Forum > General discussions > Rumours > SONY XPERIA Rumors 2014

On 2013-11-10 23:52:27, karim128 wrote:
I'm sorry but what you wrote is just pure gibberish.

Qualcomm make his own custom cores whereas A15 is designed by ARM, so how can the S800 be an A15? It's like saying an Radeon R9-290X is not even a GeForce 780. And how can it be half a gen behind when it is as powerful but uses much, much less energy? Why are Tegra 4, based on A15, used in much less devices than S800 and S600?


If you have seen how Tegra market share been divided by 4 this year, you wouldn't have said that... Perception and marketshare doesn't last and doesn't explain success.


They have been the very first with 28nm on mobile, ever since the S4 which was in Sony Xperia T.


You do know that Samsung manufactures Apple's SoC's, right? How can they be catching up to Apple with a process node? They are catching up with Intel. Besides, going to smaller nodes is hardly an idea, it's a vague as saying let's make better chips.


The fact that he's been resigned prove that he was the only one at Qualcomm thinking that.


They're on par with Apple and better than everyone else. Exynos 5 only compares with S600, but is way behind S800.

On 2013-11-11 05:24:31, vivftp wrote:
Hmm, interesting that they're gonig for a smaller overall appearance on the speakers on the bottom of that unit. Plus what's that on the top of the unit along with the headphone hole? Possible IR blaster?

On 2013-11-10 10:11:10, my ninja wrote:
Qualcomm has NOTHING ready (64bit wise) for 1H14 sorry. the S800 is an between the A9 and A15 its not even an A15, so like i said its a half gen behind. let alone A57/53 that samsung is jumping to (more precisely ARMv8 instruction set) and Q has nothing but market share holding them up.. which in tech is not enough.. plus their node is several generations behind 28nm. in terms of CPUs the s800 is widespread but archaic in terms of currently available tech exynos5 (5420 specifically). they have also been on record denouncing 64bit (the man making those statements has since been reassigned) however the way they announced their chipsets at the end of 12, they havent for 14 in the same way. i dont expect them to have a 14nm node ready the way samsung is skipping a whole node to catch up to apples ideas.. exynos has always been ahead of snapdragon, same with omap(rip).. if it wasnt for the perception from the first 1ghz soc, theyd be as popular as nvidia.. as for scale they have everyone beat and can supply the entire market which is impressive, but that can change quickly, their true anchor is their radio integration less the performance. their GPU choice is also a step behind.

android needs more ram than iOS because of the middle layer of the ecosystem (java/davlik runtime) it creates a HUGE performance hit, where if google JUST used their own release of linux that would alleviate a lot of the 'lag' and short comings of android.

4.4 introduced ART to replace Davlik, so that should be impressive as its the first implementation so in a few iterations it should become the default and have substantial gains, right now they are about the same.. but they still need to get out of the java business and just run a branch of linux, if they did that, android would explode performance wise.. but as of now, android is beholden to its roots.. Id love to see Android 6 like its own version of ubuntu, redhat, or centos.. something without that nagging layer hogging resources.

android needs to update from kernel 3.4.. ffs upstream is at 3.10.. its time.

On 2013-11-11 11:36:32, Xajel wrote:

I didn't say that Q has anything, I said I didn't hear anything 64bit'sh for 1H14, and as you said, it's only a better yields of current products + newer & faster GPU... I know that... but what I said also that I know nothing of any 2H14 plans for Q... so maybe they have something for 2H14 or maybe nothing and we will have to wait till 2015 for anything 64bit...

I know that Qualcomm have their own design of ARM compatible architecture, as you said, based on ARMv7... it's just custom... deeper than what Apple is doing with their A6+A7 SoC... which are based on ARM9/ARM57 designs with some Apple modifications and customisations to increase the ARM9 performance ( as they didn't go to ARM15 )...

While Krait is a not as powerful as A15, it came much earlier than A15, the performance is not that much different between the two, Qualcomm get it in the right time as A15 designs took longer time to mature specially that pure A15 designs are power hungry that's why ARM went for the LITTLE.big idea which combines A15 + A7 cores and switch between them depending on the work load, Krait doesn't need such logic as it has superior power management and can handle different workload... Samsung went for LITTLE.big route but it failed in the first attempt ( that octa in S4 ) as it can't run all the cores in the same time ( so it's not a real octa ) and the driver is faulty to choose the cores depending on the workload, they have a lot of improvements with newer firmwares but still faulty... thats why they designed another SoC which is real octa ( can use all cores in the same time ) and the driver is better triggering the right cores depending on the workload... Tegra4 is nothing also ( much lower design wins ) and it also have that fifth combined core which turns on to save power ( as the other four cores are power hungry A15 )...

Krait now running the mastered the design stage ( reducing leak, optimising the process ), that's why they're going higher with the clock while maintaining the target TDP... A15 is still far from being at this stage ( needs more time )...

so Krait is lower in performance than A15 not because A15 is only superior, but because it came earlier so the target performance + power is lower...

Samsung Exynos is nothing special, is just pure A15+A7 design based on big.LITTLE design... there's no customization, and when Samsung made the first SoC it's wasn't complete actually that's why the first SoC was a crap and the crappiest thing they do is that they used it in an actual flagship product ( S4 i9500 )... they was just rushing everything to say ( We have the first Octa core SoC and product )... and it wasn't octa at all !!

That Chinese newcomer MediaTek made the first real Octacore, yes it was based on A9, and older arch. but it was really Octa, real 8 cores running in the same time !! I know performance might not be that much but at least they're not lying and rushing they made it for cheap phones that's why it's not complicated and uses cheaper and older design ( ARM A9 )

Qualcomm is already working on 64bit SoC, but not Krait at all... it's a new arch but we just don't know when it will be ready, 2H14 or 2015... we just don't know... what we know that it's sure not in 1H14...

