SemiAccurate has been following a massive FUD campaign for a few months and the rabbit hole has led to some interesting places. While chasing down leads things went from normal technical problems rumors to a fairly scared giant in short order, and nothing was as expected.
You might recall a story that has been floating since early December about the then upcoming Qualcomm Snapdragon 810 SoC. The narrative essentially started off along the lines that the 810 has overheating problems and with each passing iteration had more details and occasionally a new angle thrown in. SemiAccurate has been hunting down sources, leads, and hard data on the potential problem for far too long. What did we find at the bottom of this rabbit hole? Not what we expected.
Lets start out with what the story said. Initially it was quite simple, the Snapdragon 810 was said to overheat and to have severe problems. This was broken by a Korean tech blog on December 2, 2014 and quickly spread around the world. Qualcomm flatly denied the rumors and said there was no such overheating problems. As is normal the buzz peaked and quickly died down once people got a chance to test the devices, there were no problems found.
About a month later those rumors came back again, this time it was in the Korea Times on January 7, 2015. That ‘reconfirmation’ set off a fresh round of hype in the vast echo chamber of no-source news outlets. Since this was after LG announced the G Flex 2 phone bearing the Snapdragon 810 at CES, most of the press had hands on time with the actual device. This didn’t stop many people from ramping up the breathless hype once again though, and the supposed problems still were not found to be true.
The next time the 810 overheating came up was on January 20, 2015 in a Bloomberg story. There was a new twist here though, the supposed overheating that no one could verify was ostensibly the reason Samsung dropped the 810 for the upcoming Galaxy S6 phone. For once an 810 rumor seemed to be panning out, there were strong signs that Samsung was not going to use a Qualcomm chip like they did with the S5.
This was followed two days later by a Wall Street Journal article which stated that Qualcomm would build a special version of the 810 for Samsung to win them back. Supposedly this was going to cause problems for the rest of the Qualcomm partners using the ‘vanilla’ 810. There was again no truth to this rumor and it was trivial verify.
There were also a bunch of completely off the wall versions of the rumors, and yes SemiAccurate tried to verify each one. One financial report that is not public claimed the 810 had memory controller problems, another claimed LG was suing Qualcomm for some mystery slight or other. All of these and a few more were, well, far more baseless than the overheating rumors, they were not based on anything remotely real that we could find.
All of these rumors also had another common thread, all but Samsung dropping the 810 were quickly denied by Qualcomm. All of the ‘overheating problems’ were found not to be true by many testers on dozens of production devices both at trade shows like CES and in the wild. You can buy an LG G Flex 2 now and it doesn’t overheat, it wasn’t underclocked as rumored, and has none of the ‘reported problems’.
Only one of the rumors was actually true as it turns out, Samsung did switch from the 810 to a 7000-series Exynos. This was admitted to by Qualcomm in their quarterly earnings call in January although Samsung or the SoC was not mentioned by name. While this loss is still officially a mystery, no one doubts the two are connected.
The pattern was clear, there was a rumor that was quickly denied by Qualcomm, then the press quickly and independently verified Qualcomm’s side of the story. Most rumor mills didn’t care that the story was completely incorrect though, they had a juicy headline and ran with it. Where have we heard this before? On the Internet, truth is of no concern, verification is a four letter word, and hits are all that count to the desperate sites out there. Very few have real sources or ethics anymore, and it shows but a few just don’t care about truth or ethics.
What is really interesting is that the main rumor kept coming back and coming back but each time it resurfaced there was a new twist to it. Things went, rumor, refutation, rumor+, refutation+, rumor++, etc, etc, with only the supposedly overheating 810 as the common thread. It almost seemed like someone was trying to trash the 810’s reputation and knew how to manipulate the press to keep the echoes from fading into obscurity. This was either a set of desperate web sites in a slow news cycle or a really sophisticated FUD campaign.
A real basis for the rumors:
Things could have gone either way so when SemiAccurate hears rumors about chip problems like this, we dig into them and find out what is really going on. This time was no exception and when the first of the rumors surfaced in early December, 2014, we dived in as usual. What we found was pretty much par for the course in mobile SoC development, there was a grain of truth to the original rumor after all but it was inapplicable to anything consumer facing.
Understanding what that grain is gets a bit technical, something most average news outlets would not understand. Those with even a small hint of technical knowledge about the chip development process would have heard the explanation and shrugged the rumors off, the mainstream press where each round of rumors started may not have been so astute. That said they seem to all have abdicated their responsibility as journalists by not verifying the ‘rumors’, either that or they had an axe to grind. We aren’t sure which way the mainstream press leans but the supposedly ‘technical’ web sites that made up the echo chamber have no excuse this time.
