Geekbench's John Poole discusses how Apple balances performance and efficiency, how Samsung is ramping up single-core, and where Qualcomm — and maybe Google? — fit in.
Listen to the podcast version:
The Exynos Kerfuffle
Rene: The reason I wanted to talk to you is that whenever one of these...I am going to call them a kerfuffle because it sounds like a funny word. Whenever one of these processor, kerfuffles, comes up, you have a lot of people...you have so many things to deal with.
First, you have people who don't have what's suddenly considered the good processor being all salty. You have the companies trying to say, "Nothing is wrong. Everything is fine." You have people like you, people like a non-tech trying to dig into it.
You have the blogs who just take whatever you say and try to make it into an exciting story, regardless sometimes of what you say.
I figured it'd be good to go to source. For people who aren't familiar, what happened was the "Yonhap News Agency" put out an article saying that the battery performance of the Samsung Galaxy S9 smartphone trail behind rival products.
Industry watchers said, Monday, causing consumers to express his content. I think this was based on a strategy analytic survey which may have been sponsored by Sony, but it was comparing a bunch of different phones. It turned out...we can get into awhile later, but Samsung has split processor strategy where they have Qualcomm in North America and Exynos in parts of the rest of the world.
The Exynos version was not performing as well as the Qualcomm version when it came to battery efficiency. Phone Arena also found similar. The non-tech, they're usual real deep dive. They said it absolutely felt flat on the face in this test, and post to the worst results among our tracking of the latest generation devices lasting three hours less than Galaxy S8 with an Exynos processor.
Yonhap have theorized that it was because Samsung was being more conservative, given what happened with the recalls around the Galaxy Note 7. That didn't seem to be the case. Correct me if I am wrong, John. It looked like to AnandTech that they were instead of doing race to sleep, they were doing what they called walk to burn, where the increase in the performance course was resulting in a similar decrease in the efficiency.
John: Right. One thing I want to say before I get too much into this is that we still are trying to get our hands on one of these new Galaxy S9s with the Exynos processor, the joys of living in North America and not being to access these as easily as other locales. Some of my comments, we're anxiously awaiting being able to do our work into this and take a deeper dive into it ourselves.
But what Samsung's done we think, from what we've seen from other sources such as AnandTech, is they've actually come close to hitting Apple performance in their mobile cores. I've said this on multiple occasions before. I've never understood why the Android vendors has had such a huge gap in performance compared to Apple.
What I think Samsung's done is Samsung's really the first mobile SoC that's come within striking distance of Apple. If you look at, say, the Geekbench scores, I believe the Exynos gets about 3,500 or so whereas Apple's at just a smidge above 4,000, 4,100, or something like that. They've really closed that gap.
The problem is when you have that much performance is that are easy ways to get there and there are hard ways to get there. The easy way or maybe not even necessarily, easy being a very relative term here, is you're going to use a lot more power.
With the poor battery results that we're seeing, with the complaints about battery lifetime in the Exynos chipsets, that makes a certain amount of sense because these big cores could be using a lot more power than the previous generation Exynos and the current-generation Snapdragon.
Single Core Bottlenecks
Rene: It does seem to me for a long time, they just didn't have that single-thread efficiency. People blamed that on everything from UI lag to basically roadblocks. Android, like Apple, did Grand Central Dispatch years ago to try to multi-thread things. Android has really been focusing on multi-core, multi-thread performance.
But my understanding is that you still have a highway that if it ever gets blocked down to one lane, you are restricted by the speed of that one lane. That's why single core is still important no matter what you do on multi-core.
John: Exactly. Single-core performance, this is one of the reasons why Geekbench breaks down performance in just those two separate numbers, a single-core and a multi-core number, is that there are lots of applications. Speaking as a developer, having multiple cores is great because that can greatly improve my workflow.
But for most people and for most operations, what number really matters is that single core performance because there are just certain operations on your phone, on your computer that aren't going to be able advantage of multiple cores or might only be able to take advantage of one or two cores instead of six or eight cores that are available.
You really want that single-core performance to make sure that your system on your phone or your laptop is snappy and responsive. With Samsung and the new M3, they've got that single core performance, which is great and it's amazing. The problem is it seems to have come with a huge power draw.
It's almost like the megapixel race all over again. But if we equate cores to tires, having 18 wheels on a Mack truck is very different than having four wheels on a Ferrari. It's just the number of wheels that matter.
