Author: Bob O'Donnell

One of the more interesting and challenging markets that the tech industry continues to focus on is the highly touted Internet of Things, or IoT. The appeal of the market is obvious—at least in theory: the potential for billions, if not trillions, of connected devices. Despite that seemingly incredible opportunity, the reality has been much tougher. While there’s no question that we’ve seen tremendous growth in pockets of the IoT market, it’s fair to say that IoT overall hasn’t lived up to its initial hype.

A big part of the problem is that IoT is not one market. In fact, it’s not even just a few markets. As time has gone on, people are realizing that it’s hundreds of thousands of different markets, many of which only amount to unit shipments measured in thousands or tens of thousands.

In order to succeed in IoT, therefore, you need the ability to customize on a massive scale. Few companies understand this better than Arm, the silicon IP (intellectual property) and software provider whose designs sit at the heart of an enormous percentage of the chips powering IoT devices. The company has a huge range of designs, from its high-end performance A Series through its mid-range and real-time focused R series, down to its ultra-low power M series, that are used by its hundreds of chip partners to build silicon parts that power an enormous range of different IoT applications.

Even with that diversity, however, it’s becoming clear that more levels of customization are necessary to meet the increasingly specialized needs of the millions of different IoT products. To better address some of those needs, Arm made some important, but easy to overlook, announcements at its annual Arm TechCon developer conference in San Jose this week.

First, and most importantly, Arm announced a new capability to introduce Custom Instructions into its Cortex-M33 and all future Armv8-M series processors at no additional cost, starting in 2020. One of the things that chip and product designers have recognized is that co-processors and other specialized types of silicon, such as AI accelerators, are starting to play an important role in IoT devices. The specialized computing needs that many IoT applications demand are placing strains on the performance and/or power requirements of standard CPUs. As a result, many are choosing to add secondary chips to their designs upon which they can offload specialized tasks. The result is generally higher performance and lower power, but with additional costs and complexities. Most IoT devices are relatively simple, however, and a full co-processor is overkill. Instead, many of these devices require only a few specialized capabilities—such as listening for wake words on voice-based devices—that could be handled by a few instructions. Recognizing that need, Arm’s Custom Instructions addition allows chip and device designers to get the customized benefits of a co-processor built into the main CPU, thereby avoiding the costs and complexities they normally add.

As expected, Arm is providing a software tool that makes the process of creating and embedding custom instructions into chip designs a more straightforward process for those companies who have in-house teams with those skill sets. Not all companies do, however, so Arm will also be offering a library of prebuilt custom instructions, including AI and ML-focused ones, that companies can use to modify their silicon designs.

What’s particularly clever about the new Custom Instructions implementation—and what allowed it to be brought to market so quickly and with no impact to existing software and chip development tools—is that the Custom Instructions go into an existing partition in the CPU’s design. Specifically, they’re replacing instructions that were used to manage a co-processor. However, because the custom instructions essentially allow Arm’s chip design partners to build a mini co-processor onto the Arm core itself, in most situations, there’s no loss in functionality or capability whatsoever.

Of course, there’s more to any device than hardware, and Arm’s core software IoT announcement at TechCon also highlights its desire to offer more customization opportunities for IoT devices. Specifically, Arm announced that it was further opening up the development of its Mbed OS for IoT devices by allowing core partners to help drive its direction. The new MBed OS governance program, which already includes participation from Arm customers such as Samsung, NXP, Cypress and more, will allow more direct involvement of these silicon makers into the future evolution of the OS. This allows them to do things like focus on more low-power battery optimizations for certain types of devices that specific chip vendors need to better differentiate their product offerings.

There’s little doubt that the IoT market will eventually be an enormous one, but there’s also no doubt that the path to reach that size is a lot longer and more complicated than many first imagined. Mass customization of the primary computing components and the software powering these devices is clearly an important step toward those large numbers, and Arm’s IoT announcements from TechCon are an important step in that direction. The road to success won’t be an easy one but having the right tools to succeed on many different paths should clearly help.

Few technologies are expected to have as big an impact as 5G—the next generation wireless broadband connection standard—and now that we’ve finally started seeing the first 5G-enabled devices and real-world deployments, it’s worth taking a look at where things currently stand and how they’re likely to evolve.

Fortunately, I’m now in a much stronger position to do that as the result of two different 5G-focused events I attended last week. Qualcomm’s 5G Workshop at their headquarters in San Diego emphasized core 5G technologies and the work that the company has done to evolve and integrate those technologies into semiconductor-based components. Specifically, they highlighted their work on 5G modems, RF (radio frequency) transceivers, RF front ends for fine tuning the raw radio signals, and systems that integrate all three components into complete solutions. The 5G Americas analyst event in Dallas (TX) provided the telco carriers’ and network infrastructure equipment companies’ angle on the status of today’s 5G networks throughout the US and Latin America. It also included a session with FCC commissioner Michael O’Reilly that dove into the hot topic of radio frequency spectrum issues in the US.

The two events dovetailed nicely and offered an interesting and comprehensive perspective on today’s 5G realities. What became clear is that although 5G will eventually enable a whole wealth of new applications and possibilities, for the near-term, it’s primarily focused on faster cellular networks for smartphones, or what the industry likes to call enhanced mobile broadband (eMBB). Within that world of faster 5G cellular networks, there are two very important and widely recognized sub-groups that are divided by the different radio frequencies within which they each operate: millimeter wave frequencies (typically 24 GHz and higher—so named because their wavelengths are measured in single millimeters) and those collectively referred to as sub-6, shorthand for frequencies below 6 GHz. (As a point of reference, all current 4G LTE radio transmissions are done at frequencies below 3 GHz.)

Though it might seem a bit arcane, the distinction between frequencies is an extremely important one, because it has a huge impact on both the type of 5G services that will be available and the equipment necessary to enable them. Basically, millimeter wave offers very fast performance, but only over short distances and within certain environments. Conversely, sub-6 frequencies allow wider coverage, but at slower speeds. To make either of them work, you need network equipment that can transmit those frequencies and devices that are tuned to receive those frequencies. While that seems straightforward, the devil is in the details, and there are a wide variety of factors that can impact the ability for these devices and services to function properly and effectively. For example, just because a given smartphone supports some millimeter wave frequencies doesn’t mean it will work with the millimeter wave frequencies used by a given carrier—and the same is true for sub-6 GHz support. Bottom line? There’s a lot more for people to learn about the different types of 5G than has ever been the case with other wireless network generation transitions.