The grain of truth SemiAccurate found early on was the first silicon dev boards actually ran a bit hot, not overheating but hotter than final silicon was meant to run. To a layman this many sound bad but in the grand scheme of silicon development, running a little hot is not a big deal if you know how the process works.
Chips usually go through several cycles of bug fixes called ‘spins’, each one taking between 8-12 weeks or possibly more. This is because a wafer on a modern process takes a minimum of 6-8 weeks to manufacturer, there are a lot of steps to it. Pre-silicon simulation of chips is slow and you can only find so many bugs before you have to sign off on production. Once you have functional silicon back in hand, it runs 100-1000x faster than simulation and you can test your firmware, OS, and code at reasonable speeds.
It is quite normal to find bugs after you get first silicon back, the number of complex chip designs that don’t need a few silicon change level bug fixes is likely in the low single digits, the author can only think of two that did not need bug fixes in over a decade of covering the subject. This is how the process works, silicon comes back and the added testing speed uncovers a bunch of things which are then quickly fixed in the first spin.
One of the most complex parts of a modern CPU and especially a mobile SoC is the power management hardware, firmware, and software. While it is beyond the scope of the article, don’t underestimate the magnitude of the task, there are tens of thousands of parts each of which need to coördinated on a millisecond time frame for it to all work. You can’t simulate this well pre-silicon on a chip with billions of transistors, you just have to give it your best shot. By comparison almost everything else in the process is quite straightforward in comparison. See here, here, and here for a bit on the complexities of power management.
Because of this most modern SoCs work pretty well as far a logic and correctness of operation for the earliest stages, power management usually just comes close to the target. With first silicon to test against, power management code progresses very quickly, think of it more as tuning than bug fixing. Since of this functionality is at such a low-level it is actually in hardware or needs minor hardware changes to make the tuning function as intended.
As you can imagine, having non-final power management code can lead to using too much power, overheating, or releasing the magic smoke that powers modern silicon devices. If you look at early silicon development boards for mobile phone SoCs like the 14nm Samsung Exynos that will be in the Galaxy S6, the heatsink is pretty massive, way too big for a mobile device.
Early silicon for the Galaxy S6 phone SoC
Samsung engineers SemiAccurate spoke to at ARM Techcon had no problem with the heat the 14nm part produced at this early stage of development nor did SemiAccurate think anything was amiss. No one familiar with the process would have intoned that it would not be ready for a phone in a few months because of this ‘problem’. Also note that the Snapdragon 810 was much farther along in the development process at the time Samsung was showing this Exynos off.
That is the grain of truth SemiAccurate found behind these Snapdragon 810 overheating rumors, first silicon did run hot but we couldn’t really find anyone who would go as far as saying it overheated. It worked as intended though and would fit in tablet form factor devices, if SemiAccurate saw them and played with them many months before these rumors started, we could say definitively that they didn’t overheat. Then again we can’t admit we saw anything in this time frame, we just saw 805 based devices, really.
Everything we found was normal for this class of chip at this stage of development, first silicon with pre-alpha software was not perfect as it never is. Things progressed quickly and there have been several spins since the first silicon came out, none of which ran hot, the fixes worked. At this point when things go wrong it ends up with delays to the product being released and/or specs being changed late in the game. You might recall AMD’s Barcelona CPU or Intel’s current Broadwell woes, both of which are horribly late and de-featured.
It is pretty obvious when silicon has serious development problems and the 810 is not one of those, it is unquestionably not delayed and is unquestionably not running hot, you can buy one now off the shelf and see for yourself. The grain of truth to the rumors was simply a normal and quite minor part of the development process that the mainstream press reporting all this seems to not understand and did not try to verify. We have no excuse for the ‘technical’ sites though, they should know better.
At CES we played with several LG G Flex 2’s at pre-show events and in the LG booth. None overheated. All performed as intended and within expected ranges for the CPUs in question, there was no problem. Qualcomm 810 reference devices were likewise problem free so we shrugged off the rumors. Many other sites did very detailed testing and again no one found anything like what was rumored to be lurking under the supposedly oversized heatsink.
Because these rumors kept coming up and coming up, SemiAccurate kept digging into them. We eventually went as far as to test several areas not covered by the basic power testing done by others during earlier rounds of this rumor. Why reinvent the wheel when you can look in other places? With that we looked at the radio, something others did not cover and had supposed power draw and throughput problems.
