic information or not ) are first processed by the CPU: if the CPU recognizes that they are graphics commands, they are delegated to the GPU. Unlike a CPU - which has only a few cores - a GPU has hundreds of cores.
Throttling the CPU only makes things worse.
jwoegerbauer said:
Throttling the CPU only makes things worse.
Click to expand...
Click to collapse
As i said there's NO throttle. Fully disabled. That's happening on any Kernel. Idk what is this, Adreno-related thing, no problems at Mali with that.
jwoegerbauer said:
All programming instructions ( whether they are graphic information or not ) are first processed by the CPU: if the CPU recognizes that they are graphics commands, they are delegated to the GPU. Unlike a CPU - which has only a few cores - a GPU has hundreds of cores.
Click to expand...
Click to collapse
Of course. Maybe Adreno-drivers need something for this? To say to CPU "Hey, this is an application that needs a lot of GPU power"?
_RusJJ_ said:
Of course. Maybe Adreno-drivers need something for this? To say to CPU "Hey, this is an application that needs a lot of GPU power"?
Click to expand...
Click to collapse
An app neither speaks to CPU nor to GPU: an app's program code gets loaded into RAM, then read in from there by CPU and sequentially processed.
What are you talking about, lol? The game cannot use GPU. It needs something. Maybe some function to call, idk exactly.
You're speaking about completely another thing that i know for almost 12 years, lmao.
Still waiting for any answer. Few games cannot load a GPU. Only on Adreno, Mali GPUs are good.
My last 2 cents here:
A GPU is a processor with hundreds of cores: a game never can load a GPU, it only makes use of it.
For an ideal situation, your device's CPU usage should remain in the 60% to 80% range (up to 90%), and device's GPU usage at 99% or 100%. One of the main causes of low GPU usage is due to CPU bottleneck. It means that you have an under-powered CPU that is not able to keep up with GPU's performance.
A temporary workaround for this problem is to raise a game's resolution to highest if you haven’t done it already. This will cause device's GPU to work more and it will have much higher utilization than before.
Take note that not all games use all the CPU cores. It can be possible that you are getting high usage on two cores only, and others are just not used at all. AFAIK GTA SA makes use of only 2 cores, COD also makes use of only 2 cores, , but PUBG makes use of 4 cores,
What the hell are you talking about?
Is there any CPU bottleneck im talking about?
NO. NO. AND NO. ADRENO-RELATED, HUH?
Pls, stop it, my chair burns. You dont understand a single thing that i need - ADRENO CANT HANDLE A GAME.
Looks like a joke...
Actually, I'm not interested in your problem at all.
Your posts show that you don't understand how CPU and GPU work together.
Here I'll explain it to you:
The GPU's memory consists of a series of registers. These permit the CPU to access the GPU's memory and instruct the GPU to perform operations. It's the CPU what loads the graphics instructions ( OpenGL ES ) to get executed into GPU's memory.
Simplified: the CPU feeds the GPU.
xXx yYy said:
Your posts show that you don't understand how CPU and GPU work together.
Click to expand...
Click to collapse
My post shows that GPU cant handle few games. All Adreno's. GG guys, open your eyes
Still waiting for useful info. For now stop showing your brilliant mind with yours "i think you..."
After a time and with using of a simpleperf i got more info. Playing Half-Life 2:
CPU Usage 12.12% /vendor/lib/egl/libGLESv2_adreno.so
Symbol: !!!0000!f56be09eb88f86833124f1df42e945!95db91f!
Mali HAS NO that problem. Completely.
Related
Hi there, is it possible to overclock GPU?
i guess on most devices overclocking DOES overclock cpu and gpu at the same time
elgubbo said:
i guess on most devices overclocking DOES overclock cpu and gpu at the same time
Click to expand...
Click to collapse
i don't think so bcs they are different hardware parts and when you overclock cpu it is sth about cpufrec but yeah when you overclock cpu it effects graphics too bcs it makes processes faster ...( i just think like that bcs of being a linux user open source foreveeeeeeeeeer )
If we could overclock the gpu on the android then the android would take over the world
kifno said:
If we could overclock the gpu on the android then the android would take over the world
Click to expand...
Click to collapse
android has already became a legend by being the first os in USA(Apple's IOS is losing users. )
So it is still impossible, right?
kanek06 said:
So it is still impossible, right?
Click to expand...
Click to collapse
i think it is possible ( when i think that they are like little computers) but i dunno how to do it.
to start, youd probably want to dissect the software driver and see what it offers.
Well, when we will able to overclock GPU - we should not buy new and new devices, because our present devices are old. But how about temperature? Will battery get hot like in overclocking CPU?
well, from what i understand the gpu for most devices shares space with the cpu on dye and that there is a divider of sorts that keeps the gpu speed a percentage of the cpu.. and since cpus can oc better then gpus, the whole reason we can oc is due to the fact that devs have figured out how to uncouple the clock increase of the cpu from the gpu, otherwise the gpus clock would limit the cpu.. now overclocking itself doesnt increase heat and most of the time (depending on what governor is used) it doesnt always run at full clock so theres not a high constant current drain causing the battery to get hot (unless you run an intensive app continually) also, you wont see the cpu overheat as long as the voltage stays stock, but increasing cpu voltage is how you will get past the th max (thermal maximum)
now if you could adjust that divider you could oc the gpu..
Hey there,
I've read some threads about the dangers of overclocking an Android-device in general. I'm aware of these dangers.
But in my feeling overclocking the sensation is a little different.
Most custom roms set CPU-speed to 1500MHz each core, which is the native clock of the chipset (or am I'm wrong?). No problem so far.
But with tweaking the CPU often the GPU is overclocked, too.