The increased performance and there's a couple things that can happen. One is if you were incredibly inefficient in the past, then you have a whole bunch of efficiency that you can spend when you increase performance.
But Samsung as far as I know, their chipset team is really, really good. They probably didn't have a lot of poor efficiency they could draw on to increase performance. You've got to pay for that somehow.
I know Apple's been working towards building these performance cores for years. But if you do that leap year over year, I think you sort of end up paying for that with efficiency. You're doubling the amount of performance. That has to come from somewhere.
John: There are certain things in the design that you can do when you're doing that big leap forward so to speak where you might be able to mitigate some of the energy that you're going to have to spend.
Because these performance leaps, we're well past the age of low-hanging fruit. It used to be you just ramp up the clock speed, maybe make some tweaks, and then you go home. That era's over. We can't really increase clock speed anymore.
You're in a small package. You've got a battery that you have to worry about. You've got the fact that this is going to live in somebody's pocket for two, three years at a time. It's a very constrained environment.
When they design these new processors, they have to come up with ways of where can they extract more performance. It basically involves a processor looking for ways to do more things at once. That search, that process of looking for more things to do, takes more energy. There are ways to do that that are arguably more energy efficiency than another.
If you look at how Apple's done it with their cores, there's an obvious point. You can point to them and say, "Apple's been able to figure it out." But Apple's also had a multi-year lead doing this. They've been able to move incrementally and such.
Samsung came out and basically doubled performance in on generation which, as much as Apple's continually refining, they haven't had that huge jump forward for a while now. Samsung, they're playing catch-up.
It might be that maybe they focused more on performance rather than efficiency for this round. I'm sure the next generation of Exynos will be a much more efficient core.
It's just simply a case of do you want that huge leap forward now or do you want a smaller leap and maybe then you're playing around with your energy budget a bit more. I think it's a case of trade-offs in why the energy use is so high.
The true enemy is time
Rene: I actually never thought about this, but it was explained to me when I was asking around about it that a real gating factor of processor design is time, that if these design teams had infinite time, they could make you the absolute best processor in the world.
But they don't. They have a year-over-year cycle. Maybe they have a year or two years or three years of lead time. But to your point, when you double it and you only give them a year to do it, it's not the same as saying, "Double it in the next 10 years," for an example. There's a lot less you can do.
John: That's one of the things I know, being involved in a lot of planned obsolescence discussions in the last few months, people look at Samsung and Apple and all these other companies and say, "Why don't they take that extra year and do a better design?"
It's like, one, you've got all your competitors are on a year cycle now. Everybody's expecting Apple...God forbid Apple should ever take a year off and not release a new iPhone and say, "No, we're going to do a better job next year. Don't worry. We're just taking a break." I think people would scream bloody murder. Also then their competitors would say, "Well, great. Thanks for the extra year."
There's a lot of neat things you can do if you do things incrementally and you have that nice year-over-year improvement. But I'm sure if the Samsung design team, the System LSI folks, had a two or three year design where they could focus on just getting the M3 out and getting the M3 perfect, I'm sure we wouldn't be having this discussion.
But I think it's just simply the reality of business of the ecosystem of the market. They have to be on this year cadence. Sometime you're going to have to cut corners somewhere. Unfortunately, it looks like energy efficiency was one of those places where they had to make some sacrifices.
Rene: I think it's funny that you mention that because I was talking about it. One of the chip analysts said, "Apple's got to pick up the pace. China's putting out great new phones every six months. Apple can't stick to this slow once a year pace."
They were talking about the new Pro camera that has three lenses on it. Some of them have two notches now. One notch isn't enough. Or Huawei where they're super aggressive. The one-year schedule was already too slow.
I was like, "Aw, come on. I don't think this is sustainable."
John: Just think of it from the flip side of people complaining about Apple's release schedule for software right now, saying a new operating system every year is too much. We want 18 months. We want two years. There's definitely a balance there.
I don't even know, just thinking about the realities of manufacturing pipelines, even a year seems aggressive to me. Shortening that down to six months just seems insane.
Rene: Not to get off on a tangent, and maybe we'll revisit this later, you look at what's happened with Intel and their drive to get to 10 nanometer. They went from tick tock, tick tock to tick tock tock tock, oh damn, tock tock, oh crap again, tock tock.
John: I have no idea what's going on at Intel. I've talked to a few analysts about this. They're all shaking their heads. Intel had this huge process lead. Now they're behind. They're last. It's a bizarre situation we're in now.
