Image for post
Image for post
Much cooler: heat map of a Samsung Galaxy Book S with a Snapdragon processor (left) vs. a Dell XPS 13 with an x86 processor

Qualcomm Lays Out Snapdragon Roadmap in Balmy Maui

At its annual Snapdragon Summit in Maui last week, Qualcomm showed off mobile processor silicon advances on a number of fronts. To the relief of attending analysts, the schedule was not over-packed, and those of us from the frigid North had time to savor the fine weather.

Like other chipmakers, Qualcomm is diversifying its portfolio, tweaking variants of primary designs with custom spins optimized for specific applications. First out of the gate at the summit was the Snapdragon 865, referred to affectionately as “The Beast” by Alex Katouzian, SVP and GM of mobile products at Qualcomm. Katouzian touted the “desktop-level features” of the new part, citing its 15-trillion-operations-per-second processing power and 2-gigapixel-per-second camera, which can deliver 8K video at 30 frames per second. The 865 is designed for a world of artificial intelligence (AI) — with a range of scalar, vector, and matrix computation abilities — and 5G communications — which is handled by the X55 5G modem. For now, the X55 is a separate package, but future versions are likely to include the two on a single system-on-a-chip (SoC). The 865 will be executed in the 7nm silicon process node.

Cristiano Amon, Qualcomm’s president, projected that, although the company expects to see continuing health in the smartphone market for Snapdragon, new growth will come from diversification into areas like automotive (infotainment as well as digital cockpit), personal computers (PCs), and the industrial Internet of Things (iIoT — with retail, manufacturing, and smart cities representing the top three “things”).

Spinning its flagship part in another direction, the company announced the Snapdragon XR2, a 5G and AI-enabled platform dedicated to the various X’s of reality technology (augmented — AR, virtual — VR, and MR — mixed). The part is customized to reduce latency, that all-important measure of XR. Hiren Bhinde, Qualcomm’s XR product lead, spoke of mixed reality as allowing users to “switch between immersion levels,” a sort of spectrum of augmentation, layers of overlay, spreading from clear glass all the way to a fully computer-generated world.

Then, the company announced three new processors optimized for PCs — the Snapdragon 7c, 8c, and 8cx — for entry, mainstream, and premium notebooks. As a former VP of PCs at IDC, I find this category intriguing. That Microsoft should feel free enough to partner with Qualcomm, despite its long and nearly exclusive arrangement with Intel, shows just how far we’ve come in a couple of decades. The all-day-battery-life, always-connected PC has been a no brainer since its conception years ago, but the tale has been difficult in the telling.

Early versions were underwhelming, which damaged the category’s branding, and volume remains small. At this point, however, not only is the silicon up to premium-performance snuff, but, perhaps more importantly, Microsoft is supporting Windows 10 on ARM as a full citizen, and if the platform is identical from the waist up, then OEMs and application developers are indifferent to the hardware. As Miguel Nunes, Qualcomm’s senior director of Windows products laid it out, the 8c, and 8cx will be produced in the advanced 7nm process node while the 7c will be done in 8nm. A cornerstone of the Snapdragon c series will be 5G connectivity. Although on this round, the 5G modem is not on the Snapdragon SoC, the c series is designed for 5G performance. Advanced features in include text recognition, voice activation, on-device security, fingerprint recognition, contextual awareness, and face detection. Business features, available via Win10 enterprise with its hypervisor, include secured core, biometric authentication, cellular security, and geofencing with location awareness. OEM partners ready to roll on systems include Lenovo, Samsung, Huawei, Asus, HP, Microsoft, and Weibu.

With a PC that’s always on, always connected, thin, cool, and quiet and has full Win10, premium performance, and days of battery life, what’s not to love? Qualcomm aspires to nothing less than redefining the category, and having Microsoft on board is key to that goal. Microsoft sent an emissary, Nicole Dezen, VP of device partner sales, to talk up the relationship. She said, “Customers love all-day battery life and instant on” and described the ARM-based PC market as “growing.” She waxed enthusiastic about 5G connectivity, describing how it will make the time it takes to download a typical multi-gigabyte video drop from six minutes to three seconds.

Dezen also pointed out how 5G will help Microsoft promote its holy grail, which is to convert one-time sales (the old PC license model) to service relationships (e.g., the new O365 and Enterprise subscriptions). With 5G, real computing can be done in the edge cloud, and desktops can be thick or thin, local or virtual, depending on need and desire. Existing Microsoft consumer cloud services like OneDrive and Project xCloud can be more tightly coupled with the client side.

Dezen was not the only one to talk about splitting computation between the device and cloud with 5G. Qualcomm executives referred to that bit of the cloud right near the endpoint device as the “edge cloud,” but it could just as well be called the “cloud edge,” which actually makes more sense, but the jargon is still being sorted out.

Be that as it may, this split-computing model will promote applications that are more or less agnostic to the operating system (e.g., Windows, Mac, Android). With 5G providing a fast, reliable connection to cloud, where media and program modules can reside, the endpoint has lower local storage and processing requirements. Since they are not bound by local processing and memory capabilities, applications can evolve, and the OS will become less relevant. Given the tremendous data rates of 5G, the cost per bit will decrease, and carriers will be increasingly likely to offer unlimited data plans. The new computing method will be: do everything you can on the client (subject to form factor, price, power, performance constraints), and then go to the cloud, first the edge and then farther in toward the core.

A couple of end notes on China:

When I asked Qualcomm executives about whether China’s desire to control its entire computing stack would affect Qualcomm’s business there, Amon, the company president, made a distinction between the Chinese domestic market and Chinese firms’ activity on the world market. He pointed out that the international mobile ecosystem is spread out among many suppliers and countries, and that although Chinese vendors may succeed in producing 100% made-in-China devices for the local market, its largest vendors — firms like Huawei, OnePlus, Oppo, and Xiaomi — want to sell overseas and therefore need the support of international vendors. For example, Amon noted that, outside China, virtually all non-Apple smartphones run on Google’s Android operating system. “Qualcomm is essential for [the Chinese vendors’ international] growth ambitions,” he said. “I see that business getting stronger.”

In a spectacular demonstration of AI as applied to language translation, Ziad Asghar, Qualcomm’s VP of artificial intelligence and strategy, showed how an English speaker can have their words translated into Chinese in near real-time. The multi-step process is down to time measured in hundreds of milliseconds, but by early next year, it will be fast enough to use in phone calls between English and Chinese speakers, each speaking and hearing their own language. Here’s how it goes: speech to text is easy using existing mature automatic speech recognition (ASR) modules. The translation from English text to Chinese text is handled by more recently established transformer networks that take into account a word’s position in a sentence, its context, and other cues. Transformer networks are large, but not so large that they can’t be run on the local device. Speech output is relatively straightforward, but, on its own, leads to a rather wooden result. One last step applies something called “audio style transfer,” a neural network-based module that takes the speaker’s style and lays it over the output, leading to natural sounding speech that has characteristics of the original speaker’s voice and intonation. Asghar said to expect a demonstration of a real-time two-way conversation at Mobile World Congress in February 2020.

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store