Who knows we all love to tinker, maybe a dev will develop a Kernel that enables it. I wouidt trust Samsung to do it LOL
Sent from my SM-P600 using XDA Premium 4 mobile app
8 core update for exynos? !
- Thread starter Ahmedmoataz
- Start date
Who knows we all love to tinker, maybe a dev will develop a Kernel that enables it. I wouidt trust Samsung to do it LOL
Sent from my SM-P600 using XDA Premium 4 mobile app
articles regarding note 3 exynos and HMP
http://www.nextpowerup.com/news/413...oftware-patch-which-unlocks-8-core-power.html
in russian , but use google translate, you get the point
http://hi-tech.mail.ru/news/misc/samsung-no-real-octacore.html
http://www.nextpowerup.com/news/413...oftware-patch-which-unlocks-8-core-power.html
in russian , but use google translate, you get the point
http://hi-tech.mail.ru/news/misc/samsung-no-real-octacore.html
It is completely false to say that HMP would impact battery, if anything the opposite is true. There are very few applications which could even in principle use all 8 cores. Instead most of the times 3 is where apps max out, which would make little sense to enable HMP for performance reasons. But it makes all the sense in the world to enable it for battery reasons.
Samsung released a video a couple of months ago showing just that. Having all 8 cores enabled at the same time means that you have the 4 little one take most of the workload and the 4 big ones only be enabled as needed. This would result in unbelievable battery savings. Imagine an occasion where 4 little cores are not enough, what does the current implementation of big.little does? Oh but enable the 4 big cores disabling the little ones, so it ends up at times to light all 4 big cores when 4 little ones + 1 big one would be enough, this results in far lower battery.
HMP is a great idea both in paper and in practice, see the new Note Neo, it's using HMP to perfection and manages with "lower" hardware to perform as well as it's elder brother as well as be more energy efficient. *This* should be our example. Exynos 5420 is HMP compatible and it would only take a dev to enable it. I don't think that it will happen, but *if* it will it would instantly lead to far better devices...
Note 2014 has rather mediocre battery, my gf's ipad mini with almost half the MAHs has more juice. That's partly because the exynos chip is not very energy efficient. Enable all 8 cores and it will become that...
BTW the Note Neo sources (with HMP enabled) were just released. If we had a more active community I would expect an HMP patch coming from the community soon. Sadly seeing how badly misinformed most people are about HMP I don't think anyone would put the effort required and we'd stay with gimped devices ... oh well.
Thanks to god there are people who talk about something with knowledge... Tired of people saying 8 cores will kill the battery and are unnecesary.
I own the qualcomm model but I hope exynos users will have HMP sooner as possible.
Neither the A7 or A15 clusters run "all or nothing;" meaning you don't need HMP to run between 1-4 cores on each individual cluster. In your example, what's the energy saving between running 4 A7 + 1 A15 vs. running two A15? I don't know the answer but that's the delta as to what type of energy savings can be achieved by picking and choosing across clusters vs. stepping between 1 and 4 active cores within an individual cluster. At the end of the day each individual A7 and A15 core contributes a specific amount of performance and consumes a specific amount of energy. Software engineers set performance in steps, probably between 6-8, with each threshold using a prescribed number of cores to achieve a set performance output (and consume a set amount of energy). Today they're limited to 1-4 A7 cores or 1-4 A15 cores. Not knowing what those 6-8 stepped thresholds are who's to say what the definitive impact on energy use would be if they could choose between 1-4 A7 cores and 1-4 A15 cores?
Incidentally, Media Tek's HMP big.LITTLE SoC runs eight A7 cores (two quad A7 clusters). No one else licensing big.LITTLE is using HMP across different types of cores (EG: A7 vs. A15).
http://www.fonearena.com/blog/88620...-octa-core-chip-offers-speeds-up-to-2ghz.html
My undergrad is computer electronics (though over ten years ago). Tech has changed a lot, but laws of electrical physics have stayed constant. Saying it is completely false seems as scientifically dubious as saying it is completely true. Samsung was concerned about thermals with the current devices. Seems that ties with likely more sustained peak watt consumption, so warmer running and more potential for battery consumption.