First off was a test of the modem power, while all the G Flex 2 and 810 reference design tests didn’t show any problems with the main CPU or GPU cores but the modem wasn’t really tested for power draw by anyone. The tests were pretty simple to explain, take an 810 reference platform and push as much data over the LTE connection as possible while measuring the power draw. As a reference we also did the same with a Samsung Galaxy Alpha which uses their CMC303 modem. Both were run at the same throughputs for comparison and both were done in a closed test environment free from external radio signals.
Both devices were set to the same LTE class/speed but SemiAccurate was asked not to reveal exact details. We feel this test was unquestionably representative and not a ‘best case’ for either device, the Qualcomm device could have been set to real world modes which would have made them look much better. This was a flat-out apples to apples comparison, just not using an Apple device because that would not have been an ‘apple’ in this case. The results looked like this.
Qualcomm 810 modem power draw
Samsung CMC303 modem power draw
There are two things you can see from the pictures above, the graph itself and the little average number on the bottom right. On top the Qualcomm 810 runs in a very tight power range, other than a few spikes there was only about a 5mA variance in the draw. The average power draw, spiked included, was 237mA but be aware that the PAs were not on because the transmitter and receiver were right next to each other in an RF sealed chamber.
The CMC303 in the Galaxy Alpha looks a lot messier than the 810 but that is not the case, the spikes in the Qualcomm graph blew out the scale making it look cleaner. If anything the Samsung device has a little tighter variance but we are talking far sub-1% here so a mouse sneezing in the next room during another run of the tests could flip the results. That said the CMC303 drew measurably more average power at 266mA, and yes, both were running at the same voltage.
In short the supposed power draw problems of the modem that some referenced were unquestionably not true. The modem in the 810 drew only 89% of the power that the CMC303 did and by using real-world applicable modes that the 810 supports but the CMC303 does not, we could have dropped the power per bit transferred significantly, really significantly. While SemiAccurate feels this would have been a fair comparison, it wouldn’t have been apples to apples.
For the record, sources we trust have tested an LG G Flex 2 in the same way and found it draws 251mA worst case. This was not technically a released device because it was tested about a few days before the phone went on sale but we are quite sure there won’t be changes to shipping units that will change this number appreciably. While it is a slight bit higher than the 810 reference platform tested, it is still measurably below the CMC303.
Lets just say that the 810 does not have any power draw or overheating problems, it works more efficiently than anything on the market and could do better still if we wanted to test absolute performance. Maybe some other time. That said it is on sale now, go buy a few and a Galaxy Alpha and re-run the tests we did yourself, we are very confident that you will find the same numbers. If LG wants to send us some release units, we would be happy to try a few other things too.
The next rumor that hasn’t been comprehensively debunked was the rumored 810 modem throughput problems when channel bonding. Once again it was an epic 810 vs CMC303 bout in a closed RF chamber but these were different units from the ones used in the modem power tests. You have to instrument them quite differently for the two tests, doing it with one set of devices isn’t practical in the time we had available but as we keep saying, you are welcome to do so yourself.
Please note that this test wasn’t using the full bandwidth available in Cat 4 due to a purposefully degraded signal representing real-wold conditions. The idea here was to show how efficiently the channel bonding worked and how much overhead there was. Intended results should be roughly half of the 150Mbps that Cat 4 supports, the idea is to see how efficiently that the available signal was utilized. In theory any differences would be magnified at ‘full’ bandwidth/speeds but we couldn’t test that with what we had at hand. In any case both devices were tested in exactly the same way, it is again apples to apples.
Qualcomm modem throughput
Samsung modem throughput
As you can see the Samsung CMC303 had a throughput rates of 75/74 Mbps, not bad for an LTE Cat 4 device with the available bandwidth. The Qualcomm 810 did much better at 88/87.5Mbps, more than 10+ percent more throughput than the best competitor. Modem throughput is not a problem for Qualcomm, nor is channel bonding. It is quite clear that the 810 is significantly better in both regards than its closest competitor but unfortunately SemiAccurate can’t say exactly why this is the case, our tests didn’t shed light on that.
That was only Cat4 though, the 810 can do Cat 9 which is where the purported problems might lie. It is not unusual for a device to work well on older specs but have problems on bleeding edge features that are in silicon for the first time with a new device. That could be what some of the rumors are referring to and something that a major OEM’s testing would probably uncover. We decided to test that but since there are no other Cat 9 modems out there to test against, we had nothing to use as a competitor. Because of that all we did was push it as hard as we could and see what numbers came up. This test was again done with an 810 reference device in an RF sealed chamber, it is a ‘best case’ radio throughput test.