Just to complicate things a bit more, one of the more interesting take-aways from the two events is that there’s actually a third-group of frequencies that’s becoming a critical factor for 5G deployments in many countries around the world—but is still waiting to be deployed in the US. The C-Band frequencies (in telecommunications parlance, typically the frequencies from 3.5 GHz to 4.2 GHz), though technically part of the sub-6 group, offer what many consider to be a very useful compromise of both better performance and wider coverage than other frequency options above and below that range. In the US, the problem is that this set of frequencies (which happen to measure in the single centimeter range, though that does not seem to be the origin of the C-band name) is not currently available for use by telecom carriers. Right now, they’re being used for applications in defense and private industry, but as part of its spectrum modernization and evolution process, the FCC is expected to open up the frequencies and auction them off to interested parties like the major telcos in 2020.

Another interesting insight from the two events is that there are some important differences between the theory of what a technology can do and the reality of how it gets deployed. In the case of millimeter wave, for example, Qualcomm showed some impressive (though admittedly indoor) demos of how the technology can be used in more than just line-of-sight applications. The initial concern around this technology was that you needed to have a direct view from where you were standing with a smartphone to a millimeter wave-equipped cell tower in order to get the fastest possible speeds. With the Qualcomm demo, however, people were able to walk behind walls, and even into conference rooms, and still maintain the download speed benefits of millimeter wave, thanks to reflections off walls and glass. When asked why early real-world tests of 5G devices didn’t reflect this, the company essentially said that the early networks weren’t as dense as they needed to be and weren’t configured as well as they could be. Carrier representatives and network equipment makers at 5G Americas, however, countered that the reality of outdoor interference from existing 4G networks and the strength of those signals meant that—at least for the near term—real-world millimeter wave performance will be limited to line-of-sight situations.

An interesting takeaway from the demo, and the subsequent conversations, is that millimeter wave-based 5G access points could prove to be a very effective alternative to WiFi in certain indoor environments. Faster download speeds and wider coverage mean that fewer 5G small cells would have to be deployed than WiFi access points in a given building, potentially leading to lower management costs. Plus, technologies have been developed to create private 5G networks. As a result, I think millimeter wave-based 5G indoor private networks could prove to be one of the sleeper hits of this new technology.

Even though most of the current 5G efforts are focused on speeding up cellular connections, it became clear at both events that there are, in fact, a lot of interesting applications still to come in industrial environments, automotive applications, and more. Another important point that was emphasized at both events is that the initial launch of 5G is not the end of the technological development process, but just the beginning. As with previous cellular network generations, the advancements related to 5G will come in chunks roughly every 18-24 months. These developments are driven by the 3GPP (3rd Generation Partnership Project—a worldwide telecom industry standards group originally formed to create standards for 3G) and their release documents. The initial 5G launch was based on Release 15. However, Release 16 is expected to be finalized in January of 2020, and that will enable a whole other set of capabilities, all under the 5G umbrella. In addition, a great deal of work has already been done to start defining the specifications for Release 17, which is expected sometime in 2022.

The bottom line is that we’re still in the very early stages of what’s expected to be a decade-long evolution of technology and applications associated with 5G. Initial efforts are focused on faster speeds, and despite some early hiccups, excellent progress is being made with much more to come in 2020. The two conferences made it clear that the technologies underpinning next generation wireless networks are very complicated ones, but they’re also very powerful ones that, before we know it, are likely to bring us exciting new applications that we’ve yet to even imagine.

Though the leaves may not have started changing color, there’s another sure sign that we’ve entered fall: the barrage of smartphone and other personal device announcements from major manufacturers around the world. Technically, it started in early August at Samsung’s Unpacked event in New York, where they unveiled their Note 10 line of smartphones. The bulk of the announcements, however, are happening in September, most notably Apple’s iPhone 11 line. Looking ahead, the announcements should extend at least until October, given Google’s own pre-announcement of the Pixel 4.

The most recent phone announcement isn’t actually a new one—it’s the relaunch of the Samsung Galaxy Fold with a hardened, re-engineered design. The original Galaxy Fold never shipped to the public because of a number of serious issues with the foldable display that popped up with early reviews of the first units. Though it was clearly a PR disaster for the company, to their credit, they made the difficult decision to delay the product, make the necessary changes, and are now re-releasing it.

I was fortunate enough to receive a review unit of the first edition and, as a long-time fan of the concept of foldable displays, was pleased to discover that in real-world usage, working with a smartphone-sized foldable device truly is a game-changing experience. I also had absolutely zero problems with the unit I received, so was very disappointed to have to return it. Happily, I now have the revised version of the Fold and while it’s obviously too early to say anything about long-term durability, it’s clear that the new Fold design is better conceived and feels more rugged than the original, particularly the redesigned hinge.

Samsung has been very careful this time around to warn people to be cautious with the device and frankly, the early problems with the first generation will probably serve as a good warning to potential customers that they need to treat the Fold a bit more gingerly than they do a typical smartphone. Now, we can certainly argue whether a nearly $2,000 smartphone ought to be this delicate, but the re-release of the Fold says a number of things about the state of foldable technology in general.

First, the plastic material currently used to make foldable displays is still not anywhere close to the level of scratch resistance that glass is. Companies like Corning and other display component manufacturers are working to develop more hardened foldable displays, but if you’re eager to embrace the future now with a foldable device, current material science is going to limit devices to softer, more sensitive screens. An important implication of this is that Samsung made the correct decision in choosing to go with a fold-in design on the Galaxy Fold. Fold-out designs like the Huawei Mate X and the Royole FlexPai aren’t likely to survive more than a few months of regular usage. (Unfortunately for Huawei, that’s the least of their concerns as the lack of Google Services on any of their new devices—including Mate 30 and Mate X—is going to severely handicap their opportunities outside of China.)

Second, we need to think differently about the inevitable tradeoffs between functionality and ruggedness on these new devices. While even the revised design might not be able withstand running an X-Acto blade across the screen or dropping sand into it—though let’s be honest, who’s going to do that to a nearly $2,000 smartphone—as long as the devices prove to be functional over an extended period of regular usage, that will keep most all potential customers happy. The key point to remember is that people who want a radical, cutting-edge device like Fold are interested in it because of the unique experiences it can enable. Having started using it again, I’m still excited at how incredibly useful it is and how innovative it feels to open the device and start using a tablet-sized screen on a phone-sized device. Simple perhaps, but still very cool. In fact, given all the challenges that the initial device faced, it’s pretty amazing that so many people are still interested in the new Galaxy Fold. Clearly, the lure of foldability is still quite strong.

