Why Samsung don't go full Exynos?

gongo

Regular
Their new S7 will use 820...but why? Their Exynos S6 was a superb success in light of 810 misfortune. This 810 dark period for Android also shows up the reliance on one third party SOC. All Android flagship using 810 were more or less.... junk.

If Samsung want to be-at Apple, shouldn't they follow one of Apple success tactics, using custom SOC for maximum software efficiency and performance? The time spent to customised their TouchWiz for 820 could be better used to create a better GPU for their next Exynos...

Is it something unspoken in the electronics industry about having to pat each other backs, like Samsung using third party panels in some of their TV?
 
I think it was because of the modem. But nowadays exynos already have internal modem for worldwide. Dunno since what series
 
Maybe their soc (desing, production validation, whatever) cycle doesn't align with the yearly phone cycle, and they choose what's best for current model.

If i'm not mistaken they even tried both socs in past galaxy models, depending on the market, right??
 
Yes, it depended if you had the European model or the US model.
 
Some people think that SoC choice is just about CPU and GPU, but there are a lot of stuff in a SoC. I'm sure Samsung Exynos is very good, but just look at this https://www.qualcomm.com/products/snapdragon/processors/820
A lot of stuff is integrated within the 820 SoC. Also I assure you that software optimization and hardware design is being done by separate team thus whatever Samsung doing with their UI
will not affect their chip design. Also you might fail to notice that although their high end phone is using Exynos, their lower end phone is using whatever cheap SoC that is available. I believe right now they use Mediatek for low to mid level phones.
Anyway, it shouldn't concerned most consumer. It would likely to be the overall best SoC for Android when S7 lauched and if I'm buying S7, I want it to have the best hardware. If Samsung can make a better Exynos in time for S7 launch, sure, use Exynos.
In the end they choose 820 simply because it is probably the best option. Nothing more, nothing less. Also if I'm not mistaken 820 will be made using Samsung foundries, thus they might get some financial benefit from that.
 
But is 820 really the best option?

What if Samsung had taken the cue from 810 failure and fully develop Exynos then and going foward, they will have something that competes against Apple on the performance level...
Samsung is big company, they can stomach the R&D.
They can sell Exynos as Snapdragon competitor.
They can use cheap Mediatek SOC for their lower end models.

But nope, we are stuck with Qualcomm....and another 810 failure, we would lose out on a cycle....while Apple/iOS keeps chipping forward..
 
I think we need to wait until the product is actually out on whether S7 will use 820 only or 820 and 8890. If it's the later, then there should be a reason for it which again we probably won't know until the product is out.
For me personally I will favor the one that has the lower power consumption over raw performance since these high end SoCs have more performance than what I actually need. Of course speaking of performance, I would also favor those that can sustain their good enough performance longer (preferably indefinitely). I don't really care which SoC can achieve that.
 
The 810 had some CPU efficiency issues relative to other SoCs and some reports of overheating in certain implementations, but I think the issues were overblown by sites that seem to emphasize the same drag race like benchmarks in analyzing phone performance as they did with desktop PCs. I think if you measure the system wide power efficiency on day to day tasks like taking a photo or web browsing which emphasize DSP, I/O, memory, and modem efficiency as well, the 810 was probably at least on par with other SoCs. I don't consider the 810 a failure, there are indeed 810 and 805 based devices I would buy like last year's Nexus models. It also looks like the 820 fixes the outstanding issues w/ the 810, and it's very premature to call it a failure relative to the Exynos, though I'm still curious about the L3 situation...

Qualcomm products are much more decathletes than their competitors who might shine brighter in fewer specific disciplines. The 810 is still impressive in its sheer level of integration of functions on a single die, I don't think it's seen anywhere else in the industry yet; the AX and Exynos SoCs still use separate modem and sat-nav modules, although Exynos is starting to put some of that on the same package. I also don't think the RF front end is integrated on the same die in either of those two cases either. The design wins the 810 did garner weren't just because people had warm fuzzy feelings for Qualcomm for no reason (they're considered the bad guy in the industry for their licensing practices), they probably saved them money with a cleaner circuit board layout and outperformed other SoCs when it came to I/O and memory.