A tale of two Samsungs
Rene: Some people I think don't understand this either is that Samsung is a foundry as well as a design company. They have in some years manufactured Apple chipsets, and they manufacture their own chipsets. But just by being a manufacturer doesn't automatically mean that you're a leader in the design department. Those are two separate skill sets.
John: It is two separate skill sets. I'd say definitely having the foundry in-house helps. I know Intel's talked about the efficiencies and the advantages of having the foundry in-house where the foundry design can influence the chip design and vice versa. There are definitely efficiencies there that you can have.
Samsung I don't think has really taken full advantage of that. I don't know whether they will.
I know the structure of Samsung is very different than a North American company where they very much keep their different departments siloed, but it'd be interesting to see if there are any synergies. Definitely System LSI now is really...
I still think even with the energy issues, the M3 is a great core. I'm really excited to see what the M4 does. But it's a case of it's growing pains in a way. As I mentioned earlier, they're the first Android SoC that's come within spitting distance of an iPhone in terms of single-core performance. We just haven't seen that from other manufacturers yet.
Rene: Apple is different in that all iPhones use Apple silicon. They know what the roadmap is two, three years out. The silicon team can target the features that the features team wants, and the features team can target the capabilities that the silicon team provides them.
Three years ago, they could start working on the neural engine and FaceID at the same time and deliver them both with iPhone X where if you have, like Samsung, half your phones are going to get Qualcomm, half of them are going to get Exynos, it's much harder to...You can't guarantee which chip your users are going to. You can't build those silicon-specific features.
You can layer RAM on top of the camera as a workaround to do that, but I think you lose the benefits of having in-house silicon that way.
I thought, "Oh, this is interesting. They're really trying to maybe take advantage of some of that vertical integration that Apple has where at least if they knew everybody's running the same chip and the same OS and the same hardware design and they can sort of consolidate that, that would be great."
But the next S8 or whatever it was went back to the split Exynos, Snapdragon. I thought, "OK, well, I guess not." I've never understood why Samsung does that bifurcation. I'm sure they have reasons, but I've never quite figured them out.
I think you look at Apple and that vertical integration they have, I think that's one of the things that Samsung...It's clear that they've looked into this and they've tried this, the talk of ties-in a few years ago. Clearly the more that they're investing in their in-house CPU design, the fact that they have their foundries, I think these are all pointers that they'd like to do this.
I think in an organization that big I imagine the political factors that influence we're going to do this instead of that are huge and hard to surmount.
Qualcomm is two companies. It is a patent licensing company and a chip design company. The chip design part is not their biggest business.
I think a lot of companies, unless you're Apple and incredibly stubborn, find it easier to acquiesce to Qualcomm's chipset demands in North America where they own so much of the CDMA patent portfolio than it is to fight them and force your own silicon in there.
I love that Apple bullies the carriers. I know that's a horrible thing to say. But I think if there's anyone who should be bullied, it's the carriers. Huawei could, but they have so many problems with just the US government and being China-owned that politically they can't bully the carriers.
But I think Samsung is one of the companies, maybe Blackberry back in the day, that could start exerting its will upon the carriers.
Really the smartphone market now in North America is consolidating around Apple and Samsung and then everybody else at the fringes. I'm sure it's very similar in the States where you walk into a mall or you walk into a cell phone store, and most of the advertising is either for Apple or for Samsung. That's really what people are being pushed toward.
I'd love to see Samsung use its influence as a force of good in the industry for the consumer. It's seem they're not quite at that point yet though.
That Google silicon
Rene: There's rumors that Google is going to enter the custom silicon market in a big way. It's going to take a couple years to wrap up, but they are serious about having their own iPhone. We've seen that with Pixel 1 and Pixel 2 which currently use Snapdragon systems on a chip, bu I wouldn't be surprised if they shipped with Google systems on a chip within a few years.
John: Google's been doing all sorts of interesting things with dedicated hardware for machine learning, the custom TPU units. They're more data center-focused than consumer-focused, but it's clear that Google understands the value of designing your own silicon, designing your own processors in-house and being able to fully tweak them and tune them to your needs.
It'd be really interesting to see if they do decide to design a custom SoC and a custom core where they go with that.
The efficiency gap
One of the things Andre said that was interesting is he said it felt like Samsung LSI -- which if you're not familiar with Samsung LSI, listeners, it's the silicon division, the company that makes all the chips as opposed to Samsung Electronics which makes the phones -- they're one generation behind when it comes to efficiency.