Qualcomm 810 Cat 9 performance numbers
The two numbers that matter are “Throughput – Combined” in the top picture and the Mbits/sec column in the lower picture. They are 452 Mbps and ~442Mbps respectively, the former being raw phy throughput and the latter is actual UDP throughput, encoding overhead accounts for the difference. With a transmitter and receiver sitting next to each other in a box and using the full 60Mhz of bandwidth available for three carrier aggregation, the 810 was able to hit the theoretical maximum throughput. No problems there.
That said a radio throughput test like this is obviously the best case scenario, the real world is pretty much assured to be far worse. You have multiple users fighting for the same spectrum, interference, multipath echoes, and all sorts of other little problems that can drag performance way down. Testing a Cat 9 device in the real world is problematic because the first Cat 9 network should be going live as you read this but that is nowhere near SemiAccurate. We couldn’t test it in the real world.
Luckily for your dear readers, the latest Signals Flash from Signals Research Group did test an 810 in the real world but only in Cat 6 modes. They flew to South Korea and got access to a few pre-production LG G Flex 2 phones and the LG U+ LTE-Adanced Cat 6 network. The results pretty much floored SemiAccurate, we really didn’t expect anything real world to be this close to the theoretical max throughput.
While we urge you to read the Signals Flash linked above the short story is they tested over three days in Seoul in taxis and the subway. As the graph at the bottom of P6 shows, they averaged 158Mbps over the tests with a peak of 296.5Mbps over 612.6GB of data transfers, again in mixed real-world conditions. While that number is stunning enough, if you only look at the subway testing on P8, it averaged 213.9Mbps over 113GB of data transfers. How fast is your land line again?
If any of SemiAccurate’s tests didn’t convince you, the Signals Research Group report should. Do note that on P4 it says, “This was an entirely self-funded independent study.” The radio in the Qualcomm Snapdragon 810 has no power draw problems, no theoretical Cat 4 throughput issues, and no theoretical Cat 9 issues. Real world testing of Cat 6/LTE-A devices and networks show there are no issues there, devices came within a hair of the theoretical peak too, what more assurances do you need?
Conspiracy or idiots?
In spite of the complete and utter debunking of the rumors, the Snapdragon 810 ‘overheating’ story kept coming back. Each time there was a new twist, it seemed to be just enough to refresh the story and make headlines once again. This was either chance or something very carefully engineered but at this point it could go either way. So once again SemiAccurate started digging, not into the hardware and reported ‘problems’ but where the noise causing the echoes came from.
The first thing that became clear is that each new echo started out within a few days of important events in the launch of the 810. The first overheating rumor started on December 2, about a week before the first press outing and test for the 810 and a few days after the press was notified of the event. Coincidence? Probably.
Things died down again until CES when LG outed their G Flex 2 with the 810, the first public launch SemiAccurate heard of was on January 5th at a pre-show event. The first article in the second round of echoes happened two days later and spread a bit quicker than round one. Twice is unlikely but still could be coincidence, nothing to point to as a conspiracy yet.
Times three and four have a caveat to them, quiet periods. Because of SEC rules, when a company files something with the body they are restricted from saying things that may influence investors until the filing is made public. That means when a quarterly report is filed, the company has to shut up on most topics that could affect financial folk, so just about every topic is off-limits. This quiet period is about a week or two long depending and is really determined the dates the company picked to submit and release those filings. Product launches, press briefings, and especially analysts briefings are right out during this time.
Back to the main story and echoes of overheating. Qualcomm reported on January 28, 2015 and for a bit over a week before that they were in their quiet period but SemiAccurate could not find the exact dates for it. During this time there were a bunch of Asian press events but nothing new was being revealed as it was the quiet period. Think of this as a product Q&A plus hands on time for the press that didn’t go to CES. These press tours are quite common so nothing exciting there, we are just pointing them out for completeness.
On January 20th, Bloomberg reported that Samsung dropped the 810 from the Galaxy S6 because of overheating problems. We know that this story has a few grains of truth to it, but was it correct? First silicon did run a bit hot but that was part of the normal development process and was fixed many months before the story. We know Samsung did decide not to use the Snapdragon 810 in the S6 but not because of overheating because that is patently not occurring.
Is this rumor true or false? It has two grains of truth but purported problem was not real and does not lead to the end result. Was this coincidence, a case of journalistic idiocy, journalistic malfeasance, or a conspiracy to use the press for ulterior motives? At this point things were starting to move away from coincidence and towards one of the other choices, but which one? Either way, Qualcomm was in its quiet period so it couldn’t have responded if it wanted to. If this wasn’t happenstance, someone was hitting below the belt, far below and hard.