Now the point: What are the dangers of overclocking the GPU?
Does it make sense to overclock just the CPU without the GPU? Is it possible?
What are your experiences with overclocking (e.g. with IsertCoin or RevolutionHD) the GPU?
What you have to remeber with regard to overclocking is that your not alone in doing it when you use one of the custom roms.
Many thousands of ppl are running overclocked handsets and the first things you see happening from overclocking to far is your handset not booting up or locking up well before you would do any physical damage to the handset i would say.
I have run my sensation at 1800 cpu speed and it started ok but locked up when doing one of the benchmark tools , i rebooted and reduced the speed a little but all running fine with no concerns with regard to overheating
So there is a hardware protection to prevent serious damage?
I dont know enough about the handset to say if there is or not i suspect there will be something , but have you ever tried to overclock a graphics card on a pc you get so far and it wont display or something like that so you reduce it.
Heat with any electronic component is i problem and overclocking something increases that heat, but when you get to a certain point your phone will lock freeze reboot etc usually before any damage is done.
If you have any concern about destroying your phone by overclocking the simple answere is that you dont overclock it .
Diydude said:
[...]
If you have any concern about destroying your phone by overclocking the simple answere is that you dont overclock it .
Click to expand...
Click to collapse
Yes you're right of course, but i just want to get a feeling for this topic... i just try to decide if it's a good idea to use RevolutionHD, which is overclocked from the beginning.
And to get a feeling, I just asked about the risks, because the benefits are clear for me
To answer your question, yes, the sensation comes with thermal protection that helps against overheating. If the device becomes to hot, it(in most cases) shuts itself down until it cools. The CPU in the sensation is underclocked when it's shipped at 1.2Ghz(1188), but the true spec of the CPU is actually 1.5Ghz(1536). HTC underclocks the CPU to 1.2 to save on battery consumption. I personally have ran my phone just fine at the 1.5Ghz setting, I've even pushed it up to 1.6Ghz(1620) and it ran fine with the exception it did get just a little hot, but nothing to worry me. That being said I've since underclocked my CPU to 1296 and all is well. And afterall, how fast do you really need your phone to be before you take it too far and fry the cpu?? The sensation is fast right out of the box and even faster when using a NO sense rom, such as cm7(my personal fav)
Side Note : The new sensation XE ships with the cpu at it's original clock speed of 1.5Ghz.. both sensations have the 'exact' same cpu.
In regards to your GPU question, I have mine set @ 300mhz and not 320. The sensations GPU comes clocked at 266.I've ran it @ 320 and it ran just fine.. for me it's more of a personal preference. My graphics look just fine clocked @ 300mhz.. Below is a quote about the chipset and the new one coming out.
"With Adreno 225 Qualcomm improves performance along two vectors, the first being clock speed. While Adreno 220 (used in the MSM8660) ran at 266MHz, Adreno 225 runs at 400MHz thanks to 28nm. Secondly, Qualcomm tells us Adreno 225 is accompanied by "significant driver improvements". Keeping in mind the sheer amount of compute potential of the Adreno 22x family, it only makes sense that driver improvements could unlock a lot of performance. Qualcomm expects the 225 to be 50% faster than the outgoing 220."
if this helped, give the thank you button a punch
p.s. I've ran both insertcoin & revolution HD and both ran fine.. I just really prefer no sense roms
Quakeworld said:
To answer your question, yes, the sensation comes with thermal protection that helps against overheating. If the device becomes to hot, it(in most cases) shuts itself down until it cools. The CPU in the sensation is underclocked when it's shipped at 1.2Ghz(1188), but the true spec of the CPU is actually 1.5Ghz(1536). HTC underclocks the CPU to 1.2 to save on battery consumption. I personally have ran my phone just fine at the 1.5Ghz setting, I've even pushed it up to 1.6Ghz(1620) and it ran fine with the exception it did get just a little hot, but nothing to worry me. That being said I've since underclocked my CPU to 1296 and all is well. And afterall, how fast do you really need your phone to be before you take it too far and fry the cpu?? The sensation is fast right out of the box and even faster when using a NO sense rom, such as cm7(my personal fav)
Side Note : The new sensation XE ships with the cpu at it's original clock speed of 1.5Ghz.. both sensations have the 'exact' same cpu.
In regards to your GPU question, I have mine set @ 300mhz and not 320. The sensations GPU comes clocked at 266.I've ran it @ 320 and it ran just fine.. for me it's more of a personal preference. My graphics look just fine clocked @ 300mhz.. Below is a quote about the chipset and the new one coming out.
"With Adreno 225 Qualcomm improves performance along two vectors, the first being clock speed. While Adreno 220 (used in the MSM8660) ran at 266MHz, Adreno 225 runs at 400MHz thanks to 28nm. Secondly, Qualcomm tells us Adreno 225 is accompanied by "significant driver improvements". Keeping in mind the sheer amount of compute potential of the Adreno 22x family, it only makes sense that driver improvements could unlock a lot of performance. Qualcomm expects the 225 to be 50% faster than the outgoing 220."
if this helped, give the thank you button a punch
p.s. I've ran both insertcoin & revolution HD and both ran fine.. I just really prefer no sense roms
Click to expand...
Click to collapse
some sources say adreno 220 can be overclocked to 400mhz .... bdw how can i overclock ...is der a app like setcpu for it ?????
rohit choudhary said:
some sources say adreno 220 can be overclocked to 400mhz .... bdw how can i overclock ...is der a app like setcpu for it ?????
Click to expand...