With the current SOC being a 28 nM build, how does higher performance not equal more heat and less battery? Respectfully does not add up. Added: Unless the nM goes down in correlation to increased speed which can not be the case. That is unless the core mix results in an exponential speed curve, which would also be very unlikely.
Again, the premise of point here is in regard to faster performance and not power efficency.
Added 2 : BarryH does a fine job of articulating. The benefit would be efficiency and not a speed boost that would not come with a thermal and battery life cost, since the fixed constant is the 28nM.
With the current SOC being a 28 nM build, how does higher performance not equal more heat and less battery? Respectfully does not add up. Added: Unless the nM goes down in correlation to increased speed which can not be the case. That is unless the core mix results in an exponential speed curve, which would also be very unlikely.
Again, the premise of point here is in regard to faster performance and not power efficency.
Added 2 : BarryH does a fine job of articulating. The benefit would be efficiency and not a speed boost that would not come with a thermal and battery life cost, since the fixed constant is the 28nM.
Last edited:
Here are some examples. Under light load the A7 cluster is being used with each core's output changing individually and dynamically based on demand. Under heavy load all four A15 cores are outputting at max. It's hard to capture pics of the individual cores output fluctuating at idle or light load within each cluster because using S Pen to take a screen cap ramps up load. But if you want to see the interplay between cores within each cluster as demand changes download CPU-Z and play with some different apps on your device.
Having nothing to do with Samsung Electronics who wants to sell mobile devices and is SoC agnostic, Samsung Semiconductor who fiercely competes with Qualcomm would sell their left testicle to develop a chip that provides near-identical performance to S-800 while offering 25% better energy usage. And if HMP was the "Holy Grail" and "silver bullet" that allowed that to happen it would be deployed now and mobile manufacturers would be lined up to buy it.
And this is from a Qualcomm "anti-Octa" video. 85% of apps in use don't use more than two of the four cores on Qualcomm's chips.
In the Octa-world that translates to "never leaves the A7 cluster."
So from a performance perspective adding the A7 cores on top of the A15 cores under high-demand is kind of like turbo charging and a pretty clear benefit (energy use and thermal issues aside). But outside that HMP for pure energy savings doesn't seem to add a lot of value. Which is probably why it isn't implemented.
@ barry, Rushless: Per-watt performance of A7 cores is greater than that of A15 cores, I'm relatively short of time so I cannot find sources, but when I'll return I'll you provide you some if you wish. That means quite clearly that any combination of A15 and A7 is *always* more energy efficient than A15 cores running alone. That is one...
The second is that if you run both A15 and A7 cores at once you're going to run peak watt situations that's to be sure. But apart from Benchmarks and maybe some games there are not many use cases that can achieve that. Still if it is determined that running all 8 cores exceeds the thermal envelope of a tablet (remember we are not talking about a phone here, Note 10.1 is a tablet) you can *always* throttle the cpus forcing them to run at a lower wattage. In other words cap the energy requirements of the cpu.
Those (1 and 2) would result to better battery at no reliability cost. It's a win-win scenario that it is seriously beyond me why is it not attempted by Samsung. I honestly have to think that there is more in play than simply "can't be done". The far more thermally constricted Note Neo can run 6 cores (4 small, 2 big) at once no problem, so it is no stretch to think that a tablet could run maybe 4 small cores + 3 big ones before throttling gets in place, still almost as energy efficient as running 4 A15 cores, at a higher performance.
I can see no downsides, I honestly can't...
The second is that if you run both A15 and A7 cores at once you're going to run peak watt situations that's to be sure. But apart from Benchmarks and maybe some games there are not many use cases that can achieve that. Still if it is determined that running all 8 cores exceeds the thermal envelope of a tablet (remember we are not talking about a phone here, Note 10.1 is a tablet) you can *always* throttle the cpus forcing them to run at a lower wattage. In other words cap the energy requirements of the cpu.