Two days later the Wall Street Journal reported that Qualcomm was going to whip up a special 810 for Samsung to use in the S6, essentially shafting all of their other partners. Anyone with a shred of technical knowledge would know this is farcical, you can’t whip up a new faster version of a part at the volumes Samsung needs for the Galaxy S6 in a few months let alone about a month before the product launches. The volume ramp would have needed to start months before this rumor said it was happening and the engineering would have needed to start quarters before that.
In short it was technically impossible for Qualcomm to whip up a special SoC as reported, the whole idea is simply farcical. Any reporter with half a clue about tech would have laughed this one out the door, I am amazed that the Wall Street Journal printed this drivel, it is quite pathetic. And again it happened during the Qualcomm quiet period so they couldn’t respond. Coincidence wasn’t looking like a really strong possibility at this point.
To cap the whole affair off, on January 28th Qualcomm’s quiet period ended. During their analyst conference they were pestered with questions about overheating, Samsung, Galaxy S6s, LG, and similar things, all with the unquestionably false overheating rumor at their heart. Qualcomm said that they had indeed lost a major customer but because they are honest and play by both the letter and spirit of the law, they didn’t mention who it was. We can say that it was Samsung and the Galaxy S6 but that is not official at this point in time. Coincidence my ass.
Two clusters of data:
With strong hints that there was a deeper conspiracy at play, SemiAccurate kept digging. We found two sets of interesting data points, both had multiple sources saying either exactly the same thing or large chunks of the same story with a few differing bits. One was centered in the mainstream press and was technical in nature, essentially the overheating, throughput, and related bugs started there. The other set was firmly based in the financial side, not the financial press like Bloomberg and the Wall Street Journal, they were echoes of the original ‘leaks’ during the quiet period.
A lot more digging into the overheating rumors brought out the same thing SemiAccurate found earlier, the numbers reported and/or alluded to depending on the outlet ‘reporting’ the ‘problem’ all came from real tested silicon. Real tested first silicon, the one known to run hot. One ones given only to high-end tier 1 OEMs who intend on being on the first wave of launch devices for a new SoC. The kind of board that is used only because the other option is software simulation, and the kind of board that is quickly replaced by the first spin.
What you can say is that there are very few of these board in the wild. Other than Qualcomm itself the only ones to have it are the big players with serious technical abilities who both know what they are getting and how the process works. They are tightly controlled and NDA’d to the extremes, numbers almost never leak from silicon this early. If they do leak, they are unquestionably not representative, this hardware is meant for device bring-up, not performance testing.
Why would a journalist use numbers from such silicon and call it representative of the final product? Mainly to get an interesting story and news ahead of the curve. Any journalist with even a vague clue about silicon development would know that the state of such hardware and software precludes getting representative numbers from any tests run. The closest you are going to get is in the ballpark but that is not always the case.
In short you would have to be a rank amateur, idiot, or desperate to put out numbers run on early silicon like that and claim them to be representative. SemiAccurate has had a chance to see, test, and play with such early hardware many times over the last decade, we have a wealth of experience to back up these statements.
So again, why would a journalist take data that was almost assuredly known bad and make a claim that the end product 6-9 months on, 2-3 spins newer, and countless software updates later, is represented by it? There are two reasons, the aforementioned amateur plus idiot combo or they have an axe to grind. That said, don’t preclude both, but if it is scenario A (amateur/idiot), how could they run the tests, it is doubtful they would have the technical chops to plug the power cable in correctly. If it was scenario B (axe to grind), who was it and why? This last one is especially puzzling in light of the multiple times the same rumor kept popping up with new added details on each re-surfacing.
That is what SemiAccurate has been following for months now, doggedly hunting down each and every rumor and seeing what or whom was behind every one. Unlike what we first thought, there is only a single likely culprit, one that we are sure is the root cause but can’t detail the actual reasons we dug up without burning a lot of people who trusted us. In short we fully believe we are right here but we recommend that you do your own research and testing before coming to your own conclusions on this front.
Let start with the overhearing rumor, something that we are quite convinced came from first silicon on the 810. This means the first and likely second set of leaked rumors would have had to come from a major OEM. Although anecdotal both of the first two overheating stories came sources in Korea, a place where there are at least two such OEMs would have had the right dev boards.