Click to collapse
a few time ago some other user reported the same thing about adreno 220
but i think it is too high that value and it is impossible for gpu to handle such a value
in order to overclock the gpu you have to flash another kernel
Hello guys,
Since i`ve seen so many debates about single vs dual-core, today i was bored and i tested with antutu and quadrant this thing.( I know that benchmarks isn`t a real proof in day-to-day performance )
WANT THE SCRIPTS FOR UNDERVOLTING AND FORCING CPU1 ONLINE OR OFFLINE ? See 2nd post
SO , what I actually did:
-flashed a freshly new rom ( using elegancia 3.1.0 for about 2 weeks and i found to be very stable and smooth with better battery life than any other rom i`ve tested)
-flashed latest bricked kernel min 192 mhz max 972 mhz with gpu oc @300mhz, lagfree governor, I/O deadline, both cores online via init.d script, booted and tested with antutu and quadrant and then i let the phone settle for a while to see what is the power consumption in stand-by with battery monitor widget.
-after that, i flashed bricked kernel with only max speed change to 1836 mhz, cpu1 offline via inid.d script and made those tests again
Here`s what I got:
Antutu: 4827 (both cores online, clocked @ 972 Mhz)
Quadrant: 2592
BMW: -19mA
Antutu: 4593 ( cpu1 offline, clocked @ 1.836 Ghz )
Quadrant: 3393 ( cpu1 offline, clocked @ 1.836 Ghz )
BMW: -28mA
Those numbers doesn`t reflect exactly the user experience. Some apps opened faster with 2 cores, some faster with only 1 core clecked at higher speed. The only major difference i noticed was in the browser( stock ICS browser) where the more fluid experience was with both cores on.
I`m gonna test those 2 configuration further to see which one has better battery and post some screenshots.
I will add a poll to see which configuration you think is the best.
Be aware that SoC are not created equaly, so the UV script or OC will NOT work with all devices.
Unrar the archive and choose what you want
This files needs to be copied in /system/etc/inid.d and then set the right permissions ( read: all 3 needs to be checked; write: owner checked; execute: same as read) . You need a file explorer that has access up to root ( i recommand Root exploer)
Benchmarks should always be done using Performance governor for consistency
Michealtbh said:
Benchmarks should always be done using Performance governor for consistency
Click to expand...
Click to collapse
I knew that, but i wanted a more real-life experience, and in my opinion there isn`t that much difference in lagfree vs performance. I ran benchmarks 2 times and the scores where almost identical
Please be aware that because of the instruction sets hard coded at low level into the cpu it doesn not mean that 2x 900mhz cpu cores is equal to 1x 1800mhz cpu core because of the way in which the second core is utilised. Its exactly the same with pc's hence why sometimes an AMD cpu with 2 cores can give a better real world performance benchmark than an intel cpu with 4 cores. the same applies vice versa, it just depends on what instruction sets were used and how the cpu is used. Benchmarking stuff like this is not a reliable way to test
Jonny said:
Please be aware that because of the instruction sets hard coded at low level into the cpu it doesn not mean that 2x 900mhz cpu cores is equal to 1x 1800mhz cpu core because of the way in which the second core is utilised. Its exactly the same with pc's hence why sometimes an AMD cpu with 2 cores can give a better real world performance benchmark than an intel cpu with 4 cores. the same applies vice versa, it just depends on what instruction sets were used and how the cpu is used. Benchmarking stuff like this is not a reliable way to test
Click to expand...
Click to collapse
Thanks (I'd use thanks button,but i'm limited to 8 oer day) for clarifying this for me and any other users, this post should be sticky because many members on xda think that way. I am aware that 2x900 doesn't equal 1x1800, but I didn't knew the exact explanation. The only reason that I did these tests was to see which configuration gives the best battery life. Their not equal, but acording to antutu, quadrant and end user experience they are comparable
Thanks for your time in doing these tests... It might not be reliable but it was interesting to know bout it... Im curious between the two tests though, which test puts the cpu under more stress? the first test shows awesome result on the battery so being single core might stress the cpu more, im guessing
Sent from my HTC Sensation using Tapatalk
AndroidNeophyte said:
Thanks for your time in doing these tests... It might not be reliable but it was interesting to know bout it... Im curious between the two tests though, which test puts the cpu under more stress? the first test shows awesome result on the battery so being single core might stress the cpu more, im guessing
Sent from my HTC Sensation using Tapatalk
Click to expand...
Click to collapse
33% left and it's running for15 hours with 2 cores
hi guyz, here is a little video witch looks a not standard demo video, but a real working product with hmp solution (the galaxy tab 10.1 2014 ed)
http://www.youtube.com/watch?v=wYMnu4L2-aU
girl sad it works over the same android solution, so, we could have it in an android 4.3 upgrade, and not only from the 4.4
untill better news, watch the video :good:
iba21 said:
hi guyz, here is a little video witch looks a not standard demo video, but a real working product with hmp solution (the galaxy tab 10.1 2014 ed)
http://www.youtube.com/watch?v=wYMnu4L2-aU
girl sad it works over the same android solution, so, we could have it in an android 4.3 upgrade, and not only from the 4.4
untill better news, watch the video :good:
Click to expand...
Click to collapse
"Android 4.4 can not be just a modified kernel"
Where is it?
I hope samsung this issue as soon as possible:good:
Bilux said:
"Android 4.4 can not be just a modified kernel"
Where is it?
I hope samsung this issue as soon as possible:good:
Click to expand...
Click to collapse
What??m
Sent from my SM-N900 using xda app-developers app
thats really great news hope it comes with 4.4 update
iba21 said:
but a real working product with hmp solution (the galaxy tab 10.1 2014 ed)
Click to expand...
Click to collapse
That's not a 10.1 2014. It's some sort of test device. Look how big the bezels are and how thick it is.
Umh..