Qualcomm designed their own chips based around the ARMv7 architecture. The Krait CPU from Q is based on an EARLY version of the A15 before it was finalized. It is a 28nm processor.

These companies license the architecture and design their own individual SoC AROUND their custom ARM CPUs.

Intel fabrication is at 14nm (2 fabrication nodes below the S800 @ 28nm, 20nm being the middle step between the two).

Your analogy between ATI and NVIDIA for graphics cards is inapplicable here for this discussion as its irrelevant to the points that were being made. false dichotomy.

Power management is up to the individual firm, my analogy would be that not all chocolate ice cream tastes the same, they are all chocolate but the final product can vary from manufacturer.

Your complaint about the market share is also down to numerous factors but mindshare is a hard thing to break. An example of mindshare would be 'Retina Display' which is not related to any technical metric, however its often associated with the minimum pixel density where the pixels can no longer be seen. People are stuck behind this artificial wall because of perception, it doesnt however make it the best display. Just the ones that people want because of mindshare. Likewise for SoCs... the S800 is in the Snapdragon family, and because they the 'Snapdragon' broke the GHz barrier for smartphones they continue to own mindshares, it isnt to say its a bad SoC its just an old design people dont seem to grasp this. There was a time not too long ago you could only find OMAP chips in smart devices, to the point where TI owned the market in a similar way, where are OMAP chips today?

Qualcomm was not the first to 28nm for the mobile market, thats just blatantly false. TI had OMAP at 28nm in 2010. Qualcomm had test chips out in mid 2011 I believe..

Intel is the first to 14nm, TSMC is at 20nm and may struggle to keep up, the reports say that there are some yield issues, which they will overcome. Samsung is rumored to have skipped the 20nm process in favor for the FinFet 14nm process.

Going to another node, "is not an idea" im confused by your rebuttal.. its a billion dollar decision, and in order to succesfully produce tens of millions of cpus it takes a lot of brilliant people.. It is a competitive decision, but technology needs to move forward, i dont see a reason to oppose it.

Chandrasekher has not resigned, unless my news is old, he was reassigned. Still reports to the same ppl.

Like you said, the Samsung family of SoCs are on par with apple because they make them for apple. Your assumption about the whole Exynos 5 family is misplaced. the 5420 is a monstrous CPU that technically performance wise (it would burn you like the sun) is an order higher than the S800. Remember that the S800 Krait Cores are based pre-A15 architecture (ARM had not finalized the designing, Q pushed forward). The 5420 is four A15 and four A7's (updated version of the A8), because Samsungs manufacturing process matured over 2013 they corrected the issues from the 5410, and allowed for CCI/HMP. Simply put it sits atop the S800 but remains shackled and there are several limiting factors, namely the software and Samsungs unwillingness to release kernel information regarding the processor itself. Power management from Q however is second to none, that is ALL them too, as no one else in the industry has achieved the battery life or the charging speed of the S800.

my point was Q has not released their plan yet

The future is bright!

On 2013-11-11 11:36:32, Xajel wrote:

On 2013-11-11 13:15:29, karim128 wrote:

ARMv7 is an instruction set, not an architecture (huge difference, for instance AMD and Intel have the same instruction set but their architectures are completey different.) And Krait have been designed by Qualcomm around the same instruction set, and is completely unrelated from A15. At least you recognized that you made a mistake when you said that their node is several generation behind 28nm.


Each company have the choice to either licence the architecture (like nVidia and Samsung who are using A15) or to licence only the instruction set and design their own, custom core compatible with ARM's own CPU out there. Qualcomm and Apple licenced the instruction set and designed their processor completely independantly from ARM.


Intel haven't released yet any processor based on 14nm.


Then you didn't get my point. You said Krait isn't even an A15, while Krait and A15 are completely unrelated architecture-wise and have been designed by two different companies. So that makes as much sense as saying a Radeon isn't EVEN a GeForce. We all know Radeon and GeForce are completely different, but saying a Radeon is not EVEN a GeForce, as if they were supposed to be, is wrong.


Again, Krait and A15 are two different cores which only share the same instruction set, in other words, they can run the same programs but their architecture are completely different.


My point exactly, you said Qualcomm have only mindshare and market share in it's favor. If that is what explains their succcess, then how come Tegra (which were the first dualcore, and then the first quadcore, present in the Surface and Nexus 7) have seen it's sales slow down this much? THey had the mind share and the market share but they aren't succesful.


The only chip from TI that was supposed to be 28nm was the OMAP5, which was never produced. The latest OMAP is based on 40nm.


All we have seen are roadmaps, we haven't seen a single, commercial product based on 14nm, only 22nm from Intel for now.


What i meant is that it is too vague to be considered an idea. It's obvious going to smaller nodes is always better.


I made mistake, I actually meant that he was reassigned. And that proves that everybody at Qualcomm disagrees with him about 64bits.


Actually, there have been a misunderstanding. Only S800 is on par with Apple. Once again, Krait is not pre-A15. How can it be pre-A15 when it is as fast but more power efficient? In other words, how can a pre-A15 be better than a A15?
How come S800 have far, far more design wins than Tegra 4? Have you looked at benchmarks? Only the actively cooled Shield is competitive, but that is rellevant as you should compare it with actively cooled S800.


S800 updated with (among other things) Adreno 400 with DirectX 11 and H265 decode. But yeah it's a leak (although confirmed) and not announced.


At least we agreed on something.