Plus, Samsung themselves has acknowledged the potential challenges the device faces and added two additional services to ward off concerns people may have. First, they’re providing a special concierge level service for Galaxy Fold owners that gives them access to a set of dedicated support personnel who can walk people through any types of questions they have with the phone—a nice touch for an expensive device. Second, the company is offering to replace any potentially damaged screens for $149 for the first year of ownership. While that’s not cheap, it’s certainly appears to be a lot less expensive than what it will cost Samsung to have to perform that repair.

Finally, I believe the official relaunch of the Fold will mark the beginning of a wide range of commercially available products with foldable displays and start to get people thinking about the creative new form factors that these screens enable. Lenovo, for example, has previewed their ThinkPad foldable PC, which is expected to ship around this time next year—showing that foldable screens won’t just be limited to phone-size devices.

There’s no question that the Galaxy Fold is not yet a mainstream device, but it’s equally clear to me that people who want cutting edge device experiences will be drawn to it. I, for one, am eager to continue my explorations.

If you had to name two things that best exemplify today’s modern office environments, you’d be hard pressed to come up with better choices than huddle rooms and videoconferences. Together, the two reflect both the different types of physical environments that many people now find themselves working in, as well as the different means by which they communicate with co-workers, customers, and other potential business partners.

Huddle rooms, for those who may not know, are the small meeting rooms or mini conference rooms that many companies have adopted—particularly those organizations with open floor plans—to provide small groups of people (typically 2-6) with an easy, space efficient means for holding meetings. Videoconferences are nothing new, of course, but their frequency has increased dramatically over the last few years, thanks to a combination of ubiquitous camera-equipped notebooks and smartphones, higher quality wireless networks, a younger workforce, and more emphasis on collaborative efforts within and across companies. Another big factor is the wider variety and broader usage of collaboration-focused software tools, ranging from modern chat platforms like Slack and Microsoft’s Teams, to integrated collaboration features in Google’s GSuite, to an enormous range of videoconferencing applications, including Zoom, Blue Jeans, GoToMeeting, Webex, Ring Central, Skype, FaceTime, and much more.

Arguably, huddle rooms and videoconferences are each separately having an important impact on how people work today, but when you put the two together—as organizations have started to do—that’s when you really start to understand how work environments of the late 2010s are very different than they were even earlier in the decade. In recognition of the fact, many companies are starting to set up a number of videoconferencing equipped huddle rooms to drive more collaborative efforts—as well as free employees from the noise-filled cacophony that many open office environments quickly turn into. Not surprisingly, a number of vendors are working to create solutions to address those needs.

From PC companies like Lenovo and HP, to more traditional videoconferencing companies like Polycom (now part of a merged company with Plantronics called Poly), there are quite a few interesting new approaches to creating hardware tools that can optimize collaboration and tap into the many software-based videoconferencing tools and services now available. In fact, arguably, one of the key reasons why Plantronics spent $2B to acquire Polycom last year to create Poly is because of the growing importance of video-based collaboration in work environments.

Looking specifically at huddle room-focused video collaboration tools, one of the more intriguing new options coming from the blended company is the Poly Studio, a $949 USB C-equipped soundbar and video camera system that integrates much of the intelligence of higher-end dedicated videoconferencing systems into a huddle room-friendly portable form factor. Anyone using the huddle room can simply plug the device into a USB C-equipped notebook PC and get access to a high-quality 4K-capable audio video system that works automatically with most popular videoconferencing tools—including the aforementioned Teams, Zoom, Blue Jeans, GoToMeeting, and more.

Unlike standalone webcams, the Poly Studio has the ability to automatically track whoever is speaking in a room, both through automatically focusing the camera and directing the microphones to pick up and prioritize the audio coming from the speaker. On top of that, some clever audio processing technology can create what Poly calls an Acoustic Fence that keeps voices outside the room (or walking past) from interrupting the discussion. The NoiseBlock feature will analyze and then automatically mute other sounds that may be coming from within the room or from other call participants. For those who prefer to only use the audio for a given session, there’s a slider available to physically block the lens. A key benefit for IT departments is that Poly Studios can be centrally managed and remotely updated and configured.

Though they may be small, huddle rooms are proving to be incredibly important resources for employees who want to be more productive in their workplace. Particularly in companies that chose to go with open office layouts (and that are likely regretting the decision now), huddle rooms can provide an oasis of calm that enables the kind of increased collaboration that open offices were supposed to offer. At the same time, it’s clear that collaboration of all types, but particularly video-enabled calls, is going to be increasingly important (and common) in businesses of all sizes and varieties. As a result, tools that can bring together real-time collaboration in small rooms are going to play a critical role in increased/improved communication and productivity moving forward.

Let’s put it this way—it’s a lot to take in over the course of 90 minutes. But given the extensive reach that server virtualization pioneer VMware’s products now have, once all the information starts to sink in, the company’s long-term strategy does make sense—even if it does fall a bit into the buzzword bingo trap.

Over the course of the opening day keynote at their VMWorld conference, the company managed to discuss the future of enterprise computing (it’s a hybrid cloud world where Kubernetes and containers are at the heart of corporate software transformation), driven in part by virtualized GPUs running AI/machine learning workloads, delivered to a wider range of devices (including enterprise-hardened Chromebooks from Dell), all secured from a zero trust model that’s integrated across devices, virtual machines, containers and more. Toss in a multi-faceted, but logical splitting of edge computing models that can leverage 5G connectivity and, well, there’s pretty much not many topics that VMware didn’t talk about.

Despite the overwhelming scope of the discussion, however, there was a fairly clear sense that the future of computing was on display—or at least, a vision for where that future could go. Not surprisingly, that future is extraordinarily software dependent. In fact, a quick summary of everything they said might essentially be, we’re working to virtualize everything from computing infrastructure to applications to networking and beyond into a set of software-defined services, and we’ll let you build, run and manage those services on any combination of hardware and in any set of locations that you’d like. Again, a lot to take in, but certainly a compelling idea and vision for the future.