Those (1 and 2) would result to better battery at no reliability cost. It's a win-win scenario that it is seriously beyond me why is it not attempted by Samsung. I honestly have to think that there is more in play than simply "can't be done". The far more thermally constricted Note Neo can run 6 cores (4 small, 2 big) at once no problem, so it is no stretch to think that a tablet could run maybe 4 small cores + 3 big ones before throttling gets in place, still almost as energy efficient as running 4 A15 cores, at a higher performance.
I can see no downsides, I honestly can't...
Absolutely. It's the thesis behind big.LITTLE. But the maximum output of A7 is limited because it's tuned for efficiency.
Look at it this way. Both Qualcomm and ARM say in different ways that 85% of use cases stay on two Qualcomm cores and never leave A7 on big.LITTLE. Let's be conservative and say it's really 75%. Let's assume that under high demand 10% of what's being processed never leaves A15. In both those conservative scenarios which comprise 85% of SoC usage HMP is meaningless. That leaves a potential 15% of use cases that might benefit from some combination of A7 and A15 rather than being confined to a single cluster. Let's also assume that in that scenario introducing HMP improves the energy use of the 15% by 25%. In the overall scheme of things in terms of total energy usage (battery life) that's not very meaningful and probably why there's been no rush to implement HMP (other than by Media Tek as a marketing grab). Perhaps the engineers looked at the potential gain and decided it wasn't worth the added complexity; especially if 25% less energy draw on 15% of use cases is aggressive which if I had to guess I'd say it probably is. In other words, playing with tuning and optimization within each cluster nets you similar performance vs. energy draw more simply than adding the complexity of HMP.
It would be nice to have the option to at least compare it on and off 
My point was really in regards to some folks thinking having both cores running full at the same time would be a good thing when it would not, based on how warm and how quick the 2014 can already drain battery. That is all. Core juggling all eight in some balanced fashion is different. Seems already doing that though. Question being is the power gate in series to each core, or in series to each set of four? If in series to each core, that would have a better power advantage, but would be an odd design to standard.
Samsung did mention that thermals with current devices was a reason not an option for now.
My point was really in regards to some folks thinking having both cores running full at the same time would be a good thing when it would not, based on how warm and how quick the 2014 can already drain battery. That is all. Core juggling all eight in some balanced fashion is different. Seems already doing that though. Question being is the power gate in series to each core, or in series to each set of four? If in series to each core, that would have a better power advantage, but would be an odd design to standard.
Samsung did mention that thermals with current devices was a reason not an option for now.
Last edited:
@ barry: I have to dispute that in 85% of the situations the little cluster is enough. My observations showed an about even 50-50 distribution between the big and the little cluster. Let's say now that my usage of the machine is untypical, still you have to concede that a *tablet* is used differently than a phone and qualcomm's (or whomever's) calculations would had been different if done to a tablet.
Now let us say that 4 little cores can net the performance of 2 big cores at 66% of energy usage (even though it is more like 50%) that in turn would mean 8%-16% better battery for most users, or maybe even closer to 12%-25%. Even if just 8% is given to the user it means he will have 7% left in the tank by the time a non-hmp tablet would shut down and it can be scaled up to 20-25% left in the tank when heavy usage is concerned, that's not little by any means and since exynos socs rather suffer battery wise I dont see why this is not pursued in tablets at least.
It is not a hard implementation considering it has already been implemented hardware wise. All it needs is a patch to be released which will say that when more than 4 little cores and 3 big ones are maxed for more than a few seconds throttle the machine. This is another thing that has already be done, try to get a higher clock from exynos 5420 and see what happens (it is throttled even if you run at lower wattage)
@rushless: Again I have to say that I'm more concerned with higher efficiency than higher performance. What Samsung mentioned sounds more like sth to say to keep people calm. There are no thermal consideration to the much smaller note 3 neo, why should there be any to a much better ventilated/roomy tablet especially when throttling is taken into consideration.
@ dully.noted : Sure take a look at note neo's and note 3's battery review
Note 3: http://www.gsmarena.com/samsung_galaxy_note_3-review-996p3.php
Note neo: http://www.gsmarena.com/samsung_galaxy_note_3_neo-review-1038p3.php
Look at 3g and Web browsing numbers where is where the cpu is mostly used (the playback test tests screen technology efficiency which is not what we care about here).