He's comparing them to ARM, but I think it's probably fair to compare them to Apple as well. They continue to be one generation behind when it comes to efficiency.
But he says currently at 17, 20 percent performance lead does not seem worth the 35 to 58 efficiency disadvantage along with the two times higher silicon area cost. That was one of his key takeaways here. Time again as a gating factor, do you see a way for them to catch up or, unless they have some sort of big revelation, is this going a pattern going forward?
John: I was thinking about this. What I found interesting was that if you look at what Apple's done for their die sizes, and Apple's dies are quite large as well, I'm sure there are non-linearities there where to get 10 percent extra performance, you're going to end up spending 20 or 30 percent extra die because nothing's ever easy when it comes to electronics.
But if you look at what Samsung's done and compare it to what Apple's done, Apple doesn't put four big performance cores on their SoCs. They put two. Maybe they put three in an iPad, but they really do limit the number of performance cores which in turn also helps minimize battery life use.
Right now if I were to take the same workloads -- let's say hypothetically speaking I had something that used four threads -- and I ran it on an iPhone and I ran it on a Samsung Exynos chip, the iPhone's going shunt some of that work off to the low power, the high-efficiency cores whereas Samsung might happily light up the big performance cores.
Even something like that I was surprised to see that they still stuck with that four big, four little design that they used to use when single core performance wasn't as great. I'm almost wondering if they would have been better off to limit the number of big performance cores.
Now that they've got really got single core performance, there's not the need to have four of them. Why not put two of them in and then shunt other stuff off to the low power cores, the efficient cores? Part of that might just be...
That would reduce your die size, that would reduce your power usage, that would reduce your thermal output. It might have solved a lot of the issues that people were seeing because Android much more so than iOS will happily use all the cores that you've got in your phone.
Eight core chips in Android have been common for quite a while now. Android takes full advantage of them when doing all sorts of things that maybe aren't necessarily latency-sensitive. Even a really trivial example like rotating your phone when you go from portrait mode to landscape mode, there's a lot of work that goes on behind the scenes to make that happen. That will use all your cores.
If you've got four big, power-hungry performance cores, it'll happily use those in which case then that's going to impact your battery life. There might be other ways that this comes out too.
But even still, even if that's not the case, then you could have done some other interesting tweaks with the performance cores, the big cores, where if you only have two of them, maybe you can run them at higher frequency, maybe you can get more performance out of that.
I know Andre did all these experiments with writing his own custom scheduler and whatnot for the Exynos and found if you do all these sorts of interesting things, then you get more performance or you get better battery efficiency. There are ways to do trade-offs that Samsung hadn't done in their stock implementation. It'd be interesting to see if Samsung actually does any of that work as well.
To go back to my original point, just from a pure efficiency standpoint, I think even doing something as simple as removing two of the cores would have been a huge advantage for them.
Rene: I always try to assume that the chip companies aren't stupid. It's easy to say, "Oh, they're just dumb. They made a mistake. They can fix it in software." But I try assume at first that they're not.
It's not even a guess, but I have this wonder that they know what they're producing in terms of chips. They might have made a decision to run it this way because they knew, depending on the yield or the bins or the performance of individual cores, that they would have a bigger problem if they implemented the fixes at scale.
Because anything's easy on one chip, but to do it on a million chips, you might have outliers. Either they run too hot, they disperse too fast, they have other problems. This might have been the safest balance they could have found for the entire run of the chipset.
John: It might have even been something as easy as just simply a spec war because I know six core designs, the two big, four little designs, in the Android market haven't been well-accepted in the past. I know outside North America, people are much more attuned to those sorts of spec sheets than, say, the folks in North America.
It might have just been viewed as people would view do you want the eight course Snapdragon with the four performance cores or do you want the Exynos with the two performance cores? It might have just been something as simple as it just would come off as a less powerful processor, even though it wasn't. It could be even marketing decisions like that. Who knows, though?
The benchmark business
Rene: I hate relitigating this, but we have in the past where they would amplify their frequencies for Geekbench and other benchmarking apps even if those weren't the actual frequencies for the chip because they wanted so much face out of getting those high numbers.
It's always unfortunate when you see people playing games whether it's boosting stuff past the stock frequency, whether it's removing thermal lineaments, whether it's changing the way that this operating system interacts with threads in the cores just to provide higher benchmark scores. I'm really glad that that's something that seems to be as the industry we've moved past.