On the enterprise application and cloud computing side of the world, the company debuted Tanzu, a set of products and services designed to make Kubernetes-focused containers the default type of structure for virtually all applications, including modern cloud-native ones and even older legacy apps. Specifically, Tanzu Mission Control will provide a centralized place to manage both containers and virtual machines in a single place, even when they’re spread across multiple cloud environments, local data centers, co-location facilities and more. As companies start to transition towards more cloud-native programs, they’re running into challenges in trying to manage all these new applications alongside existing applications running in virtual machines. Mission Control is designed to help with that process.

In addition, the company debuted Project Pacific, a technology preview of a capability being built into the next version of their vSphere virtual machine platform. Project Pacific embeds Kubernetes capabilities into vSphere and allows virtual machines to be converted into and run as containers without having to rewrite, or refactor, the underlying application. The idea is to give companies tools to turn all their applications into containers and then provide a unified management console that makes the process of running and managing these applications much easier.

Within the applications themselves, VMware also announced a partnership with Nvidia that lets server-based GPUs be virtualized in order to run AI and machine learning workloads more efficiently. CPU virtualization has been at the heart of VMware’s offerings since the company’s debut, but the move to virtualize GPUs for AI is a recent phenomenon and a logical extension to both VMware and Nvidia’s business. The companies also specifically partnered with Amazon Web Services (AWS) to offer an Nvidia T4-accelerated EC2 bare metal computing instance, allowing companies to move GPU-dependent workloads from their private data centers to the cloud with AWS.

From a client perspective, VMware also made several interesting announcements, including some additions to their Workspace One digital workspace offering, as well as the debut of extended management services for Chromebooks. One of the highlights of Workspace One is a new IBM Watson-powered voice assistant that’s designed to provide intelligent, contextual information to employees as they interact with their devices, corporate applications and company services. It’s an interesting choice not to go with Alexa or Google Assistant for this, but also highlights how both VMware and IBM likely want to take a different, more corporate-focused path for their assistant. For client hardware, VMware made a joint announcement with its partner Dell to add Unified Workspace management support for Dell’s new Latitude Chromebook Enterprise notebooks—emphasizing the increasing diversity of client devices that these new cloud-computing focused application delivery models enable. In addition, the two companies announced an extension of Workspace One that lets it monitor and verify the state of Dell SafeBIOS on supported Dell hardware. It’s a yet another small but interesting example of how the two companies are starting to leverage the connection between them to deliver a better together experience.

There’s no question that the vision VMware CEO Pat Gelsinger laid out in his opening keynote is a broad, bold view of a completely software-defined, virtualized and services-driven world. Realizing the vision is, of course, much harder than simply describing it, but it appears the company is taking a number of key steps in the right direction, particularly through an aggressive series of new acquisitions. Integrating all the new pieces into the vision puzzle won’t be easy, and the process of getting companies/customers to modernize their applications and infrastructure is likely to take much longer than VMware would like to see. Still, VMware has been building off the same core strategy of software defined capabilities for several years now and as they extend that strategy to integrate newer developments like containers and Kubernetes, they certainly appear to be on a compelling path forward.

For a long time, most of the biggest innovations in semiconductors happened in client devices. The surge in processing power for smartphones, following the advancements in low-power CPUs and GPUs for notebooks, enabled the mobile-led computing world in which we now find ourselves.

Recently, however, there’s been a marked shift to chip innovation for servers, reflecting both a more competitive marketplace and an explosion in new types of computing architectures designed to accelerate different types of workloads, particularly AI and machine learning. At this week’s Hot Chips conference, this intense server focus for the semiconductor industry was on display in a number of ways. From the debut of the world’s largest chip—the 1.2 trillion transistor 300mm wafer-sized AI accelerator from startup Cerebras Systems—to new developments in Arm’s Neoverse N1 server-focused designs, to the latest iteration of IBM’s Power CPU, to a keynote speech on server and high-performance compute innovation from AMD CEO Dr. Lisa Su, there was a multitude of innovations that highlighted the pace of change currently impacting the server market.

One of the biggest innovations that’s expected to impact the server market is the release of AMD’s line of second generation Epyc 7002 series server CPUs, which had been codenamed “Rome.” At the launch event for the line earlier this month, as well as at Hot Chips, AMD highlighted the impressive capabilities of the new chips, including many world record performance numbers on both single and dual-socket server platforms. The Epyc 7002 uses the company’s new Zen 2 microarchitecture and is the first server CPU built on a 7nm process technology and the first to leverage PCIe Gen 4 for connectivity. Like the company’s latest Ryzen line of desktop CPUs, the new Epyc series is based on a chiplet design, with up to 8 separate CPU chips (each of which can host up to 8 cores), surrounding a single I/O die and connected together via the company’s Infinity Fabric technology. It’s a modern chip structure with an overall architecture that’s expected to become the standard moving forward, as most companies start to move away from large monolithic designs to combinations of smaller dies built on multiple different process size nodes packaged together into an SOC (system on chip).

The move to a 7nm manufacturing process for the new Epyc line, in particular, is seen as being a key advantage for AMD, as it allows the company to offer up to 2x the density, 1.25x the frequency at the same power, or ½ the power requirements at the same performance level as its previous generation designs. Toss in 15% instruction per clock performance increases as the result of Zen 2 microarchitecture changes and the end result is an impressive line of new CPUs that promise to bring much needed compute performance improvements to the cloud and many other enterprise-level workloads.

Equally important, the new Epyc line positions AMD more competitively against Intel in the server market than they have been for over 20 years. After decades of 95+% market share in servers, Intel is finally facing some serious competition and that, in turn, has led to a very dynamic market for server and high-performance computing—all of which promises to benefit companies and users of all types. It’s a classic example of the benefits of a competitive market.

The prospect of the competitive threat has also led Intel to make some important additions to its portfolio of computing architectures. For the last year or so, in particular, Intel has been talking about the capabilities of its Nervana acquisition and at Hot Chips, the company started talking in more detail about its forthcoming Nervana technology-powered Spring Crest line of AI accelerator cards, including the NNP-T and the NNP-I. In particular, the Intel Nervana NNP-T (Neural Networking Processor for Training) card features both a dedicated Nervana chip with 24 tensor cores, as well as an Intel Xeon Scalable CPU, and 32GB of HBM (High Bandwidth Memory). Interestingly, the onboard CPU is being leveraged to handle several functions, including managing the communications across the different elements on the card itself.

