Sign in with
Sign up | Sign in

Samsung's Exynos 5 Octa Functions as a Quad-Core Processor, Says Developer

By - Source: VR-Zone | B 23 comments

According to developer AndreiLux, a poor implementation of the big.LITTLE driver means that a maximum of four of the Samsung Exynos 5 Octa’s cores are actually operational.

Though the Samsung Exynos 5 featured in the Galaxy S4 is technically an octa-core processor, Android kernel developer AndreiLux has reported that only four of its cores work at any given point in time. The developer attributes this to a remarkably poor implementation of the big.LITTLE driver, whereby the Linux kernel doesn’t differentiate between the various architecture based cores, and the following takes place:

  • Cluster migration: When load rises, just like when a phone increases its frequency per DVFS, all the cores will jump from the LITTLE processor to the big one. You either have four A7 cores or four A15 cores online.
  • Core migration: When load rises, it will only jump those cores from the LITTLE processor whose load exceeds the capacity of the A7 cores; you can have a single A15 online with three A7′s on, or any other combination of a total of four processors.
  • HMP (Heterogeneous Multi-Processing): This is a vastly more complex working mechanism, and its implementation is also an order of more sophisticated magnitude. It requires the kernel scheduler to actually be aware of the differentiation between the A7 and A15 cores. Currently, the Linux kernel is not capable of doing this and treats all CPUs as equals. This is a problem since we do not want to use the A15 cores when a task can simply be processed on an A7 core with a much lower power cost.

Thankfully, this problem can be resolved by simply updating the kernel so the Exynos 5’s eight physical cores operate in tandem and allocate more demanding tasks to the chip’s more powerful cores. Doing so should result in significant performance gains and improvements to battery life which were not, coincidentally, Samsung’s stated reasons for using an octa-core processor in the first place.