Edit: right.. it has a different form factor as the galaxy tab 10.1
zeyadhan said:
thats really great news hope it comes with 4.4 update
Click to expand...
Click to collapse
If samsung wont upgrade the note3.. it will be only a commercial choice.. and sure.. too mouch peoples will disappoint it
iba21 said:
If samsung wont upgrade the note3.. it will be only a commercial choice.. and sure.. too mouch peoples will disappoint it
Click to expand...
Click to collapse
Design, performance, and the thermal profile of Samsung's current Octa devices weren't optimized to run all eight cores at once. Battery life would suffer as could the longevity of some of the components with the additional heat not being able to be properly dissipated. Even if they could activate HMP on existing devices via s/w that doesn't mean they will or that it would be smart. How many mainstream Octa-device owners do you think know or care about HMP?
BarryH_GEG said:
Design, performance, and the thermal profile of Samsung's current Octa devices weren't optimized to run all eight cores at once. Battery life would suffer as could the longevity of some of the components with the additional heat not being able to be properly dissipated. Even if they could activate HMP on existing devices via s/w that doesn't mean they will or that it would be smart. How many mainstream Octa-device owners do you think know or care about HMP?
Click to expand...
Click to collapse
Design performance and thermal profile.. well.. how do you know the thermal profile has issues? Do you have a datasheet?
And.. please.. respond..
The ACTUAL use is the CORE MIGRATION.. traduct for all, 4 cortexA7 at the base.. when touch the screen, in the 99% of the touching situation, cluster switchs from 4 a7 to 4 A15
Why? IKS LOGIC
So.. are you writing a text? Well you are using 4 cortexA15
With hmp the difference is.. an ammount of 4 cortexA7 at the base.. and.. if really necessary, turn on 1-2-3-4 cortexA15
If you look in the video you see angry birds runs over 4 cortexA7, and sometimes turns on the A15 for "other" things..
Well, with the actual cluster migration, in the same situation, game computation is compiled by 4 cortexA15
If you would i can prove it.. i'm on an n900
Well, do you thing 4cortexA7 + SOMETIMES 1-2-3-4 A15 has an HIGHER battery drain than 4 cortexA15?
Because.. when my phone reaches the maxfreq, cpu temperature goes to 80 degreese Celsius in about 4 seconds.. after that thermal ID decreses the maxfreq value..
Well.. the WORST condiction is when there are 4 A15 turned on..
All thinks about hmp as a pure 8 cores with always all 8 cores turned on.. but.. looks at the s800.. does it have the quadcore always turned on??
do you know s800 has an huge heating issue?
With 4 cores at maxfreq (2.3ghz) in about 3 seconds cpu temperature goes to 65 celsius.. after that, maxfreq changes to 1.3-1.5ghz
So.. 4 krait 400 have thermal issues
4 cortex A15 have thermal issues
4 cortexA7 works perfect without any issue
That's the real goal of hmp
Use the cortexA7 for ALL POSSIBLE.. and the other compiled by cortexA15
The 8 core condiction is reached only in heavy benchmarks and "maybe" in heavy games
Web rendering is compiled by A7.. panning needs 1 cortexA15..
So.. do you really think this 5420 has thermal issues?
Well, last stupid question..
The next cpus will have a lower building process, but, will have an higher frequency, so, high consuption caused by high performance arch.. an the 64bit needs more transistors.. so.. more heating issue
How to fix it? An external dissipator with a 120mm fan?
Com'on.. sometimes use the brain
iba21 said:
Com'on.. sometimes use the brain
Click to expand...
Click to collapse
Back at you. You and a bunch of enthusiasts all have their panties in a bunch because one of the two modes Octa runs in isn't being offered by Samsung. There's nothing wrong with the approach they've taken assuming it was to deliver a SoC with comparable performance to Qualcomm's latest chips. With that goal being reached and the right balance between energy efficiency vs. performance already achieved, what reason would Samsung have to, months after the N3 was released, release a feature they could have released at launch? No one but enthusiasts could give a crap about HMP and they make up an estimated 5% of the device population. Samsung builds devices for the mainstream, not enthusiasts. KNOX and eFuse should pretty much confirm that. My logic as to why Samsung chose not to enable HMP and most likely won't until a later generation of Octa chips is released is more based on business logic and common sense than anything else. But you and others keep rationalizing and justifying its imminent appearance via an update. That worked out really well for the SGS4 Octa owners who've been doing the same thing since March.
Oh.. second responce..
With my usage.. i forced the ondemand + my settings + battery saving... why? With this setting the 650mhz (1.3ghz of cortexA7) is the second most freq used.. the first is the 250mhz.. (500mhz of cortexA7)
With this setting i never go over 39 celsius with a normal usage..
Where is the issue?
When i'm writing a text.. cpu increases over the 650mhz.. and.. changes the cluster to A15
So.. for 1 keyboard task.. all 4 cores are used..
The same situation is when i move the web pages, the panning..
With the same situation, with hmp, 4 A7 will used for the most of the things (web rendering too)
And 1 or 2 cortexA15 will used for web panning and keyboard writing
Do you think 4 A7 At maxfreq + 1 A15 at maxfreq has an higher battery drain than 4 A15 at maxfreq??