If SemiAccurate wanted to get into the spirit of the anti-810 FUD campaign, we would have pointed out how the 14nm Exynos pictured above has a massive heatsink and is not able to fit in a mobile form factor device. Then again we didn’t do that because we know how development works, we actually understand what early silicon is, how normal bugs are worked through, and how devices are made. In short we aren’t rank amateurs, aren’t idiots, know how the system works, and don’t have an axe to grind. We reported the early silicon and dev boards as just that, normal early hardware, we didn’t even mention the heatsink size.
When LG showed off the GFlex2 phone during CES, the overheating rumors were there along with it. No one actually bothered to logically connect the dots that if a late stage prototype device that will be on the market in a few weeks didn’t overheat, the volume variant wouldn’t either. Most of the so called ‘journalists’ and ‘analysts’ were all trying to figure out how much LG had underclocked the early GFlex2’s they were showing off, or how they had rigged the demos. If you tried one of the phones, and SemiAccurate did, you could see they were not underclocked at all. We did, they showed 2.0 GHz for the A57 cores and 1.5GHz for the A53 cores, exactly as expected.
If you test them you could easily see that there was no faking going on, benchmark results were all very close to where they should have been for the stated clock speeds. If you held them in your hands while you did this, and SemiAccurate did, it was also clear that they didn’t overheat, they didn’t even get really warm much less hot.
In short there were dozens of barely pre-production GFlex2’s at CES, all of which ran fine. These were plentiful enough to make cherry-picking silicon on the level needed to hide the rumored ‘problems’ impossible, satellite photos show that there are now new holes in the desert next to Atari 2600 ET cartridge graveyard either. The performance and temperature of the GFlex2’s shown at CES were exactly where Qualcomm and LG said they would be, not where the rumor mill breathlessly claimed they were. Anyone who reported otherwise didn’t test or is actively distorting the issues at hand because of ulterior motives.
The second cluster of data points was in the financial world where SemiAccurate has some rather deep ties. Digging around in this world lead to some very interesting information and a very clear picture quickly emerged. SemiAccurate had multiple sources say the exact same thing and quite a few more corroborated a large portion of the same version of events.
The story they related was quite simple, the Qualcomm Snapdragon 810 simply wasn’t ready for prime time. They said it was overheating badly and had to be downclocked for the LG G Flex 2 even though both were patently and provably false. When pointed to the fact that the G Flex 2 was not overheating at CES and other events, and it was on sale so it could be independently, they immediately responded that it had been downclocked for the design.
When pressed farther, many of these non-technical people could not explain what downclocking was or why it affected power other than to define the individual words. When it was pointed out to them that SemiAccurate dug into the rumors and found it was normal first silicon issues underlying them, again most did not understand what first silicon was much less how the normal development process worked.
At this point it was quite clear that someone was feeding a large number of financial oriented people a quite complex and convincing, at least convincing if you are not versed in silicon development, tale. It was complete, based on the grains of truth listed earlier, and unquestionably designed to paint the 810 in a very bad light. The tale was a classic case of several correct data points that did not lead to a correct conclusion, that being Samsung dropped the 810 because it overheated.
At this point it was quite clear that there was zero chance of things being coincidence, there really was an underlying conspiracy. In cases like this there is one way to find out who is behind it, you ask people who trust you. We did. A few declined to answer but more than we expected said that the information came directly from Samsung. No we are not joking, SemiAccurate has multiple sources who independently named Samsung as the one spreading the rumors.
They were also said to be repeating the overheating information from the two Korean news articles although they did not tell the Wall Street folk the numbers originally came from them. In SemiAccurate’s opinion it is standard operating procedure for FUD campaigns like this to seed a tame news site with FUD and then use it as an “unbiased” source. While we can’t say that Samsung planted the first two news articles, we can say they both originated with Korean sources using numbers that could have only come from a large OEM with sophisticated equipment and people who understand how to use early silicon. Could have been anyone, eh?
But what is the motive?
Now that we know who, and we know the methods of attack, what we lack is a good motive as to why this was happening. Samsung and Qualcomm are close partners, Samsung buys a large amount of Qualcomm radios and SoCs at the moment and will continue to do so for the foreseeable future. Qualcomm is a Samsung supplier, not really a competitor so why go after them? True they can’t fight back at all but what does trashing Qualcomm, the Snapdragon 810, and all the rest buy them? If there was a real problem, that could be a reason to leak the problem, but this FUD campaign goes way beyond that, in fact it goes beyond anything SemiAccurate has seen in more than a decade of covering such campaigns. What is the motive?
