Read more.Firm says users will experience better gaming, quicker and smoother multi-tasking.
Read more.Firm says users will experience better gaming, quicker and smoother multi-tasking.
8 cores good, in theory... but it's marketing BS - the ARM A7 (not the confusingly named Apple A7) is slow per core compared to the best A9 or better cores.
Far better to have a quad core of something faster per core, actually load those single thread apps at better than snail pace. Hell even the dual core SoCs based on Apple A7, Atom or high clocked Krait would be a better user experience than this.
Hmm. 8 cores @2Ghz. Would be pretty quick IF IPC is good. Otherwise a waste, still waiting on software to catch up to hardware.
The only problem with MediaTek chips is that MediaTek refuse to release their frameworks, leading anyone who wishes to create custom ROMs completely in the dark and unable to do anything.
Slow compared to what though? My stepson has a phone with a dual A5 core @ 1GHz and it's plenty fast enough for all the tasks he does including mobile gaming. A7 cores are faster than A5 cores, and in this SoC they're clocked at up to twice the clock speed. So the processor isn't slow by any reasonable real-world definition. Now, whether the software makes good use of it or introduces lag is another matter - my LG O2X is fairly laggy sometimes, but that's not the Tegra 2 SoC's fault...
um... I thought the samsung octa (or more precisely the arm big little architecture which it's based on) was a true eight core.... well after it has/had a software update anyways....
Still don't see the point in 8 cores, hell I'm not even convinced 4 cores is 'needed' on mobile's. It's all marketing these days, same as 64bit on mobile phones or even 'consumption orientated' slates like the iPad... I can see a point for a server though....
Could be wrong but i think Samsungs effort was 2 Quad Core chips on the same die, meaning it could switch between the two depending on load and usage not really a octa core. Whole thing seems a tad pointless tbh given that we can't even get desktop CPUs to use 8 cores correctly in most tasks.
Non-symmetric SMP is a minefield where performance is held back by the slowest core, so no I don't think you can call a 4+4 core setup like in the Samsung part an octa core.
As for 64 bit ARM, consider it a needed facelift of the whole instruction set. The fact that 64 bit processing falls out is a bonus.
To an A9, A15 or Krait - see my other post, I suspect user experience of a high clocked quad or dual core with better cores would be superior even if total performance was technically lower. Whilst an A5 or A7 might well be "good enough" or "reasonably good" it doesn't mean an 8x A7 is really a sensible option. ARM themselves estimate that A9 has ~30% more performance per MHz than A7, Krait is something like ~70% more... that really counts in the single thread mobile workloads.
An 8x A7 just seem like a bizarre design choice to me, A7 is aimed at the lower end, is 8x SMP really needed there... it's exactly like PCs where Intel have kept the mainstream market on faster clocked quad cores.
Does he have a desire x ? Not many phones ever used the a5 core. Not to be confused with the apple a5 of course as that is based on an a9 core.
I would suspect 8x a7's are slower than 4x a15's in all but extreme multi threading. As when the a15 first came out a dual a15 was more than a match for a quad a9 clock for clock
The A7 is by no means a slow core. As an example, the new Moto G which is getting positive reviews everywhere, and praise for its excellent performance, uses a quad A7 processor. It's also worth remembering how *tiny* the A7 cores are; nearly 3 would fit in the same area as an A15, nearly 13 of them in a Clovertrail Atom.
As for whether a 8 core A7 is well suited to a phone is debatable ATM though; it may not offer much more than a quad A7 in a good deal of scenarios. But yeah, there's sure to be a fair amount of marketing buzz behind an 'true 8 core CPU'.
On a phone? Pointless, what are u going to do with them, play BF4 or use solidworks? Browsing facebook and playing angry birds at the same time is more likely lol Why don't they put their efforts into something useful like power consumption?
it is 2 sets of 4 (1 set is low power a7, 1 set is high performance A15), hence the software comment, supposedly it can be made to work 'together' with a software update...
You could argue the benchmarks show how slow the a7 is versus the a15, 8 a7's only just beat 4 a15's.
The thing is with hardly any mobile apps using 4 cores you could (assuming said software patch worked) use it in the exact same way the mediatek would likely work, it would just distribute apps to their own core(s) to improve multitasking, well in theory, and at equal clock they'd perform no worse than the mediatek as the 'slower cores' in the octa are a7 (albeit slower) like the mediatek lol.
I don't see 64bit (apple has just used arms 64bit design/code earlier than anyone else) as a needed facelift, Qualcomm haven't gone that way to 'improve the instruction set' and you can see how much can be done without the need for 64bit with their chips. I will say Arm does need to optimise some aspects of their code, maybe removing certain legacy elements would help...
As I said further up, A15 is over three times the size of A7 so of course it's faster. However it also uses more power and is less efficient.
As for 64 bit, as danceswithunix said earlier, there's more to ARMv8 than 64 bit!!
What exactly would you have ARM optimise or drop; ARMv8 makes a fair amount of changes to the ISA. But don't confuse the ISA with IP cores - Apple/Qualcomm/etc design custom cores but still work from ARM ISAs. A12 is also out soon, a mid-range successor to A9, along with the ARMv8 A53 and A57 cores of course.
Last edited by watercooled; 21-11-2013 at 04:27 PM. Reason: Clarity + spellings.
All old instruction sets have their quaint features that were a good idea when they were new but hold them back now.
MIPS has the branch delay slot, which only makes sense if you have a single issue simple pipe.
Sparc has those wacky register windows that add complication but still leave you with management problems.
ARM has the expectation that leaving a barrel shifter in the main pipe will speed things up, and predication is always a good thing.
x86 has a lack of registers, variable instruction length, bottleneck on a single condition code register and all that segmentation baggage.
So yes, it did need a face lift, in the same way that the conversion from x86 to AMD64 more than made up for the penalty of caching 64 bit values.
Or put it another way, the best ARM cores out atm can issue three instructions at once, and struggle to keep up with a dual issue Atom. Dual issue is the sweet spot for energy consumption due to the average instructions between jumps (about 6) so something had to change before ARM energy use had to shoot through the ceiling to go any faster.
Oh, on the 8 core Samsung thing there was some research on this years ago. Latency is set by the slowest core in the system, so the mind bending result is that you are actually better off disabling the slow cores and just using the fast ones as that makes the system feel smoother. If sounds wrong, because less cores means less work can get done, but the slowest core will end up running something you don't want it to and that is what the user will feel. I'm sure that can be fixed in the schedular with enough hints from an application, but as I said before it would be a minefield and frankly not worth doing when the 4+4 system they have seems to work (I say seems as I haven't tried it, but my phone uses the Nvidia 4+1 layout and that isn't too bad).
There are currently 1 users browsing this thread. (0 members and 1 guests)