Keep in mind that the Note 3 in question is using the even more efficient qualcomm soc than exynos one and still manages to come short compared to hmp, despite using a bigger battery (3200 mah). What's more according to antutu and most multi threaded cpu tests note neo's cpu delivers very equivalent results to note 3's at a much more efficient manner.
...again I don't see why hmp is not enabled given that all it needs is a darn patch ...
Now let us say that 4 little cores can net the performance of 2 big cores at 66% of energy usage (even though it is more like 50%) that in turn would mean 8%-16% better battery for most users, or maybe even closer to 12%-25%. Even if just 8% is given to the user it means he will have 7% left in the tank by the time a non-hmp tablet would shut down and it can be scaled up to 20-25% left in the tank when heavy usage is concerned, that's not little by any means and since exynos socs rather suffer battery wise I dont see why this is not pursued in tablets at least.
It is not a hard implementation considering it has already been implemented hardware wise. All it needs is a patch to be released which will say that when more than 4 little cores and 3 big ones are maxed for more than a few seconds throttle the machine. This is another thing that has already be done, try to get a higher clock from exynos 5420 and see what happens (it is throttled even if you run at lower wattage)
@rushless: Again I have to say that I'm more concerned with higher efficiency than higher performance. What Samsung mentioned sounds more like sth to say to keep people calm. There are no thermal consideration to the much smaller note 3 neo, why should there be any to a much better ventilated/roomy tablet especially when throttling is taken into consideration.
@ dully.noted : Sure take a look at note neo's and note 3's battery review
Note 3: http://www.gsmarena.com/samsung_galaxy_note_3-review-996p3.php
Note neo: http://www.gsmarena.com/samsung_galaxy_note_3_neo-review-1038p3.php
Look at 3g and Web browsing numbers where is where the cpu is mostly used (the playback test tests screen technology efficiency which is not what we care about here).
Keep in mind that the Note 3 in question is using the even more efficient qualcomm soc than exynos one and still manages to come short compared to hmp, despite using a bigger battery (3200 mah). What's more according to antutu and most multi threaded cpu tests note neo's cpu delivers very equivalent results to note 3's at a much more efficient manner.
...again I don't see why hmp is not enabled given that all it needs is a darn patch ...
Last edited:
Two posts above you : http://www.gsmarena.com/samsung_galaxy_note_3_neo-review-1038p6.php
Not only it does have HMP but it also posts better battery and performance than equivalent hardware. What is more its sources are released a talented Dev (ala patwip in the original galaxy phone) may convert it for use to exynos 5420, letting us enjoy all the benefits associated with it, but I kinda doubt it will happen...
Nothing you can buy from Samsung has HMP enabled. I hope that it worded better.
Sent from my SM-T320 using XDA Premium HD app
As far as we are concerned though (according to the original question) there is an HMP implementation out there. It's in the source code of Note 3 Neo, maybe it's not directly transferable to our devices, but there is a way to study the nuts and bolts of HMP.
As for the device itself, it will be out within this coming week in certain places.
As it has been stated earlier in this post, It does not benefit Samsung in any way so I do not see it happening on their part.
Similar threads
Top Liked Posts
-
There are no posts matching your filters.
-
7
It is completely false to say that HMP would impact battery, if anything the opposite is true. There are very few applications which could even in principle use all 8 cores. Instead most of the times 3 is where apps max out, which would make little sense to enable HMP for performance reasons. But it makes all the sense in the world to enable it for battery reasons.
Samsung released a video a couple of months ago showing just that. Having all 8 cores enabled at the same time means that you have the 4 little one take most of the workload and the 4 big ones only be enabled as needed. This would result in unbelievable battery savings. Imagine an occasion where 4 little cores are not enough, what does the current implementation of big.little does? Oh but enable the 4 big cores disabling the little ones, so it ends up at times to light all 4 big cores when 4 little ones + 1 big one would be enough, this results in far lower battery.