Oooh.. if i shut down the batterysaving, cortexA15 will be used for the most of the time.. the A7 will.be used only for iper base functions, as like calling and multimedias
so.. if you think it's normal, you should buy one n900 and looks in the kernel sysfs code
BarryH_GEG said:
Back at you. You and a bunch of enthusiasts all have their panties in a bunch because one of the two modes Octa runs in isn't being offered by Samsung. There's nothing wrong with the approach they've taken assuming it was to deliver a SoC with comparable performance to Qualcomm's latest chips. With that goal being reached and the right balance between energy efficiency vs. performance already achieved, what reason would Samsung have to, months after the N3 was released, release a feature they could have released at launch? No one but enthusiasts could give a crap about HMP and they make up an estimated 5% of the device population. Samsung builds devices for the mainstream, not enthusiasts. KNOX and eFuse should pretty much confirm that. My logic as to why Samsung chose not to enable HMP and most likely won't until a later generation of Octa chips is released is more based on business logic and common sense than anything else. But you and others keep rationalizing and justifying its imminent appearance via an update. That worked out really well for the SGS4 Octa owners who've been doing the same thing since March.
Click to expand...
Click to collapse
Are you a flamer?
1) you sad n900 has thermal dissipation problem.. it's as like as on normal people say the sun is cold..
2) Galaxy s4 coudn't have hmp due to a CCI transistor issue.. it means it's impossible to turn on 2 different cluster or system crashes..
3) i don't know why samsung doesen't releases the hmp until.. it's difficult to compile the code, all the scheduler, with the new function.. it's the FIRST TIME in the world.. so.. there aren't any other base code to copy (ex.. ondemand governor was wrote by intel developers.. INTEL..)
4) samsung sad exynos 5420 is ready for hmp.. new exynos versions? Well where do you see those? Post me an official link, not an ipotetic exynos 6xxx with 64bit arch.. because.. i'm rocco siffredi by using words :thumbup:
The only thing witch n900 doesent have is the widcon technology.. only that..
So.. the situation are 2:
1) samsung will release hmp soon for n900
2) samsung will release hmp for galaxy s5 as a marketing pubblicity
Anyway.. n900 has the worst cpu logic.. core migration should be better.. hmp is the maximize of battery saving with the same task
Oooh.. i forgot..
More frequency = more performance = more battery drain = more heat
If i'm using 4 A15 when only 1 is really need.. i'm spending more energy for other 3 A15 FOR NOTHING
Do you think 3 A15 for nothing is what you sad a good battery drain condiction?
Do you think 4 cortexA7 has an higher batterydrain than 3 cortexA15
With hmp i could shut down those 3 A15 and direct use the 4 cortexA7.. with the same performance
With hmp A15 CORES WILL BE USED ONLY IF REAL NECESSARY.. SO.. NO MORE A15 TURNED ON FOR NOTHING
now you understand?
Oh.. why samsung doesen't release hmp from the day 1.. i probably have 1 answer
Since s800 and 5420 has a simillar benchmark results.. 5420 with all 8 core will have more benchmark results
So.. samsung released the s800 version in the most of the world, and the 5420 in the other even if the 5420 should reach an higher benchmark ammount?
Who bought the s800?
Peoples witch wants a slower cpu?
iba21 said:
Are you a flamer?
Click to expand...
Click to collapse
My thanks count is almost three times as high as yours. Why does everyone assume Octa without HMP is broken? 5420 paces S-800 just fine the way it was released. Maybe to pace S-805 we'll see Samsung move to HMP. But for now there's no motive I can see other than placating a bunch of enthusiasts who somehow feel cheated. As for any "marketing" benefit that's a stretch. Do you think 95% of the mainstream that buy a Samsung eight-core phone do so for that reason? If so making the majority of N3's quad-core was a huge mistake. And whether all eight cores running at the same time means anything outside XDA is highly doubtful.
BarryH_GEG said:
My thanks count is almost three times as high as yours. Why does everyone assume Octa without HMP is broken? 5420 paces S-800 just fine the way it was released. Maybe to pace S-805 we'll see Samsung move to HMP. But for now there's no motive I can see other than placating a bunch of enthusiasts who somehow feel cheated. As for any "marketing" benefit that's a stretch. Do you think 95% of the mainstream that buy a Samsung eight-core phone do so for that reason? If so making the majority of N3's quad-core was a huge mistake. And whether all eight cores running at the same time means anything outside XDA is highly doubtful.
Click to expand...
Click to collapse
Other "majority" with a quadcore s800 only because s800 has LTE in the SoC
Qualcomm sells his s800 at less than double than the single LTE chip
Would you buy a top range phone without lte in states witch have it?
Anyway from heating issue to marketing choice.. well.. if it's only a marketing choice, i have to say fu samsung
Simply, i hope to use my exynos with it's own completely tech.. 5420 has the ability to do..
iba21 said:
Are you a flamer?
1) you sad n900 has thermal dissipation problem.. it's as like as on normal people say the sun is cold..
2) Galaxy s4 coudn't have hmp due to a CCI transistor issue.. it means it's impossible to turn on 2 different cluster or system crashes..
3) i don't know why samsung doesen't releases the hmp until.. it's difficult to compile the code, all the scheduler, with the new function.. it's the FIRST TIME in the world.. so.. there aren't any other base code to copy (ex.. ondemand governor was wrote by intel developers.. INTEL..)
4) samsung sad exynos 5420 is ready for hmp.. new exynos versions? Well where do you see those? Post me an official link, not an ipotetic exynos 6xxx with 64bit arch.. because.. i'm rocco siffredi by using words :thumbup:
The only thing witch n900 doesent have is the widcon technology.. only that..
So.. the situation are 2:
1) samsung will release hmp soon for n900
2) samsung will release hmp for galaxy s5 as a marketing pubblicity
Anyway.. n900 has the worst cpu logic.. core migration should be better.. hmp is the maximize of battery saving with the same task
Click to expand...
Click to collapse
any idea if the octacore version for mediatek helps.. ie. if they releases the source for hmp?? may be jus a noobish question :silly:
cooldoc007 said:
any idea if the octacore version for mediatek helps.. ie. if they releases the source for hmp?? may be jus a noobish question :silly:
Click to expand...