But at the same time, I could see that sort of thing influencing decisions where maybe you do something where it's not maybe necessarily the ideal for most applications but it'll give you better benchmark number. I can definitely there would be people in the industry that would think, "Oh, this is the reasonable trade-off to make."
Rene: A long time ago, I'm not going to name any names, there was a vendor who said, "It doesn't mean so much to me what camera we have as long as Best Buy puts the CNET best camera sticker on it on the shelf." That was the most important, not the actual component but the perception of the component.
We see that with the DxOMarks and we see that with Geekbench marks, fiercely competing for what are in the end very slice-of-time specifications.
The other thing that you said that I thought was so interesting is when you look at what Apple did, they came out with 64 bit with the A7. That was a bit of a shock in the system to mobile processors.
But then when those started being super-high performance, they had so much gap underneath that they slid in the efficiency cores to make sure that they weren't sacrificing too much performance for efficiency on the high end.
Then with the next iteration, they close to doubled the performance of the efficiency core. Even the efficiency cores now are fast. They also broke them up so that they didn't have to be paired. You could use them all separately or jointly.
That looks like an interesting step. I called them performance.EFFICIENCY and got a really nasty look, but I do like the idea of that kind of architecture.
Because Android vendors have been doing this for years, Apple's now doing this too, I think it's a great way to balance the power consumption that comes with these big, fast cores where you'll need them for certain things.
You're logging in. You're taking a picture of your kids. There's just that brief moment when they're being adorable and they're not hitting each other. You want to capture that versus when I'm streaming Spotify. There are lots of things that are time sensitive. You really want that great performance, but a lot of the rest of the time, you don't care as much because it's not as critical.
I think having those efficiency cores paired with the performance cores is a great way to achieve that because there are so many design decisions that go into performance, that you're going to sacrifice so much power usage for that. Whereas if you've got these nice efficiency cores, then you can say, "Well, OK, yeah. We'll use all of our energy budget on these when we really need it."
But then we can fall back and use these really still performant but not as fast, but they use so much less energy. I think that's a really great design. I'm glad to see Apple's adopted that.
Because one of the tricky bits has always been when do you move stuff from one set of cores to another? When do you use the performance cores? When do you use the efficiency cores?
I'm hoping now that the industry, I think Intel's really the only manufacturer for mobile parts that doesn't do this, I'd love to see more open research into how do we make the best advantage of this because even talking to people...
I'm completely unsurprised Andre's found some low-hanging fruit when he did his work on custom building the Exynos stuff because it's such a fertile area of research. It's such a fertile area for improvements. I'm sure Apple internally has been doing amazing things with that because they've got such a tight coupling whereas Android obviously it's a bit more Wild West out there.
There's so much interesting stuff people could be doing to improve efficiency even without hardware changes. I'm really curious to see it. I'm really curious to see if Samsung, especially with the M3, takes some updates as an opportunity to tweak behavior and improve battery life or improve battery life or do both.
The controller keys
Rene: I think Apple hasn't been too secretive about this. They've said part of their secret sauce is the custom controllers that really handles all this and that's an integral...Having the chips is one thing. Knowing when and how to apply them is something else.
Otherwise, instead of race to sleep, you do get walk to burn. You're just spending battery on nothing which is what you were talking about earlier with so many performance cores.
The other thing that's interesting to me about all of this is you have this market where you have customers who want infinite battery life, limitless performance, and zero weight on the device. If it would float and be lightning fast and last forever, they would be happy. Anything less than that...
So you're not only balancing the incredibly hard...It seems like dark arts honestly that this stuff works because you have to battle...
You have to balance the thermals of the processor against the size of the battery, the RF transparency, how fast these things ramp up, all of these factors to get to this product. Then we just complain about whichever one they took the biggest compromise on. [laughs]
John: Exactly. Looking back on what the Android phones have been, I've publicly complained about this, and I've said they're slow. Looking at single-core performance, they'd fallen so far behind Apple that a new phone would come out and be a, "OK, great, you're still hovering around the same point in performance that you've been for the last two years. What's going on?"
Now, of course, that Samsung's caught up, it's like, "Well, your battery life's gotten bad." The question is I'd love to see daily usage, because I know, speaking as someone who has written a synthetic battery test, these things can be torture tests, and they can exacerbate poor design decisions that were made that may not necessarily have a huge impact on the day-to-day use.