Also, as far as Android falling behind Apple performance wise, I think a lot more of that has to do w/ the OS and app environment than the hardware. The iOS app model is much more memory efficient than Android's Java VM based stack which they threw together in a hurry to get things working. (Hopefully ART improves this situation in the future.) As such none of the benchmarks done with Android devices against iOS devices are very Apples to Apples (no pun intended). In typical use, these OS' are also different in how they handle multitasking, and the synthetic benchmarks are not compiled w/ the same tool chain in many cases.
 
Last edited:
I tried most flagship 810 phones and they get easily hotter in much shorter times than 801/805 and AX SOC. I am not surprised if the reviews of throttled performance are true and indicative. And with 810 phones choosing smaller mA batteries for slimmer body for god knows why.. it just compounds the poor period for Android last year....

I think 810 is just bad*bad SOC. The whole ecosystem dependent on 1 guy is bad, and 810 is not Qualcomn first misstep, IIRC 600 series were also bad. It is no wonder we hear rumors of LG and Sony researching their own SOC. I am just befuddled why Samsung is not brave enough to venture out on their own wholly .
 
What is the point you are trying to make? Even if the 810 was a relatively poor chip Qualcomm is not the only player and you seem to conveniently forget that there is no guarantee that whatever Samsung makes will actually be better than what Qualcomm has to offer. That and chances are Samsung's chip manufacturing devision will earn more producing chips for the likes of Apple and AMD than for the Samsung mobile phone decision.

If a company would always make the best product we wouldn't have more than one company in each industry.
 
A few points:

Samsung Mobile is not Samsung LSI. Mobile buys SoCs from vendors and LSI is just one of them.

a) LSI has no CDMA license or modem. This by default requires Mobile to use a Qualcomm SoC or external modem. By default this means QC will get silicon in CDMA markets. In the S7 this will probably be the 820 because:
b) Both 8890 and 820 have integrated modems: Using an external modem on top of the 8890 is basically a $25 increase in BOM. Mobile just wants profits so the choice comes down to:
c) Even if the 820 is worse than the 8890 it's financially a smarter choice to just use 820 in CDMA markets. The only way they would avoid a QC SoC again is if the difference would be just too much, as seen in the 810 vs 7420. The S6 was planned to be 810 in CDMA markets but got dropped due to the abysmal power efficiency.

Again, I expect the S7 to be 820 on CDMA carriers/markets and 8890 in everything else. Apparently the ratio is 6:4 for 8890:820.

Also people are overstating the amount of integration in Snapdragons. They're just the best at marketing it, but Exynos and other SoCs have just as much integration.

I think if you measure the system wide power efficiency on day to day tasks like taking a photo or web browsing which emphasize DSP, I/O, memory, and modem efficiency as well, the 810 was probably at least on par with other SoCs.
No it wasn't. It was really bad.
 
Last edited:
Pretty much every mobile device in the world incorporates a 3.25% bundled patent license to Qualcomm's QTL branch off of wholesale prices if it interfaces with 3G or 4G mobile networks, regardless of if it uses Qualcomm chips or not. (In China it's off 75% of wholesale.) Most of Qualcomm's profits actually come off of this 25% facet of its total revenues rather than the 75% from the chips it produces, and Qualcomm has resisted efforts to break itself up into a more profitable licensing branch and a less profitable chip selling branch because it claims (probably correctly) that this process of chip development and patent filing is synergistic. Anyway, the parts that companies pick is afaik based on the cost, availability, and performance of the part rather than licensing considerations, though Qualcomm has more ability to enforce its terms if they are an important supplier in your chain so you might want to cut them out for that reason.