Click to collapse
Hard to say. Their using eight identical A7 chips which makes task migration and scheduling easier.
For the cores MediaTek has opted to use eight Cortex-A7 cores arranged in a big.LITTLE configuration. Which doesn’t make much sense as big.LITTLE is meant to be for a true Heterogeneous Multi-Processing design where some of the cores are faster (i.e. Cortex-A15 cores) than others. It seems that MediaTek has arranged the eight cores in a kind of little.LITTLE arrangement. In this arrangement MediaTek is using its own scheduling algorithm that also monitors temperature and power consumption to ensure optimum performance at all times.http://www.androidauthority.com/mediatek-mt6592-318700/
cooldoc007 said:
any idea if the octacore version for mediatek helps.. ie. if they releases the source for hmp?? may be jus a noobish question :silly:
Click to expand...
Click to collapse
Mediatek uses all the same cluster as octa, it means scheduling doesen't have to switch and chose what core need.. simply.. it's a false big.little.. because it's a little.little!
Probably there are 2 clusters in the core.. and the scheduling logic probably is this:
First case..
there are 1 cluster (4 cortexA7) always on.. the second shutted down by power gating tech.. when scheduling needs more power.. turn on the second cluster
Second case..
All cores are indipendently, so, in the base situation 1 core on and the other shutted down..
When scheduler needs more power.. it activates the other core..
But.. for me the first case will be used.. it's easier to write the code..
###########
Some free informations about big.little tech
###########
Big.little has 2 clusters
http://www.imgur.com/amCLDKU.png
If you open the terminal, and you write:
cat /sys/devices/system/b.L/b.L_core_stat
You'll see what cluster are on.. and.. if you red what i wrote.. you'll see in the 99% of case the A15 cluster..
In that stat you see 0 and 1.. 0 is shutted down, 1 is on..
If you see there are the L2 value too.. 2 clusters has 2 different L2 (cache second level), so, cache is shutted down too..
A7 Cluster has 512kb of cache.. A15 has 2048kb of cache (2mb) (4 time more)
Cache absorbs a lot of energy.. that's one of the cause of the low power consuption by a7 arch..
Well, how to shut down the cluster?
Big.little uses the power gating tech, it's created by intel, and, simply, when there aren't an usage of the core, chip AUTOMATICALLY shut down it..
The old logic uses a dvfs command to say to the chip to shut down the core.. now it's all absolutely automatically.. so.. if scheduler doesen't want to send a taks in one core.. that core will be shutted down until one scheduling call
What's the scheduler?
linux kernel has this code in it witch chose the task destination..
How it work?
It runs the task for little time (ex.. 10milliseconds) and looks about the usage of the cpu.. it trys to find the possible cpu load caused by that task, and it move it into one core witch can compile it without increasing frequency..
So.. the base logic is to always have the minor freq possible
Why?
More freq = more and more power consuption
one chip at 500mhz absorbs less than an half than the same chip at 1000mhz
so.. 4 cores at 500mhz are better than 1 at 2ghz.. absorbs less energy and have the same base computation
Sure.. scheduler is not a mage, it can't be perfect, and sometimes it moves the task in the worst core.. it means, the worst situation for energy efficiency
Anyway.. the actual cluster migration logic chose from 2 cluster.. you could have OR 4A7 OR 4A15
why? Code is easyer to write..
android scheduler is the same as normal scheduler.. there are only an IKS (in kernel switching) module witch chose what cluster use..
From 250mhz to 650mhz there is used A7 cluster
From 700mhz to 1900mhz there is used A15 cluster
Frequency of A7 cluster is an half of the real freq of the chip
So.. 250mhz in kernel value means 500mhz of the chip a7 cluster.. 650mhz of kernel means 1300mhz of the cluster (650 x 2)
A15 has the same frequency as kernel.. so.. 800mhz is really 800mhz..
There are 2 BIG BUGS with that logic:
1) scheduler doesen't have a good code.. it sees 4 cores.. so.. it moves tasks in those 4 cores.. the IKS is indipendent out of the scheduler logic.. so.. there are not so mouch time witch one core has NO LOADING.. so.. the most of the time if the cluster is turn on, all cores are really on
2) it's the conclusion of the first case.. when the A15 are turned on.. A7 are off.. and.. if all tasks force all 4 A15 cores to be active, frequency grow up for all 4 cores.. and.. you have to know that cortexA15 has an HIGH POWER DISSIPATION..
IT REALLY NEEDS A PHYSICAL DISSIPATOR + A FAN AS LIKE PC
In simple words.. the actual cluster migration logic USES TOO MOUCH THE A15 CLUSTER FOR NOTHING..
A car example.. you use the car in a normal street with a low (2nd) gear at 6000rpm at 80km/h (50miles/h)
You'll have an high fuel consuption for nothing
You could increase the gear (5th) and engine will slow down at, for ex, 2000rpm with the same 80km/h
That's a real efficiency
We actual use too mouch cortexA15 cores for nothing.. it means.. spend energy for nothing
The HMP REWRITES THE SCHEDULING LOGIC..
Scheduler knows what are the cores and what architecture have.. so.. the base situation is 1 A7 core.. all other shutted down..
It runs a task in A7 core, looks of it "size", and move into one other A7 core (if size is low) or in one A15 cores (if size is high)
So.. this logic is completely different.. now the scheduler uses less cores as possible, and prefere A7 cores over A15 if they could run the task
Scheduling has a better powergating code.. so.. it choses to turn on one core OR increase frequency of the core..