As part of its development process, Nervana determined that a number of the key challenges in training models for deep learning center on the need to have extremely fast access to large amounts of training data. As a result, the design of their chip focuses equally on compute (the matrix multiplication and other methods commonly used in AI training), memory (four banks of 8 GB HBM), and communications (both shuttling data across the chip and from chip-to-chip across multi-card implementations). On the software side, Intel initially announced native support for the cards with Google’s TensorFlow and Baidu’s PaddlePaddle AI frameworks but said more will come later this year.

AI accelerators, in general, are expected to be an extremely active area of development for the semiconductor business over the next several years, with much of the early focus directed towards server applications. At Hot Chips, for example, several other companies including Nvidia, Xilinx and Huawei also talked about work they were doing in the area of server-based AI accelerators.

Because much of what they do is hidden behind the walls of enterprise data centers and large cloud providers, server-focused chip advancements are generally little known and not well understood. But the kinds of advancements now happening in this area do impact all of us in many ways that we don’t always recognize. Ultimately, the payoff for the work many of these companies are doing will show up in faster, more compelling cloud computing experiences across a number of different applications in the months and years to come.

My, how things have changed. Not that many years ago Microsoft was trying to compete in the smartphone market and PCs were considered on their way out. Samsung was a strong consumer brand but had virtually no presence (or credibility) in the enterprise market.

Fast forward to today—well, last week’s Galaxy Unpacked event to be precise—and the picture is totally different. Microsoft CEO Satya Nadella joined Samsung Mobile CEO DJ Koh onstage to announce a strategic partnership between the companies that highlights what could (and should) prove to be a very important development, not only for the two organizations, but for the tech industry in general.

The partnership also signifies a profound shift in the landscape of devices, companies, platforms, and capabilities. By bringing together Samsung branded hardware and Microsoft software and services, the two companies have formed a powerful juggernaut that represents a serious threat to Apple and an oblique challenge to Google, particularly in the enterprise market.

To be clear, the companies have worked together in the past and some have argued that the exact details of the partnership remain to be flushed out. Fair enough. But when you put together the number one smartphone market share presence of Samsung with the deeply entrenched position of Microsoft software and cloud-based services, it’s not hard to imagine a lot of very interesting possibilities that could grow out of the new arrangement.

For one, it helps each company overcome long-running concerns that they’ve been missing out on important markets. Samsung has been chided for not having the software and services expertise and offerings of an Apple, which was theoretically going to make the Korean giant vulnerable as the hardware markets started to slow. On the other side, Microsoft rather notoriously failed to make any kind of dent in the smartphone market. Together, however, the complementary capabilities offered by the partnership give customers a wide range of powerful and attractive devices, along with leading edge services and software in the business world. Plus, the two companies don’t really compete, making the collaboration that much more compelling. The consumer story is clearly much tougher, but even there, Microsoft’s forthcoming game streaming services could certainly be an intriguing and compelling option for certain consumers. On top of that, the combination of Samsung and Microsoft is likely to attract interest from other third-party consumer services (Spotify or Netflix anyone?) that would be interested in joining the party.

But there are additional benefits to the partnership as well. For one, it clearly helps tie PCs and smartphones together in a much more capable and blended way. To Apple’s credit, their Continuity features that link iPhones, iPads and Macs in an organized fashion were the first to make multiple devices operate as a single entity. However, despite Apple’s overall strength, the percentage of people who only own Apple devices is actually pretty small—in the single digit percentage range according to my research. The percentage of people who have Windows PCs and Android-based phones, on the other hand, is enormous. Obviously, Samsung only owns a portion of that total, but it’s a big enough percentage to make for a very significant competitor.

More importantly, the combination of Microsoft and Samsung also further breaks down the walls between different operating systems and highlights the value the cloud can bring to a multi-device, multi-platform world. Samsung is still committed to Google’s Android as its smartphone OS, but by integrating things like Office 365, OneDrive and more into its devices, they are making life easier for people who spend much of their time in the Windows world. Conversely, Microsoft’s expanding efforts with their Your Phone app in Windows 10 highlight the effort they’re making to turn the process of using multiple devices into a more coherent experience. Unique Samsung-specific extensions promised for that app should make it even more compelling. For Google, the challenge will be continuing to build the presence of apps like G Suite and other enterprise-focused services in spite of its leading Android partner choosing Microsoft for some of its business-based offerings.

The deal extends beyond smartphones as well. Though Samsung has been tiny player in the Windows PC market for several years, they made a bit of a splash at the event by introducing the Samsung Galaxy Book, the first Windows-based always connected PC (ACPC) using Qualcomm’s third generation PC-focused processor the 8CX. While there are clearly still challenges in that space, the fact that it’s a Samsung, Microsoft, Qualcomm partnership in PCs exemplifies again how far the tech industry has evolved over the last several years.

We’re clearly still in the very early days of analyzing what the potential impact of the Samsung and Microsoft partnership will mean. But even a casual glance would suggest that there are very interesting things still to come.

The US telco industry has seen its share of upheavals and evolutions over the last few years, but one of the biggest potential changes got kickstarted late last week when the US Dept. of Justice finally gave the green light to the long-awaited proposed $26.5B merger between T-Mobile and Sprint. Ironically, it took the introduction of Dish Network—a company best-known as a satellite TV provider, but one that has had its eye on being a more general-purpose service provider for some time now—to get the deal over the final hump of federal regulatory approval. (An antitrust lawsuit backed by several state attorneys general could still end up blocking the final merger, but the DoJ approval is widely seen as a strong argument for its completion.)

A tremendous amount of ink has already been spilt (or should I say, pixels rendered) discussing the whats and wherefores of the proposed merger, but in the end, it seems the most critical factor is 5G and what it will mean to the future of connectivity. Sure, there are arguments to be made about how our individual cellphone plan pricing may change or what services may or may not be offered, but those are all short-term issues. Strategically, it’s clear that the future of not just the mobile wireless industry, but connectivity in general, is increasingly tied to 5G.

In the near-term, of course, lots of people and companies are interested in building out 5G-capable networks, as well as devices that connect to them and services that can leverage them. That is indeed a huge task and something that’s going to take years to complete. Not surprisingly, some of the most compelling arguments for the merger—as well as for the new fourth 5G-capable network that Dish is now on the hook to complete—were around 5G-compatible spectrum, or frequency holdings, that each of the new entities would have if the deal was to go through.