I'm not going to defend the 810 to the end since it is a flawed part, but I think the 810 is really only a B- part and not the D- minus part that some would suggest it is. I get the feeling that people's overly negative impression of it vs overly rosy impression Exynos is colored by something like extended runs of SpecINT2k or 3DMark on the chips. I've yet to see any reasonably holistic quantitative measure of whole SoC efficiency performance on the various sites that do this kind of thing that constitute a stress test of all its components, correct me w/ some data if I'm wrong. That said, the physical implementation likely trumps any design choices that went into the chip by a wide margin as well, so it's likely the 810 would lose to any 14nm fin-fet fabbed SoC, but my point is it would not lose by as much on tests that stress other bottlenecks like DSP, memory, and I/O than if the test just leans into one or two corners of the part.

I do think for now Qualcomm has an important edge in integration in packing its modem on die (I believe the 8890 has it on package) and just as importantly, making developers aware of its software stack for its various components. An aspect of this is marketing for sure and they need to do this as they don't make phones themselves, but this makes me think that their parts probably have better drivers and APIs than Exynos or MediaTek. It may not matter for Exynos if they decide to keep it in house just for the S7 and has just one set of developers for it, but I think they did have some teething issues w/ the S6 on the software side.
 
Last edited:
AnTuTu-GPU-performance-scores.jpg


Now we see why Samsung did not go full Exynos....
What just happened at Qualcomm to produce such epic turn of event? It is not like Samsung went slack with the Mali GPU, they went with the full configuration! The Adreno 530, what is inside it? Did Qualcomm trolled their buyers with the crappy 430 last year?

Is such a generation gap even possible? Are we in a dream?
So many questions...
 
Probably off topic but there's something odd with that Antutu graph...

I've not looked in detail at the Galaxy S7 / Qualcomm 820 so may be missing something obvious but the S7 looks to score only a small (10-20%) higher than the A9 in the iPhone 6S/6S+ on the other common GPU benchmarks (GFXBench TRex, Manhattan, 3DMarkNext-IceStorm, etc) but the Antutu score is much higher in favour of the 820. Not familiar with Antutu's latest benchmark but there must be something going on that makes it significantly different from the other standard graphics tests.

edit:

I downloaded Antutu and had a play. 6S data is from my phone, Galaxy S7 data from here:
http://www.slashgear.com/antutu-ben...-version-has-better-gpu-performance-16427082/

iPhone 6S (A9): 3D = 39726 (Garden = 17748, Marooned = 21978)
Galaxy S7 (820): 3D = 52410 (Garden = 22816, Marooned = 29594)

So, it does looks like a straight 30-35% speedup across both tests Very impressive! Wonder what's so different about the workload in this bench...
 
Last edited:
Absolutely irrelevant. They did it because of CDMA. ...

While I agree that Samsung skipped out on the 810 because they just had a better chip in the Exynos mainly due to their 14nm process, I don't think they're picking the 820 this year due to licensing cost considerations. Bundled licensing fees for essential wireless patents are paid on a per device percentage of wholesale value basis rather than per relevant chip included. This is why Qualcomm made such a nice profit last year despite the lukewarm reception to the 810 since it was raking it in from the sale of all those iPhones and other 3G/4G devices regardless of if they had a chip in them.

So why source two types of chips at all? I think the choice of chipset comes down to foundry capacity.

Previously, ordering half its parts from Qualcomm was probably more a method to ensure a reliable and redudant supply of chips as Snapdragons were fabbed with TSMC; now it's probably more because Samsung is trying to better utilize to capacity its cutting edge foundries that they spend so much to upkeep each year by keeping an external partner around with an S7 supply contract and a corresponding fab contract. The Samsung / Qualcomm foundry deal extends beyond the S7: it certainly helps Samsung's bottom line that Qualcomm is buying up so much of their foundry capacity for the 820 demands of other customers, not just in the S7 and that this year, it's not just Samsung themselves masturbatorily fulfilling their own orders. An integrated modem on die rather than on package means Qualcomm is making better use of those transistors than Samsung is and it passes some of those savings on a package level to Samsung's phone division for its S7 orders. Furthermore, even if all chips do come from Samsung foundries and there isn't physical redundancy anymore, having two companies plunk down cash to buy foundry capacity for APUs does ease the initial capital commitment by Samsung itself and ensures the supply for Samsung's phone unit in a very crucial way.