I did some examples in other posts
Anyway.. HMP tryes to use A7 in the most of the condictions, and, moves sometimes some tasks in 1-2-3-4 A15 cores dipendent of the condiction
I repeat one simple example
CLUSTER MIGRATION (ACTUAL)
touch the screen, 4 A15 on
Web rendering (loading) 4A7 on (in the most of the case)
Web page panning (sliding) 4A15 on
HMP
Tuch the screen 4 A7 on and SOMETIMES one A15
Web rendering 4 A7 on and SOMETIMES 1-2 A15 ON
Web page panning 4A7 on and 1-2 A15 on..
Why in hmp always 4 A7 on?
Because OS has a lot of tasks to do.. low tasks (as like sensors) are compiled by A7.. and huge tasks by A15
In the actual cluster migration, if you reach more than 650mhz (always if you need to compile 1 stupid high task) all 4 A15 are on even to compile other stupid tasks.. so.. 1huge task increases the power consuption for all other 300 little tasks
With HMP.. 300 little tasks are compiled by A7 cluster.. and 1 huge by 1 A15 cluster
THAT'S WHAT I CALL EFFICIENCY
Sure.. more efficiency = less energy drain = less cpu heating
Sorry for my kilometrical post.. i hope someone read it, and understand what are the benefit of this "code"
iba21 said:
Mediatek uses all the same cluster as octa, it means scheduling doesen't have to switch and chose what core need.. simply.. it's a false big.little.. because it's a little.little!
Probably there are 2 clusters in the core.. and the scheduling logic probably is this:
First case..
there are 1 cluster (4 cortexA7) always on.. the second shutted down by power gating tech.. when scheduling needs more power.. turn on the second cluster
Second case..
All cores are indipendently, so, in the base situation 1 core on and the other shutted down..
When scheduler needs more power.. it activates the other core..
But.. for me the first case will be used.. it's easier to write the code..
###########
Some free informations about big.little tech
###########
Big.little has 2 clusters
http://www.imgur.com/amCLDKU.png
If you open the terminal, and you write:
cat /sys/devices/system/b.L/b.L_core_stat
You'll see what cluster are on.. and.. if you red what i wrote.. you'll see in the 99% of case the A15 cluster..
In that stat you see 0 and 1.. 0 is shutted down, 1 is on..
If you see there are the L2 value too.. 2 clusters has 2 different L2 (cache second level), so, cache is shutted down too..
A7 Cluster has 512kb of cache.. A15 has 2048kb of cache (2mb) (4 time more)
Cache absorbs a lot of energy.. that's one of the cause of the low power consuption by a7 arch..
Well, how to shut down the cluster?
Big.little uses the power gating tech, it's created by intel, and, simply, when there aren't an usage of the core, chip AUTOMATICALLY shut down it..
The old logic uses a dvfs command to say to the chip to shut down the core.. now it's all absolutely automatically.. so.. if scheduler doesen't want to send a taks in one core.. that core will be shutted down until one scheduling call
What's the scheduler?
linux kernel has this code in it witch chose the task destination..
How it work?
It runs the task for little time (ex.. 10milliseconds) and looks about the usage of the cpu.. it trys to find the possible cpu load caused by that task, and it move it into one core witch can compile it without increasing frequency..
So.. the base logic is to always have the minor freq possible
Why?
More freq = more and more power consuption
one chip at 500mhz absorbs less than an half than the same chip at 1000mhz
so.. 4 cores at 500mhz are better than 1 at 2ghz.. absorbs less energy and have the same base computation
Sure.. scheduler is not a mage, it can't be perfect, and sometimes it moves the task in the worst core.. it means, the worst situation for energy efficiency
Anyway.. the actual cluster migration logic chose from 2 cluster.. you could have OR 4A7 OR 4A15
why? Code is easyer to write..
android scheduler is the same as normal scheduler.. there are only an IKS (in kernel switching) module witch chose what cluster use..
From 250mhz to 650mhz there is used A7 cluster
From 700mhz to 1900mhz there is used A15 cluster
Frequency of A7 cluster is an half of the real freq of the chip
So.. 250mhz in kernel value means 500mhz of the chip a7 cluster.. 650mhz of kernel means 1300mhz of the cluster (650 x 2)
A15 has the same frequency as kernel.. so.. 800mhz is really 800mhz..
There are 2 BIG BUGS with that logic:
1) scheduler doesen't have a good code.. it sees 4 cores.. so.. it moves tasks in those 4 cores.. the IKS is indipendent out of the scheduler logic.. so.. there are not so mouch time witch one core has NO LOADING.. so.. the most of the time if the cluster is turn on, all cores are really on
2) it's the conclusion of the first case.. when the A15 are turned on.. A7 are off.. and.. if all tasks force all 4 A15 cores to be active, frequency grow up for all 4 cores.. and.. you have to know that cortexA15 has an HIGH POWER DISSIPATION..
IT REALLY NEEDS A PHYSICAL DISSIPATOR + A FAN AS LIKE PC
In simple words.. the actual cluster migration logic USES TOO MOUCH THE A15 CLUSTER FOR NOTHING..
A car example.. you use the car in a normal street with a low (2nd) gear at 6000rpm at 80km/h (50miles/h)
You'll have an high fuel consuption for nothing
You could increase the gear (5th) and engine will slow down at, for ex, 2000rpm with the same 80km/h
That's a real efficiency
We actual use too mouch cortexA15 cores for nothing.. it means.. spend energy for nothing
The HMP REWRITES THE SCHEDULING LOGIC..
Scheduler knows what are the cores and what architecture have.. so.. the base situation is 1 A7 core.. all other shutted down..
It runs a task in A7 core, looks of it "size", and move into one other A7 core (if size is low) or in one A15 cores (if size is high)
So.. this logic is completely different.. now the scheduler uses less cores as possible, and prefere A7 cores over A15 if they could run the task
Scheduling has a better powergating code.. so.. it choses to turn on one core OR increase frequency of the core..