This is where things get tricky, Samsung is not actually attacking Qualcomm at all. Sure they are trashing Qualcomm’s products with a baseless smear campaign, pushing out tons of technically correct but known inapplicable data, and sending it to media and financial sources that can’t or won’t question things, but it isn’t actually an attack on Qualcomm. Qualcomm is actually an innocent bystander, the guy walking his dog who gets gunned down in a gang related drive-by shooting. The real targets are not ever named by the rumor source, or only in passing.
Stepping back a minute, lets talk about the psychology of such campaigns and FUD in general. A lot of people think that these campaigns are done by companies with a strong product line and are used to do backhanded PR. Having seen dozens of such campaigns over the past 10+ years we can definitively say that this is never the case, no one attacks like this when they are up. Most don’t even do it when they are down, only the special few resort to such unethical sleaze, and they only do it when they are down and afraid. The more afraid they are, the greater the temptation to take the low road.
Coming back to the devices at hand, if Samsung is behind this, that would intone they are down and afraid, right? One look at the phone sales numbers over the past few years will show that Samsung is either #1 or #2 in worldwide phone sales, up from roughly nowhere a mere single digit years ago. The Galaxy S line of phones is the premium Android brand, no one else is close to their brand image or marketshare. This is down and out? This is cornered and afraid?
Actually yes it is. If you look at the last quarterly results, Samsung’s phone and tablet marketshare cratered, they aren’t even in the top three in China at the moment. The division that is phones and related devices had sales drop 21% in 2014, enough to badly hurt even a company the size of Samsung overall. If you look at the current Galaxy S5, it may be the premium brand phone but it was released to rather tepid reviews. It brought something between little and nothing to the table over the older S4, a device that was very well received. It was a chink in the armor.
Chinks in armor are harmless unless there is someone there to exploit it. You could say that Samung’s mobile line is doing pretty darn well at the moment, dominant marketshare and billions in revenue. The last quarterly results show that such views are only cursory, the loss of marketshare says that someone is exploiting this weakness and exploiting it hard. Basically the snapshot number is pretty good but the trends paint a dire outlook.
That brings us to the question of who is doing the exploitation and the answer is not one but several companies working from two directions. On the top you have LG with their current G3 phone, arguably the most advanced premium Android device on the market. If you compare it to the previous G2, it is a pretty minor advance. Luckily for LG the G2 was a home run smash hit, it singlehandedly took the company from floundering in phones to being the premium benchmark in almost all respects. It is a damn good phone.
To say the G3 is a minor advance on the G2 is missing the point, it started from the top and moved up. The Galaxy S4 was a fine phone but the G2 was far better, the S5 was a tepid improvement on its predecessor. The G3 was minor but not tepid and grew the gap between the flagship lines of the two companies. SemiAccurate’s sources with access to markeshare numbers said the G2 took a chunk of sales from the S4 but the real pain came from the G3 eating into the S5’s sales far quicker than anyone predicted. Curiously though LG’s mindshare and perception gains have lagged far behind their sales but most companies would not be too upset by this state of affairs.
On the other end of the market, that would be the low-end, Samsung is under even worse pressure. Chinese ODMs who were once derided as not being able to engineer their way free from a wet paper bag burst out of the low-end over the last 12 months or so. Anyone who follows the market closely will have seen their progress over the last five or so years, it has been steady, clear, and quite rapid. Just because they started from the basement doesn’t mean they weren’t good, it is arguably harder to do low-cost devices well than high budget halos.
On the bottom end these ODMs have pushed out of the mid-range device category and started putting out some premium devices. While there are dozens of players with a huge number of devices, most can’t match a Samsung Galaxy S5 in all regards but many come very close. Several have better specs than Samsung’s flagship device in key areas and a few even beat it on paper in all ways although none have the panache of a Galaxy. Yet. Do you think the cool kids at the coffee shop will let you sit at the ‘in’ table if you have a Xaiomi phone? How about a year from now?
While SemiAccurate thinks those cool kids will be eating their words in 12-24 months, at the moment Chinese ODM devices don’t impress most westerners. Currently they are somewhere between almost as advanced to better than the so called premium players in most measurable areas, they too are usually damn good phones. That is a problem for Samsung but the real pain is because most of these devices cost half as much as a Samsung or LG equivalent. Do you want to pay $500+ for a name or $299 for something every bit as good but without the fancy logo? In the US most opt for the logo, in the rest of the world, well just look at Samsung’s 2014 results.
So that is the problem, LG on the top and the Chinese ODMs like Xaiomi, Lenovo, and many more on the bottom. Samsung is being eaten alive and it costing the company staggering sums of money, it is a massive problem. They are down and dropping fast without a real way to fight back directly, social and corporate norms effectively prevent direct corporate attacks hence the FUD campaign.