Specifically, the new T-Mobile would gain a large chunk of Sprint’s mid-band, 2.5 GHz range frequencies (a subset of the larger group known as sub-6), which many have argued is an important middle ground for 5G. AT&T, Verizon and now T-Mobile have focused their early 5G efforts on millimeter wave frequencies (around 39 GHz for all three of them, although T-Mo also has some 28 GHz spectrum), which offers extremely fast speeds, but extremely short range, and essentially only works outside (or near an interior mounted, millimeter wave small cell access point). Late in the year, T-Mobile plans to add 600 MHz frequencies, which is on the bottom end of the sub-6 frequency range and offers significantly wider coverage—but at speeds that aren’t likely to be much faster (if even as fast) as some of fastest 4G LTE coverage now available. The Sprint frequencies will allow the “new” T-Mobile to also offer faster download speeds at 2.5 GHz, rounding out their 5G offering. (AT&T and Verizon have committed to bring sub-6 frequencies into their 5G offerings sometime in 2020.) Dish, the mobile carrier, for its part, will be able to leverage some existing spectrum it already owns in the 1.7-2.1 GHz range, as well as use some of the 800 MHz frequency that Sprint was forced to sell to Dish as part of the deal. All of it fits into the sub-6 category of spectrum, but the combination should allow Dish to create a 5G network with both good coverage and decent performance.

The one interesting twist on the mobile wireless side is that 5G heavily leverages existing 4G infrastructure investments, and in fact, 4G LTE service is getting better and faster as 5G is being deployed. As a result, the 5G buildout will, ironically, lengthen the usable lifetime of 4G LTE technology, as well as devices that use it—particularly those equipped with LTE Advanced capabilities and some of the spectrum sharing and compression technologies like 256 QAM, 4×4 MIMO (Multiple Input, Multiple Output), and carrier aggregation. Toss in technologies like Dynamic Spectrum Sharing (DSS), which in the world of 5G mobile infrastructure was pioneered by Ericsson and allows telcos with the appropriate equipment to share 4G and 5G spectrum, and the transition from 4G to 5G in mobile wireless should be very seamless (and almost invisible).

However, there’s more to 5G than mobile wireless, and that’s where things start to get really interesting. First, there are some very interesting options for building private 5G networks that companies could leverage across campus sites, or inside large manufacturing buildings, and essentially replace their WiFi network. While no one expects WiFi to completely go away, there are some very intriguing opportunities for network equipment makers and carriers to address this market because of the faster transfer speeds, higher levels of security, and the decrease in manageability costs that private 5G could provide versus WiFi.

There’s also the opportunity to replace broadband network connections and even supplement or replace WiFi in our homes as well. As in the business world, WiFi isn’t going to go away overnight in the consumer world (there are just too many WiFi devices that we already have in place), but it’s already possible to get 5G connections (heck, even some of the new 4G LTE Advanced networks—like AT&T’s confusingly labelled 5Ge) that are faster than a lot of home WiFi. Think of the potential convenience both at work and at home of not having to worry about two different types of wireless connections, but instead connecting everything through a single wireless broadband connection like 5G. In the future, it could be a very intriguing possibility.

Above and beyond the pure network “pipes” discussion, 5G also potentially enables a host of new services through technologies like network slicing. Essentially a form of virtualized or software-defined networks, network slicing will allow carriers to do things like provide a combination of different services to different companies or even individuals with a guaranteed quality of service, and much more. Innovative companies are likely to dream up interesting ways to package together existing services like streaming video and music, along with lots of other things that we haven’t even thought of yet to take advantage of the opportunities that network slicing could create.

The bottom line is that the transition to 5G opens up a world of interesting possibilities that go well beyond the level of competition for our current cellphone plans. In the short term, as we start to see the first real-world deployments and 5G-capable devices come to life, we’re bound to see some frustrations and challenges with the early implementations of the technology. Strategically and longer term, however, there’s no question that we’re on the cusp of an exciting new era. As a result, big changes, like the T-Mobile-Sprint merger and the launch of Dish as a new fourth US carrier, are likely only the beginning of some large, industry-shifting events that will be impacting not just the tech industry, but modern society, for some time to come.

For most major tech advancements, the more mature and better developed a technology gets, the easier it is to understand. Unfortunately, it seems the exact opposite is happening in the world of artificial intelligence, or AI. As machine learning, neural networks, hardware advancements, and software developments meant to drive AI forward all continue to evolve, the picture they’re painting is getting even more confusing.

At a basic level, it’s now much less clear as to what AI realistically can and cannot do, especially at the present moment. Yes, there’s a lot of great speculation about what AI-driven technologies will eventually be able to do, but there are several things that we were led to believe they could do now, which turn out to be a lot less “magical” than they first appear.

In the case of speech-based digital assistants, for example, there have been numerous stories written recently about how the perceived intelligence around personal assistants like Alexa and Google Assistant are really based more around things like prediction branches that have been human built after listening to thousands of hours of people’s personal recordings. In other words, people analyzed typical conversations, based on those recordings, determined the likely steps in the dialog, and then built sophisticated logic branches based on that analysis. While I can certainly appreciate that it represents some pretty respectable analysis and the type of percentage-based predictions that early iterations of machine learning are known to do, it’s a long way from any type of “intelligence” that actually understands what’s being said and responds appropriately. Plus, it clearly raises some serious questions about privacy that I believe have started to negatively impact the usage rates of some of these devices.

On top of that, recent research by IDC on real-world business applications of AI showed failure rates of up to 50% in some of the companies that have already deployed AI in their enterprises. While there are clearly a number of factors potentially at play, it’s not hard to see that some of the original promise of AI isn’t exactly living up to expectations.

Of course, a lot of this is due to the unmet expectations that are almost inevitably part of a technology that’s been hyped up to such an enormous degree. Early discussions around what AI could do implied a degree of sophistication and capability that was clearly beyond what was realistically possible at the time. However, there have been some very impressive implementations of AI that do seem to suggest a more general-purpose intelligence at work. The well-documented examples of systems like AlphaGo, which could beat even the best players in the world at the very sophisticated, multi-layer strategy necessary to win at the ancient Asia game called Go, for example, gave many the impression that AI advances had arrived in a legitimate way. In addition, just this week, Microsoft pledged $1Billion to a startup called OpenAI LP in an effort to work on creating better artificial general intelligence systems. That’s a strong statement about the perceived pace of advancements in these more general-purpose AI applications and not something that a company like Microsoft is going to take lightly.