HMP is a great idea both in paper and in practice, see the new Note Neo, it's using HMP to perfection and manages with "lower" hardware to perform as well as it's elder brother as well as be more energy efficient. *This* should be our example. Exynos 5420 is HMP compatible and it would only take a dev to enable it. I don't think that it will happen, but *if* it will it would instantly lead to far better devices...
Note 2014 has rather mediocre battery, my gf's ipad mini with almost half the MAHs has more juice. That's partly because the exynos chip is not very energy efficient. Enable all 8 cores and it will become that...
BTW the Note Neo sources (with HMP enabled) were just released. If we had a more active community I would expect an HMP patch coming from the community soon. Sadly seeing how badly misinformed most people are about HMP I don't think anyone would put the effort required and we'd stay with gimped devices ... oh well.2
Of course it can handle it. The rest is really just up to samsung whether they decide to roll out Linux 3.10+ for our device. Alternatively someone could simply backport the patchset from Linaro to Linux 3.4 or compile an actual 3.10+ kernel that boots on the SM-P600.
On a similar note, I recently tested Linux 3.14 with the normal IKS configuration (based on Samsung's defconfig for the SM-P600 with some updated values from the config for the Arndale Octa - an Exynos5420 dev board) but while it compiles just fine it doesn't get past the stock Samsung bootloader, and I'm sure there are some drivers missing as well. The only stuff I have currently merged in are the proprietary ARM Mali gpu drivers.
If I remember correctly dorimanx managed to get the 3.14 kernel to work with AOSP on the SGS2 but I'm not quite sure how he went about it and I'm too much of a kernel newbie to figure it out myself. If anyone here has any suggestions it would be deeply appreciated. Source: http://github.com/sigma-1/linux-linaro-3.142Here's your HMP. You just need to buy a device with Exynos 5422 (we have 5420) to get it.
Samsung announced two new mobile SoCs at MWC today. The first is an update to the Exynos 5 Octa with the new Exynos 5422. The 5422 is a mild update to the 5420, which was found in some international variants of the Galaxy Note 3. The new SoC is still built on a 28nm process at Samsung, but enjoys much higher frequencies on both the Cortex A7 and A15 clusters. The two clusters can run their cores at up to 1.5GHz and 2.1GHz, respectively.
The 5422 supports HMP (Heterogeneous Multi-Processing), and Samsung LSI tells us that unlike the 5420 we may actually see this one used with HMP enabled. HMP refers to the ability for the OS to use and schedule threads on all 8 cores at the same time, putting those threads with low performance requirements on the little cores and high performance threads on the big cores.
http://www.anandtech.com/show/7811/samsungs-exynos-5422-the-ideal-biglittle-exynos-5-hexa-52601aac and ac3 seem to cause the most problems SD and HD. I generally keep the nitrate around 2.5 Mb/s just for portability sake. container doesn't matter as avi, mkv, and mp4 all have the drops. I've used mx, BS, Archos, xbmc, and a few others and installed custom codecs if needed. heck even Netflix has a few frame drops. maybe I'm just too sensitive to it, but my s4, gn2, and gn8 have no problems. my old gn10.1 had no problems either. this 10.1 2014 is also the first Samsung device I've had that doesn't natively support ac3. hardware playback with alternative players do play the ac3 audio, but with dropped frames.1My undergrad is computer electronics (though over ten years ago). Tech has changed a lot, but laws of electrical physics have stayed constant. Saying it is completely false seems as scientifically dubious as saying it is completely true. Samsung was concerned about thermals with the current devices. Seems that ties with likely more sustained peak watt consumption, so warmer running and more potential for battery consumption.
With the current SOC being a 28 nM build, how does higher performance not equal more heat and less battery? Respectfully does not add up. Added: Unless the nM goes down in correlation to increased speed which can not be the case. That is unless the core mix results in an exponential speed curve, which would also be very unlikely.
Again, the premise of point here is in regard to faster performance and not power efficency.
Added 2 : BarryH does a fine job of articulating. The benefit would be efficiency and not a speed boost that would not come with a thermal and battery life cost, since the fixed constant is the 28nM.