This is, again, coming back to why I'd love to have one of these phones in my hand, so that we could play around with it here in our lab to see what's going on with it. To see how much of this is a you run a intensive battery test app on it and it's going to drain really quickly, versus a "I'm able to make it through a day without a charge."
For most people, that's their gold standard, is that if they can start in the morning and get to the end of the day without having to recharge their phone, they're happy.
It could be that the Samsung S9, regardless of whether it's a Snapdragon or an Exynos, can do that. It's a case of maybe you can't get through two days with it, which is one of the things I will say.
I've been using the iPhone X lately, and I love the fact that I can now go about three days without having to charge it. It feels like magic to me. If all of a sudden it dropped to two, I don't think I'd complain that loudly.
Limitless power, endless battery life, zero weight
It's not like they have a static target, say, "OK, this is what people are doing. We're going to double battery life." It used to be that you had PIN-based messaging and WAP browsers. With the iPhone we went to the WebKit browser and you checked your email.
Now you might have Snapchat running a lot of the day, with Geofilters, screen on, always downloading large video files, playing them over, keeping the radio lit up, keeping the screen lit up, keeping the GPS lit up. Like, "Oh, my battery wore down by 3:00 PM because I was out being an influencer all day." [laughs]
I expect to be able to, regardless of where I am, load up Tweetbot, refresh things, and see what's going on in the world. I expect my watch to tell me when stuff's going on, and that needs the phone to be in reach, and be able to use a transmitter, and all that stuff.
I'm sure if we went back and started using an original iPhone we'd be shocked at how little functionality we actually used on there. Even looking at screen resolutions.
If you look at the screen resolution, this is a slight exaggeration, but not too much, the whole screen on the original iPhone is roughly the size of an icon on the iPhone X, give or take a little bit. You're driving more pixels. You've got such nicer displays, so much faster processors that are doing more.
The web pages we're viewing, even over the last 10 years, have gotten more complicated and there's more interactivity. There's so much we expect from our phones. When it falls short in one area, we throw our hands up, and we're like, "Oh, this is horrible, and this is the worst."
There are times where I've definitely done that. There are times where people are being a little unreasonable. There are other times where it's like, "Yeah, this makes total sense."
With this particular issue with the S9 and the Exynos chips, I'm not quite sure where that falls. If you've got a phone, you're using it lightly throughout the day, and it dies at 3:00 PM, that's not great.
If you're out being an influencer, you're doing all that sort of thing, and your phone would die anyway, regardless of which one you're using, maybe take a step back from Instagram a little bit.
State of the mobile silicon
Intel's probably not there anymore. Nvidia seems like they could do a lot, but they don't do a lot yet.
How do you look at it now, someone who's been doing all this for years and has seen desktop performance end up where it is? What's your sense of the market now for mobile silicon?
Apple's got a little bit of an odd point in the market, where a lot of people aim to be where they are, but it's more from a whole package. You're not necessarily going to have someone say, "Well, we'd pick an Apple chip set, but we can't, so we're going to go do this," or something like that.
It's going to be more of a whole phone. It's going to be Samsung worried about selling an S9 to somebody, instead of them going off and buying an iPhone X.
If you look at -- ignoring Apple -- Qualcomm still has that issue that they've struggled with in the past, of the lower performance thermals and whatnot, which they seem to be getting a handle on. The 845 looks like it's not as huge of a leap forward, but they're moving in the right direction.
The 845 is a definite improvement over the 835. These things, they'll continue to move forward.
Looking at MediaTek, MediaTek's going at it from a different angle. If you think about Apple, Apple really targets the high end of the market, and not much else. MediaTek's more interested in the budget smartphones, if you will, like something where you go, you pay $200 or $300 out of pocket, and you get the whole phone.
They're doing a great job going after that market. Every once in a while they do something completely crazy, like they had a 10-core smartphone chip that I thought was ridiculous, and I don't actually know if anybody ended up using it.
You've got MediaTek coming in at the bottom, and Qualcomm, of course, exists at both the high end and the low end. Exynos exists at both the high end and the low end, as well.
It looks like they're mostly licensing ARM's existing cores, which are great cores, but there's not much of a competitive differentiator there if you're pulling those cores off the shelf.
What it seems to be that Huawei's doing that's kind of interesting is that they seem to be spending more time on doing AI, the neural network accelerators and whatnot, and seem to be getting some really promising results from their early silicon in that.