Profit and supply arrangements aside, Qualcomm does have a well-balanced winner this year for the standard specs of 2016 flagship phones in the 820, and it more than clears the hurdle for inclusion in the Galaxy. There's probably all sorts of other advantages over the Exynos like a better driver and compiler stack, better integration of parts resulting in a simpler PCB or a cheaper package, more efficiency than the Eyxnos for GPU bottlenecked things like VR etc. I'd be surprised if the Exynos is able to run those VR demos as well as the Snapdragon. I think Samsung might be leaning into the Snapdragon models for their GPU intensive marketing demos and hoping the Exynos users don't notice too much...

... Also the GPU power at full performance for the 820 is something fit for a tablet, not a phone.

An APU's GPU performance is much more related to screen resolution rather than screen size so it's perfectly applicable to the S7's smaller screen running apps at native res and few android tablets have higher res than flagship phones. If anything, Samsung is overdoing it w/ the 1440p display on a phone, but it needs to keep up in the war of specs vs its rivals and Qualcomm is just keeping pace w/ customer demands by integrating GPU that might finally be suitable for more demanding apps with a native res of 1440p. I blame the LG G3 for starting this trend; using the 801 to power a 1440p screen was a ridiculous overreach. The only reason it does make sense is for VR, but maybe the Adreno 530 isn't powerful enough for this to make sense yet.
 
Last edited:
Sometimes I get the impression that people think Samsung's Galaxy S line is the majority of what they make.. it gets most of the press but they make so many other mobile Android devices (and even Tizen devices). Probably over 50 a year. Putting Exynos in each and every one of them would be a very tall order just by volume, and a very risky move. But if you look at device specs over the last few years it does look like Samsung Mobile has gradually been selecting Exynos SoCs more and more frequently. Part of this is to fill the void left by the likes of TI, Sony-Ericcson, and nVidia no longer being useful SoC suppliers.
 
While I agree that Samsung skipped out on the 810 because they just had a better chip in the Exynos mainly due to their 14nm process, I don't think they're picking the 820 this year due to licensing cost considerations.
I didn't say anything about licensing. The 8890 has an integrated modem. For the 7420 with an external modem they could have just replaced it with a Qualcomm modem for CDMA carriers, which is exactly what they did. Making use of CDMA with the 8890 means you have to add an external modem, not just replace an existing one. The added BOM is astronomical ($20-25+).

Bundled licensing fees for essential wireless patents are paid on a per device percentage of wholesale value basis rather than per relevant chip included. This is why Qualcomm made such a nice profit last year despite the lukewarm reception to the 810 since it was raking it in from the sale of all those iPhones and other 3G/4G devices regardless of if they had a chip in them.

So why source two types of chips at all? I think the choice of chipset comes down to foundry capacity.

Previously, ordering half its parts from Qualcomm was probably more a method to ensure a reliable and redudant supply of chips as Snapdragons were fabbed with TSMC; now it's probably more because Samsung is trying to better utilize to capacity its cutting edge foundries that they spend so much to upkeep each year by keeping an external partner around with an S7 supply contract and a corresponding fab contract. The Samsung / Qualcomm foundry deal extends beyond the S7: it certainly helps Samsung's bottom line that Qualcomm is buying up so much of their foundry capacity for the 820 demands of other customers, not just in the S7 and that this year, it's not just Samsung themselves masturbatorily fulfilling their own orders. An integrated modem on die rather than on package means Qualcomm is making better use of those transistors than Samsung is and it passes some of those savings on a package level to Samsung's phone division for its S7 orders. Furthermore, even if all chips do come from Samsung foundries and there isn't physical redundancy anymore, having two companies plunk down cash to buy foundry capacity for APUs does ease the initial capital commitment by Samsung itself and ensures the supply for Samsung's phone unit in a very crucial way.
There are no deals with Qualcomm. The mobile division doesn't care what the foundry is doing. It will use whatever supplier provides the best performance per dollar.

