At Mobile World Congress, there were few companies with a swagger in their step. The rapid growth of the mobile market over the last few years has created just a handful of winners, and a lot of losers. Intel, after one too many failures in the smartphone arena, seemed almost bashful about its mobile efforts this year. Nokia, no longer the massive powerhouse of mobile tech that it once was, hid behind a cutesy Finnish facade that was no doubt designed to discourage body blows from journalists. Microsoft, despite rebranding an entire hotel with 10-storey Windows decals, had almost nothing to say, asking us yet again to hold on for a bit longer.
And then there was Qualcomm. As I watched the Qualcomm execs strut around, their chests puffed out, fresh from uncountable victories over the last few years, I wondered if they had actually stopped to scoop up the bloody dribbles of confidence left by Intel and the fellow vanquished upon the canvas floor of the mobile boxing ring. For a moment, I considered if there was a Second Law of Confidence – that confidence can’t be created or destroyed, only transferred from loser to victor.
Qualcomm has had a few very good years. This is a company that, 10 years ago, was basically just an R&D company that specialised in developing mobile IP (intellectual property). Today, just 30 per cent of Qualcomm’s revenues derive from IP licenses (chip designs, patents) – in particular, Qualcomm owns most of the patents behind the UMTS air interface present on almost every modern smartphone. The rest of the company’s (rapidly growing) revenues stem from what it calls its chipset business (QCT) – the SoCs and modems that power a large percentage of the world’s mobile phones, but also other parts of the mobile telephony stack as well, such as base stations and femtocells.
Qualcomm’s quick growth in these markets has been almost entirely down to two factors: Being first to market with key products (the first 28nm LTE modem), and highly integrated, high-performance SoCs that are easy for OEMs to deploy. These two factors then led to an even more important third: Incumbency. Because Qualcomm is already in so many shipping devices, and because it’s much easier to stay with a platform that your engineers already have experience with rather than move to a new one, OEMs will stick with Qualcomm until something far greater comes along. It is not enough for Intel to offer an Atom chip that merely breaks even with Snapdragon – it must blow Qualcomm completely out of the water.
Over the last couple of weeks, I’ve spoken to executives from both Qualcomm and Intel about their future efforts in mobile. Both companies, as you’d expect given their almost diametrically opposed backgrounds, are taking very different approaches. Intel is continuing its efforts to develop competitive wireless, graphics, and other important IP blocks for mobile applications, but ultimately it’s still relying on its process advantage – specifically 14nm later this year – to deliver the product wins. Qualcomm, on the other hand, is hammering home its advantage in home-grown, highly integrated and efficient IP blocks to create mobile platforms that are more eloquent.
The difficulty of building a truly great chip
At MWC in Barcelona, I sat across from Raj Talluri, Qualcomm’s senior vice president of product management. When it comes to laying out Qualcomm’s future SoC roadmap, the buck stops at Talluri – but as it turns out, his background is actually in computer vision and psychovisual work, and so our conversation actually began with me questioning him about 4K.
Are we actually ready to move to 4K? Isn’t it just a case of trying to market the Next Great Thing to hapless consumers? I point out that there’s almost no 4K footage out there, no 4K broadcast standard, and no 4K optical disc format. Isn’t 1080p enough? Doesn’t resolution cease to matter once you reach the PPI of Apple’s mystical Retina display?
Eventually, I steer the conversation back to the topic of SoCs: But what about power consumption?
“That’s the challenge,” Talluri says. “With this one,” he waves around a Sony Xperia Z2 with a Snapdragon 801 SoC inside, “we’ve really nailed it. Our power consumption figures are so good that you can record 4K with the same power consumption that you used to record 1080p.” I ask if that’s purely down to a dedicated 4K hardware block on the SoC, but Talluri says it’s much more complex than that. “We do a lot of things that people don’t quite think about.” He talks about how shooting a video involves encoding, camera stabilisation, auto-focus, processing (white balance), audio synchronisation, and writing huge amounts of data – while being able to pause it all if you receive a phone call. “Dedicated hardware is one aspect of it, but it’s more about system engineering.”
And so we finally get to the crux of why Qualcomm has such a competitive advantage over other mobile chip makers. Talking to Talluri, I start to appreciate that building a high-performance, power-efficient SoC is only partially about the specific hardware blocks. Much more important is how those blocks are connected, and – this is the key – how those blocks are actually used.
“The way I look at the problem [of efficient mobile computing] is heterogeneous compute,” says Talluri. “You have a set amount of computation to be done” – say, taking a photo – “that can be done on the CPU, the GPU, or the DSP, or the camera engine, the video engine. You know, we have so many blocks.” Here he’s referring to each of the discrete blocks of transistors on a Qualcomm SoC that are designed to do just a handful of very specific tasks. “You have to work out where that particular kind of compute fits best – but if that resource is tied up – say you take another picture while the camera engine is tied up, you could use the DSP instead.” So the photos can be processed on the CPU, GPU, camera engine, and DSP? “Right,” says Talluri. “We are constantly migrating between compute elements.”
“But, ultimately, the best power [efficiency] you can get is when you build dedicated hardware for that function,” Talluri says. “The risk is, sometimes you don’t know what you need, or the algorithms are still being defined.” He’s talking about the fact that, for example, you probably don’t want a dedicated H.264 block if phones are about to switch to H.265. Talluri actually calls out Pelican Imaging, a computational photography company we’ve covered before, as another good example, “We could hardwire the processing chain and get great performance and power, but they’re still changing their algorithm. At times like that, what you want to do, is take that processing and map it to a combination of the CPU, GPU, DSP, and all that to make it work.”