The problem is, these seemingly contradictory forces, both against and for the more “magical” type of advances in artificial intelligence, leave many people—myself included—unclear as to what the current state of AI really is. Admittedly, I’m oversimplifying to a degree. There are an enormous range of AI-focused efforts and a huge number of variables that go into these efforts, so it’s not realistic to expect, much less find, a simple set of reasons for how or why some of the AI applications seem so successful and why some are so much less so (or, at the very least, a lot less “advanced” than they first appear). Still, it’s not easy to tell how successful many of the early AI efforts have been, nor how much skepticism we should apply to the promises being made.

Interestingly, the problem extends into the early hardware implementations of AI capabilities and the features they enable as well. For example, virtually all premium smartphones released over the last year or two have some level of dedicated AI silicon built into them for accelerating features like on-device face recognition, or other computational photography features that basically help make your pictures look better (such as adding bokeh effects from a single camera lens, etc.) The confusing part here is that the availability of these features is generally not dependent on whether your phone includes, for example, a Qualcomm Snapdragon 835 or later processor or Apple A11 or later series chip, but rather what version of Android or iOS you’re running. Phones that don’t have dedicated AI accelerators still offer the same functions (in the vast majority of cases) if they’re running newer versions of Android and iOS, but the tasks are handled by the CPU, GPU, or other component inside the phone’s SOC (system on chip). In theory, the tasks are handled slightly faster, slightly more power efficiently, or, in the case of images, with slightly better quality if you have dedicated AI acceleration hardware, but the differences are currently very small and, more importantly, subject to a great deal of variation based on software and software layer interactions. In other words, even phones without dedicated AI acceleration at the silicon level are still able to take advantage of these features.

This is due, primarily, to the extremely complicated layers of software necessary to write AI applications (or features). Not surprisingly, writing code for AI is very challenging for most people to do, so companies have developed several different types of software that abstract away from the hardware (that is, put more distance between the code that’s being written and the specific instructions executed by the silicon inside of devices). The most common layer for AI programmers to write is within what are called frameworks (e.g., TensorFlow, Caffe, Torch, Theano, etc.). Each of these frameworks provide different structures and sets of commands or functions to let you write the software you want to write. Frameworks, in turn, talk to operating systems and translate their commands for whatever hardware happens to be on the device. In theory, writing straight to the silicon (often called “the metal”) would be more efficient and wouldn’t lose any performance benefits in the various layers of translation that currently have to occur. However, very few people have the skills to write AI code straight to the metal. As a result, we currently have a complex development environment for AI applications, which makes it even harder to understand how advanced these applications really are.

Ultimately, there’s little doubt that AI is going to have an extremely profound influence on the way that we use virtually all of our current computing devices, as well as the even larger range of intelligent devices, from cars to home appliances and beyond, that are still to come. In the short term, however, it certainly seems that the advances we may have been expecting to appear soon, still have a way to go.

It’s tough enough when you have a business model that not a lot of people understand, but then when you make some adjustments to it, well, let’s just say it’s easy for people to potentially get confused.

Yet, that’s exactly the position that Arm could find themselves in today, as news of some additional offerings to their semiconductor IP licensing model are announced. But that needn’t be the case, because the changes are actually pretty straightforward and, more importantly, offer some interesting new opportunities for non-traditional tech companies to get involved in designing their own chips.

To start with, it’s important to understand the basic ideas behind what Arm does and what they offer. For over 28 years, the company has been in the business of designing chip architectures and then licensing those designs in the form of intellectual property (IP) to other companies (like Apple, Qualcomm, Samsung, etc.), who in turn take those designs as a basis for their own chips, which they then manufacture through their semiconductor manufacturing partners. So, Arm doesn’t make chips, nor are they a fabless semiconductor company that works with chip foundries like TSMC, Global Foundries, or Samsung Foundry to manufacture their own chips. Arm is actually two steps removed from the process.

In spite of that seemingly distant relationship to finished goods, however, Arm’s designs are incredibly influential. In fact, it’s generally accepted that over 95% of today’s smartphones are based on an Arm CPU design. On top of that, Arm-based CPUs have begun to make inroads in PCs (Qualcomm’s chips for Always Connected PCs, sometimes called Windows on Snapdragon, are based on Arm), servers, and even high-performance computing systems from companies like Cray (recently purchased by HP Enterprise). Plus, Arm designs more than just CPUs. They also have designs for GPUs, DSPs, Bluetooth/WiFi and other communications protocols, chip interconnect, security, and much more. All told, the company likes to point out that 100 billion chips based on its various designs shipped in the first 26 years of its existence, and the next 100 billion are expected to ship between 2017 and 2021.

Part of the reason they expect to be able to reach that number is the explosive growth predictions for smart connected devices—the Internet of Things (IoT)—and those devices’ need for some type of computing power. While many of the chips powering those devices will be designed and sold by their existing semiconductor company clients, Arm has also recognized that many of the chips are starting to be put together by companies that aren’t traditional tech vendors.

From manufacturers of home appliances and industrial machines, to medical device makers and beyond, there are a large number of companies that are new to smart devices and have begun to show interest in their own chip designs. While some of them will just leverage off-the-shelf chip designs from existing semi companies, many of them have very specific needs that can best be met—either technically, financially, or both—with a custom designed chip. Up until now, however, these companies have had to choose which pieces of Arm IP that they wanted to license before they created their own chip. Needless to say, that business model discouraged experimentation and didn’t provide these types of companies with the options they needed.

Hence the launch of Arm’s new Flexible Access licensing model, which will now let companies choose from a huge range (though not all) of Arm’s IP options, experiment with and model chip designs via Arm’s software tools, and then pay for whatever IP they end up using—all while receiving technical support from Arm. It’s clearly an easier model for companies that are new to SOC and chip design to make sense of, and it essentially provides a “chip IP as a service” type of business offering for those who are interested. However, Arm will still offer their traditional licensing methods for companies that want to continue working the way they have been. Also, Arm’s highest performing chip designs, such as their Cortex-A7x line of CPUs, will only be available to those who use the existing licensing methods, under the presumption that companies who want that level of computing power know exactly what they’re looking for and don’t need a Flexible Access type of approach.