Profit and supply arrangements aside, Qualcomm does have a well-balanced winner this year for the standard specs of 2016 flagship phones in the 820, and it more than clears the hurdle for inclusion in the Galaxy. There's probably all sorts of other advantages over the Exynos like a better driver and compiler stack, better integration of parts resulting in a simpler PCB or a cheaper package, more efficiency than the Eyxnos for GPU bottlenecked things like VR etc.
Utter nonsense. Qualcomm's sole advantage lies in their modem and RF performance, everything else is just wild theories. The 8890 should actually be far cheaper than the 820 due to it have smaller die size (you heard it here first).
I'd be surprised if the Exynos is able to run those VR demos as well as the Snapdragon. I think Samsung might be leaning into the Snapdragon models for their GPU intensive marketing demos and hoping the Exynos users don't notice too much...
Again, load of nonsense. At MWC all the devices and VR demos were running on Exynos.

An APU's GPU performance is much more related to screen resolution rather than screen size so it's perfectly applicable to the S7's smaller screen running apps at native res and few android tablets have higher res than flagship phones. If anything, Samsung is overdoing it w/ the 1440p display on a phone, but it needs to keep up in the war of specs vs its rivals and Qualcomm is just keeping pace w/ customer demands by integrating GPU that might finally be suitable for more demanding apps with a native res of 1440p. I blame the LG G3 for starting this trend; using the 801 to power a 1440p screen was a ridiculous overreach. The only reason it does make sense is for VR, but maybe the Adreno 530 isn't powerful enough for this to make sense yet.
You completely missed the point. GPU performance is directly related to a device's size because it is limited by thermal dissipation. The Adreno 530 is oversized for smartphones and it will throttle substantially, any performance advantage it has is irrelevant if it can't sustain it more than 5 minutes.
 
Your point about the modem sounds reasonable, and it would indeed count as one of the Snapdragon's advantages for this gen. I'm speculating about what Samsung did behind closed doors; that Qualcomm won a spot in the S7 around when it switched over from TSMC to Samsung is no small thing. As for lower cost based on Exynos' die size, I'd take the modem and packaging process for the modem into account too (assuming rumors of that package level integration are true).

I look forward to more detailed analysis of this situation w/ the Snapdragon vs. Exynos, but XDA's initial numbers indicate about a 20% throttling after a few tests w/ the S7, not enough to make up the difference between the Snapdragon and the Exynos in what we've seen in AnTuTu so far even w/o throttling by the Exynos:

http://www.xda-developers.com/s7-ed...rk-stress-tests-of-sd820-810-808-exynos-7420/

Thus, I'm interpolating that the 820 will do better on 3D workloads like VR than the Exynos.

The point of having these gargantuan phones instead of the hyper efficient but anemic sliders of yore with a week of battery life is that they're able to clear certain qualitative hurdles of performance like sufficient FPS for VR, a day's worth of battery life, a stable software stack for general operations, fluid UI, fast and high quality photos etc. I don't believe a pure efficiency metric as a quotient is enough to convince me choose a phone w/ a Mediatek chipset over a Snapdragon or overstate the Exynos' advantage based on some edge cases. Despite the 810's heat issues under certain testing conditions, I'd happily buy a Nexus 6P and give the 810 credit where it deserves it (e.g. its GPU performance relative to its generation). My conclusions about the 820 are just from what I've seen so far, I'll change my mind if there's damning evidence of course.
 
Last edited:
Back
Top