“That’s how mobile processors have always been built – different from PCs.”
I have to admit that most of this information is new to me, and so I ask a few more probing questions. I assume there is a lot of analysis into how people will be using smartphones and tablets in a year or two? “So, I’m actually working on anything that you will use four years from now,” says Talluri. What will I be using in four years, I ask? He laughs. “Well, we usually classify it [future product specifications] in more generic terms. You’ll be using a lot more sensors which are extremely low power – you saw Samsung talk a little bit about health and heartbeat [at its Galaxy S5 event the night before the interview], but there could be a whole bunch more. You could be running 8K video [in four years], too; now that we’re doing 4K, you can clearly see the migration to 8K, it’s just a question of time.”
Which eventually leads me to the question that I’ve been sitting on since the interview started: “So…” I begin with no small amount of trepidation “… what’s after Krait?” Again, Talluri laughs, and I know I’m not going to get a straight answer. “We haven’t announced it yet. But we’ll have something good.” Come on, maybe you can give me something, in general terms? “We’ll move instruction set to ARMv8, we’ll move the performance up, reduce the power, new features, better caches, all the good stuff.” ARMv8 is the new 64-bit instruction set used by Apple’s A7 chip, and the new Cortex-A53 and A57 cores.
How close are we? Another laugh – this time he cackles a bit. The best I can get out of him is a kind of shrugging agreement that Krait’s successor is coming soon. Personally, I know that Apple’s 64-bit A7 SoC surprised a lot of people in the mobile industry, and I’m sure that Qualcomm wants to get Krait’s 64-bit successor out the gate as soon as possible, so that it doesn’t lose its perceived edge. I would be surprised if Krait’s successor doesn’t arrive before this time next year.
“Well, the way I think about it is – which is something we haven’t done such a good job of – since we do our own processors and our own SoCs, almost every time we make a new SoC, we make a change in the architecture to the processor. We may not have done a good job of naming them differently, though,” says Talluri. “Like, for example, there is actually a Krait 100, 200, 300, 400, 500 – they’re actually very different processors. There’s a lot of performance gains. Moving blocks around, changing caching structures, pipelines, IPCs – IPCs actually went up a lot. We just don’t call it a new processor.”
“You shouldn’t just think of Scorpion, Krait, when is the next big thing on the timeline – every year we’re innovating on a new CPU.”
I ask whether Krait’s successor will debut in “Snapdragon 1000,” or whether it’ll be paired with a whole new mythical creature. All I get is a grin in response.
The shift to 64-bit, graphics, and TSMC vs GlobalFoundries
Apple, by virtue of controlling the entire hardware and software stack, had a relatively painless transition to 64-bit with iOS. How do you think Android will cope with the shift to 64-bit? “The way the ecosystem works,” says Talluri, “is that we have to make a processor, and then work with the OS guys – not just Android; Windows, Tizen, whatever OS we support – and then all the app developer stuff. When you’re a processor maker, you have to work through the whole chain, and we are.”
I shift the conversation to Qualcomm’s graphics efforts. This morning, I lazily ask, I think I was reading this morning about a new 192-core mobile GPU… that was the Adreno 400, right? “No, no, we don’t talk about cores in GPUs,” Talluri fires back quickly. “We’ll leave core counting to Nvidia,” the Qualcomm PR rep who’s sitting in on the interview says a little snidely. “It can be hard to define what is a core,” says Talluri. “We usually talk about features; OpenGL, DX11.”
By this point, the interview is starting to lose steam. We’ve covered a lot of ground, and I think we’re both a little mentally drained. So I throw a curve ball. So I assume you’re staying with TSMC for the foreseeable future? I ask tentatively. “We… use multiple foundries,” Talluri cautiously responds. Not for Krait, though, right? That’s all one foundry? There is some squirming on the other side of the table, but eventually the PR rep shuts us down. “We don’t parcel out which products go to which foundry. We’ve never disclosed that. But we use multiple foundries.”
But you don’t move products around, right? They stick at one foundry. “Sure. You make the decision to take one product to one foundry, and one into another. You don’t just mix and match,” the PR rep says. Talluri cuts back in: “[In choosing which foundry to use] it’s about the whole SoC. The CPU is only about 15 per cent of the die size. So it’s where we put the whole SoC; it’s complex, there’s a lot of blocks, digital blocks, analogue blocks, GPUs… and so we make a decision based on the whole system, where it best fits, the volumes of it, the timing. Again, unlike the PC space, we have to make a system-level decision.” This is probably as close to an on-the-record statement that Qualcomm chose TSMC over GlobalFoundries for its high-end Snapdragon parts because its 28nm processes are superior.
Finally, it’s time for Talluri to move onto his next interview. From what I can gather, for most execs, MWC basically consists of interview after interview. As we stand up, Talluri tells me that I had “some good questions,” and the PR rep congratulates me for being “persistent.” As I walk out of the room, I wonder if good questions and persistence will be currency when it comes to getting my hands on an early sample of Krait’s successor.
After talking to Raj Talluri, and then later to a handful of other Qualcomm execs, I got the overwhelming feeling that the company knows its strengths, but that it really wasn’t expecting to be rapidly thrust into the limelight with the explosion of the smartphone and tablet markets. The company knows it’s onto something good – but it has a little trepidation about whether it can successfully bring all of its strengths to the marketplace and repeat the success of Snapdragon. Talluri seemed (quietly) confident about its continued success in the mobile market – and you know what? So am I.