One problem though, if the enemy is LG and ODMs, why attack Qualcomm? Why go after a supplier who is not a competitor when the real enemy is a direct competitor? The answer to that lies in the phone SoC market itself, Qualcomm is the only tier 1 SoC vendor that is not captive. You could argue that they are the only tier 1 SoC vendor period, no one comes close to their radios and their CPU cores tend to be a cut above too. The GPU side is a lot more questionable, Qualcomm doesn’t release enough information to state anything solid about their cores there but they seem to be in the same ballpark as ARM and Imagination.
Not coincidentally LG and all of the ODMs trying to be premium players need a premium radio and SoC to put in their devices. That means they all use, wait for it, Qualcomm SoCs. This round the overwhelming majority of them use the Snapdragon 810, the LG GFlex2 does, the upcoming G4 is rumored to, and most of the ODM’s halo devices do too. The Snapdragon 810 is a damn good part that deserves its premium label.
Samsung on the other hand is the only one of the top-tier premium Android phone players with their own in-house SoC, Exynos. (Note: We know, we know, but we can’t talk about that one yet for reasons we also can’t talk about yet, sorry.) The last Exynos was a disaster which quite ironically overheated badly. Samsung botched their test suite on that part badly leading to thermal mis-characterization on common workloads. Users saw this as overheating and the problem was bad enough for Samsung to abandon the SoC in most markets in favor of the Qualcomm 800 series for their own flagship phone. It likely wasn’t a good day to be in Samsung management when that was decided.
This time around Samsung is using their own internal SoC instead of the 810. You could say that it is because they didn’t botch this one badly like the Exynos 5. You could also say that it is a way for Samsung to differentiate their phones from all their competitors, better yet by doing so with an SoC that the competition can’t use. It is built on Samsung’s 14nm process so it also fills their rather empty bleeding edge fabs. It is a win/win/win for Samsung, maybe with another /win or two tacked on.
…If it performs. And if they can make it in enough volume. And if the radio they use is worthy of the premium name. Since the Exynos uses the same A57 cores as the 810 but is on a 20nm process, CPU performance is going to be a bit better here. The GPU side is unlikely to be a win for Samsung but we will wait to see if Qualcomm actually talks about theirs to the degree that we can make a solid call. GPU power use may again be better than Qualcomm for the same process reasons as the CPU cores. On the radio side, as we have shown above, Samsung isn’t close and the next generation part won’t close the gap anywhere near enough, nor will the process help much here.
By going with their own Exynos, Samsung fills their fabs, has a 14nm pipecleaner part, and doesn’t have to pay the margins required by an outside vendor. They also get to differentiate by being the only premium Android phone vendor who is not using a Snapdragon 810 in their halo product but this may not end up being a win, consumers are fickle about such things in premium devices. It is SemiAccurate’s semi-accurate view that at best Samsung’s Exynos will be a bit better than the Qualcomm 810 in most respects but any difference either way will not be noticeable by end users.
That a problem, a big problem. Samsung’s competition is taking huge chunks of marketshare, has better products on the high-end, significantly cheaper but technically equivalent parts on the low-end, and is advancing faster than Samsung. Samsung needs to hit one out of the park with the Galaxy S6 to hold share, basing designs on non-differentiating performance from the SoC and lesser radio is not a good start.
This is why we say Samsung is scared, they have good reason to be frightened. With the launches of the S6, G4, and countless ODM Snapdragon 810 products happening in the coming weeks, what can they do? If you can’t win on merit, FUD. They are. The thoroughness of the FUD and smear campaigns strongly intone that Samsung is going to take a pounding during the next product cycle and that they know it too. If they had a winning product, they wouldn’t have gone to the extraordinary lengths they did to attack LG and the ODMs via the proxy of Qualcomm.
While SemiAccurate has to tip their hat to Samsung for the complexity and thoroughness of this campaign, everyone else should see it as a big red warning sign for the phone market. The times they are a’changin and the future is still up for grabs. About the only thing you can say for certain is that the 810 was an innocent bystander in this war, go buy one for yourself and see.S|A
Latest posts by Charlie Demerjian (see all)
- ARM adds CMN-600 interconnect and DMC-620 memory controllers - Sep 27, 2016
- Globalfoundries 7nm process isn’t even close to the name - Sep 26, 2016
- ARM upgrades realtime offerings to v8-R and adds Cortex-R52 - Sep 21, 2016
- Everspin and Globalfoundries team up for embedded ST-MRAM - Sep 15, 2016
- Intel’s Xpoint is pretty much broken - Sep 12, 2016