I did some examples in other posts
Anyway.. HMP tryes to use A7 in the most of the condictions, and, moves sometimes some tasks in 1-2-3-4 A15 cores dipendent of the condiction
I repeat one simple example
CLUSTER MIGRATION (ACTUAL)
touch the screen, 4 A15 on
Web rendering (loading) 4A7 on (in the most of the case)
Web page panning (sliding) 4A15 on
HMP
Tuch the screen 4 A7 on and SOMETIMES one A15
Web rendering 4 A7 on and SOMETIMES 1-2 A15 ON
Web page panning 4A7 on and 1-2 A15 on..
Why in hmp always 4 A7 on?
Because OS has a lot of tasks to do.. low tasks (as like sensors) are compiled by A7.. and huge tasks by A15
In the actual cluster migration, if you reach more than 650mhz (always if you need to compile 1 stupid high task) all 4 A15 are on even to compile other stupid tasks.. so.. 1huge task increases the power consuption for all other 300 little tasks
With HMP.. 300 little tasks are compiled by A7 cluster.. and 1 huge by 1 A15 cluster
THAT'S WHAT I CALL EFFICIENCY
Sure.. more efficiency = less energy drain = less cpu heating
Sorry for my kilometrical post.. i hope someone read it, and understand what are the benefit of this "code"
Click to expand...
Click to collapse
I really enjoy reading your explanations and learning about HMP.
I hope Samsung realeases this for the exynos variant of the Note 3.
2 examples :
http://www.imgur.com/vmXPhNx.png
Normal status.. nothing open.. only system tuner.. airplane mode.. ALWAYS 4 CORTEXA15 on
http://www.imgur.com/P2HFAxp.png
Same condiction with maxfreq of 650mhz (1.3ghz of cortexa7)
More apps opened, no airplane mode.. ALWAYS CORTEXA7
....
Why?
When i touch the screen to press the run button, cpu goes over 650mhz and switch to cortexA15.. after that come back..
If i force maxfreq to 650mhz.. iks can't switch to cortexA15
It's the opposite of hmp solution! Ahahahahaha
Now it's easyer to understand?
That's why i need it..
If you woukd do the same test.. download system tuner and run that string in terminal.. you'll see
After that, if you have root.. go to CPU tab, move maxfreq to 650mhz.. and redo terminal command.. you'll have the same results as mine!!
This is and this will be until hmp upgrade (only kernel upgrade)
Sure.. i remember, more time in A15 = more energy absorb = more heat..
Now.. open this video
https://www.youtube.com/watch?v=wYMnu4L2-aU&feature=youtube_gdata_player
And look.. guy press the hone button.. 4A7 on + 1 A15 for 500ms..
Do you think 4A7+ 1A15 absorbs more energy than 4A15?
Well.. SAMSUNG PLEASE RELEASE HMP
Some weeks ago @Quarx told me that he tried to overclock his MotoG. At a first look it works. The system is stable and cpuinfo apps show 1.9GHZ.
BUT: there isn't any performance change. Then I checked it on my device(Xiaomi Mi2) - exactly the same result.
I watched a video about Nexus4 overclock: https://www.youtube.com/watch?v=96hsPe-JH2M
As you can see there isn't any change, too(the minor differences are just random).
Then Quarx found a very interesting commit in MotoG's kernel log:
https://github.com/Quarx2k/kernel-moto-g/commit/ed82c3f26ec4a7ab39a436647643bdaf71dcb67a
The message tells us, that QCOM devices normally are "screened" which basically means, that if your max freq is 1.2GHZ, it doesn't matter what you change in the kernel source. it will always be 1.2GHZ even if it shows 100GHZ because the SoC is limiting that.
There are some older SoC's with adreno 2xx which seem to be overclockable without any problems.
But all new SoCs seem to be screened which prevents overclocking.
Some more tests of other devices and SoC's would be helpful.
I'd recommend you get a buttload of single-thread Linpack benchmarks at different frequencies and do some statistical analysis. Maybe make a Google Form where you can input Device, Frequency, and MFLOPS
Linpack because it's exclusively CPU, doesn't vary much between devices of the same model, and scales nicely with frequency.
scy1192 said:
I'd recommend you get a buttload of single-thread Linpack benchmarks at different frequencies and do some statistical analysis. Maybe make a Google Form where you can input Device, Frequency, and MFLOPS
Linpack because it's exclusively CPU, doesn't vary much between devices of the same model, and scales nicely with frequency.
Click to expand...
Click to collapse
I think Linpack isn't reliable enough because it's completed too fast. For me it takes 1-2s to complete and the result variate's very much.
Sth like Antutu(only the CPU score) with performance governor and maybe even on a clean install would be better.
m11kkaa said:
I think Linpack isn't reliable enough because it's completed too fast. For me it takes 1-2s to complete and the result variate's very much.
Sth like Antutu(only the CPU score) with performance governor and maybe even on a clean install would be better.
Click to expand...
Click to collapse
So how we can really test it? test in game to see if FPS get more high? I really dont believe that is just a placebo effect, in my kernel with 1.5Ghz I really feel the speed and will test when i finish the OC of 1.8Ghz.
BryanByteZ said:
So how we can really test it? test in game to see if FPS get more high? I really dont believe that is just a placebo effect, in my kernel with 1.5Ghz I really feel the speed and will test when i finish the OC of 1.8Ghz.
Click to expand...
Click to collapse
well I only said that linpack looks unreliable. Antutu or other cpu intensive benchmarks should work just fine. Just ignore gpu benchmarks, because the gpu is definitely overclockable.