Display 23 Comments.
This thread is closed for comments
Top Comments
  • 12 Hide
    tipoo , May 5, 2013 10:19 AM
    This is what bugged me about every blog calling it an octo core, sure it technically has eight cores in there but only four are ever active at once.
Other Comments
  • 9 Hide
    DRosencraft , May 5, 2013 9:12 AM
    Driver development and implementation continues to be the bane of technology development. I feel for the people who have to read through and write all that code. When it doesn't work we get pissed and tend to forget the work that went into it. Usually it is something relatively small that makes a huge difference. Just glad they can fix this one without too much headache.
  • 0 Hide
    siddallj , May 5, 2013 9:25 AM
    Will we be able to get the 8 core S4 in the UK, im not bothered if it doesn't have 4G, i live in Wales, we're not gonna get it till S5/S6 time
  • 12 Hide
    tipoo , May 5, 2013 10:19 AM
    This is what bugged me about every blog calling it an octo core, sure it technically has eight cores in there but only four are ever active at once.
  • -2 Hide
    ojas , May 5, 2013 10:22 AM
    Har har. Who's surprised, except the Tom's news team and Zak Islam.
    +1 to what tipoo said.
  • 6 Hide
    vmem , May 5, 2013 10:27 AM
    I think most of us figured this was the case just by reading the release specs on that processor. besides, it would be a bit strange for cellphone apps to be suddenly capable of using 8 cores when half of the phones are still dual-core. just remember how long it took us to make that jump on PCs. of course, it'll be faster, but we still need the majority of users to upgrade first before developers will bother doing it
  • 1 Hide
    guru_urug , May 5, 2013 10:54 AM
    Am I missing something here or is there a glaring mistake in the table? Shouldn't the 1st quad-core be mentioned as being based on the cortex A15 architecture?
  • 2 Hide
    guru_urug , May 5, 2013 11:48 AM
    Oops I just realised, they meant the instruction set which is indeed ARM v7a. In my haste I read it A7. My bad, guess I need some sleep.
  • 3 Hide
    rwinches , May 5, 2013 1:51 PM
    The big/little is about power/efficiency. There is no driver issue. I think this is great. ARM is moving at a lighting pace. Yeah OK so at some point there will be powerful multi-core low drain Soc's. Many commenters keep wanting more complexity and faster cores yet claim they hate bloat, long boot times and short battery life.
  • 0 Hide
    soundping , May 5, 2013 2:13 PM
    I wish I'd never bought a SHV-E300S. Biggest waste of money.
  • -7 Hide
    redeemer , May 5, 2013 2:49 PM
    With Tegra 4i and SnapDragon 800 around the corner how cares about Exynos 5, seriously.
  • 3 Hide
    alextheblue , May 5, 2013 3:11 PM
    Rwinches, that's exactly what I was thinking. Isn't this how big.LITTLE was intended to function? 4 cores at a time, which ones are in use depends on load? It's simpler and safer that way, plus the apps don't need to be aware.
    The real problem is that they started marketing it as Octa-core. Stupid marketing mistake - all the geeks knew it was 4 cores at a time, with idle and near-idle power consumption being much lower than running high powered cores all the time.
    Nvidia didn't advertise their Tegra 3 as being a Penta-core CPU, and for good reason.
  • 0 Hide
    Xu En , May 5, 2013 6:16 PM
    http://www.arm.com/files/downloads/big_LITTLE_Final_Final.pdf
    quote (page 7)
    "Since a big.LITTLE system containing Cortex-A15 and Cortex-A7 is fully coherent through CCI-400 another logical use-model is to allow both Cortex-A15 and Cortex-A7 to be powered on and simultaneously executing code. This is termed big.LITTLE MP, which is essentially Heterogeneous Multi-Processing."
  • 0 Hide
    Xu En , May 5, 2013 6:18 PM
    http://www.arm.com/files/downloads/big_LITTLE_Final_Final.pdf
    quote (page 7)
    "Since a big.LITTLE system containing Cortex-A15 and Cortex-A7 is fully coherent through CCI-400 another logical use-model is to allow both Cortex-A15 and Cortex-A7 to be powered on and simultaneously executing code. This is termed big.LITTLE MP, which is essentially Heterogeneous Multi-Processing."
  • 0 Hide
    rdc85 , May 5, 2013 7:58 PM
    This make me remember the old days when first hyper threading, multi core, or amd module core...
    they had the hardware but the software is still not able to utilize it ....
    (in the other hand without the hardware, no way developer can test/implement better code)
  • 0 Hide
    rwinches , May 5, 2013 11:57 PM
    Xu En Yes MP for Big.LITTLE is not excluded.
    SCSI spec allows for 16 devices per channel because each address has two sub addresses. The first and only implementation of this I came across was in '94 in a CD ROM tower using tech from ATTO. Just because a configuration it available does not necessarily mean it will or should be used.
  • 0 Hide
    Djhg2000 , May 6, 2013 12:19 AM
    Can't they just cheat and use cgroups to prioritize the cores?
  • 1 Hide
    mjv1121 , May 6, 2013 1:38 AM
    It is operating as correctly as designed. It is a big OR LITTLE operation. The idea is to allow for high performance when required (big=A15s) while keeping overall power usage to a minimum LITTLE=A7s). IT WORKS => pocketnow found that the octa S$ uses less power than the S600 Quad. People's capacity for willfully failing to comprehend was has been explained by ARM and Samsung a thousand times, never ceases to annoy.
  • 0 Hide
    tului , May 6, 2013 5:04 AM
    Waiting to replace my Note 2 with the Note 3. Hopefully they can get the Exynos in the US versions like the did with the 2. Although a Snapdragon 800 wouldn't be too bad I suppose.
  • 0 Hide
    Non-Euclidean , May 6, 2013 6:05 AM
    Quote:
    With Tegra 4i and SnapDragon 800 around the corner how cares about Exynos 5, seriously.


    Gee, I dont know. Maybe the company that makes the most smartphones out there, which also has the largest market share? You know, Samsung. Maybe they care about the perfomance of their processor in their phones. Maybe anyone who will be buying a Samsung phone in the future cares also?

  • 0 Hide
    kartu , May 6, 2013 6:13 AM
    Quote:
    ...this problem can be resolved by simply updating the kernel...

    Yeah. Not a big deal, but leaves a bit taste. Yet another FUD compain vs yet another nVidia's competitor, sigh?
Display more comments