For those who don’t follow the semiconductor market closely, the Arm chip IP business can certainly be confusing, but with this new option, they’re making a significant portion of their IP library available to a wider audience of potential customers. And that’s bound to drive the creation of some interesting new chip designs and products based on them.

Talk to anybody in the semiconductor industry these days and all they seem to want to talk about is chiplets, the latest development in SOC (system on chip) designs. The basic rationale behind chiplets is that several different developments are making the industry’s traditional method of building increasingly larger chips less appealing, both technically and financially. So, instead of designing sophisticated, monolithic chips that incorporate all the important elements on a single silicon die, major semiconductor companies are designing products that break the larger designs into smaller pieces (hence “chiplets”) and combine them in clever ways.

What makes chiplet design different from other SOC design methodologies that have existed for many years is that many of these new chiplet-based parts are putting together pieces that are made on different process technologies. So, for example, a chiplet design might link a 7 or 10 nm CPU with a 14 nm or 22nm I/O element over some type of high-speed internal interconnect.

The reason for making these kinds of changes gets to the very heart of some of the transformational developments now impacting the semiconductor business. First, as has been widely discussed, traditional Moore’s Law advancements in shrinking transistor size have slowed down tremendously, making it difficult (and very expensive) to move all the elements inside a monolithic chip design down to smaller process geometries. Plus, even more importantly, it turns out that some important elements in today’s chip designs, such as analog-based I/O and some memory technologies, actually perform worse (or simply the same, but at a significantly higher cost) in smaller-sized chips. Therefore, some semiconductor components are better off staying at larger process manufacturing sizes. In addition, the processing requirements for different types of workloads (such as AI acceleration) are expanding, leading to the need to combine even more types of processing technology onto a single component. Finally, there have been some important advancements in chip packaging and interconnect technologies that are making the process of building these multi-part chiplets more efficient.

Most large chip companies have recognized the importance of these trends and have been working on advancing their various chiplet-related technologies for the last several years. To that end, Intel just announced some important new additions to its arsenal of chip packaging capabilities at the Semicon West conference this week, all designed to enable even more sophisticated, more flexible, and better yielding chiplet-based products in the years to come. At past events, Intel has talked about its EMIB (Embedded Multi-die Interconnect Bridge) technology, which provides horizontal, or 2D, connections across different chiplet elements. They’ve also talked about Foveros, which is their 3D stacking technology for putting multiple elements in a chip design on top of each other. The latest development is a logical combination of the two, which they call Co-EMIB, that enables both 2D-horizontal and 3D-vertical connections of components in a single package.

In order to efficiently deliver power and data to these various components, Intel also developed a technology called ODI (Omni-Directional Interconnect), which works through and across chips to provide the power and low latency connections needed to perform closer to monolithic chip designs. Finally, the company also announced a new version of their AIB (Advanced Interface Bus) standard called MDIO that provides the physical layer connect for die-to-die connections used in EMIB.

Together, the new advances give Intel more flexibility and capability to build increasingly sophisticated chiplet-based products—the real fruits of which we should start to see later this year and for several years to come. In addition, these developments help to address some of the challenges that still face chiplets, and they should (hopefully) help to drive more interoperability across multiple vendors. For example, even though the interconnect speeds across chiplets are getting faster, they still don’t quite meet the performance that monolithic designs offer, which is why a technology like ODI is important.

In terms of interoperability, there have been some notable examples of chiplet designs that combine pieces from different vendors, notably the Kaby Lake G, which combines an Intel CPU core from Intel’s 14nm+ process with an AMD GPU built on Global Foundries 14 nm, along with HBM (High Bandwidth Memory). However, right now more vendors are focused on their own inter-chip connection technologies (NVLink for Nvidia, Infinity Fabric for AMD, etc.), although there have also been some industry-wide efforts, such as CCIX, Gen-Z and OpenCapi. Still, the industry is a very long way away from having a true chip-to-chip interconnect standard that would allow companies to use a Lego-like approach to piece together chiplets from whatever processor, accelerator, I/O, or memory elements they would like.

Practically speaking, Intel recognizes the need to drive open standards in this regard, and they have made their AIB (and now, MDIO) standards available to others in an effort to help drive this advancement. Whether or not it will have any real-world impact remains to be seen, but it is an important step in the right direction. Particularly in the world of AI-specific accelerators, many companies are working to create their own chip designs that, ideally, could dramatically benefit from being combined with other components from the larger semiconductor players into unique chiplet packages.

At Baidu’s Create AI developer conference in China last week, for example, Intel talked about working with Baidu on Intel’s own Nervana-based NNP-T neural network training processors. Baidu has also publicly talked about its own AI accelerator chip called Kunlun (first introduced at last year’s Create conference), and although nothing was said, a logical connection would be to have future (or more likely, custom) versions of the NNP-T boards that incorporate Kunlun processors in a chiplet-like design.

Though they represent a significant diversion from traditional semiconductor advances, it’s become abundantly clear that the future of the semiconductor industry is going to be driven by chiplets. From this week’s official launch of AMD’s 3rd generation Ryzen CPUs—which are based on chiplet design principles that interconnect multiple CPU cores—to future announcements from Intel, AMD, Nvidia and many others, there’s no question that the flexibility that chiplets enable is going to be critically important for advancements in semiconductors and computing overall. In fact, while there’s no doubt that improvements in process technologies and chip architectures will continue to be important, it’s equally true that advances in the previously arcane worlds of chip packaging and interconnect are going to be essential to the advancement of the semiconductor industry as well.

As a long-time PC industry observer, it’s been fascinating to watch the evolution in quality that computer graphics have gone through over the last several decades. From the early days of character-based graphics, through simple 8-bit color VGA resolution displays, to today’s 4K rendered images, the experience of using a PC has dramatically changed for the better thanks to these advances. The improvements in computer graphics aren’t just limited to PCs, however, as they’ve directly contributed to enhancements in game consoles, smartphones, TVs, and virtually every display-based device we interact with. The phenomenal success of gaming across all these platforms, for example, wouldn’t be anywhere near as impactful and wide-ranging if it weren’t for the stunning image quality that today’s game designers can now create.

Of course, these striking graphics are primarily due to graphics processing units (GPUs)—chips whose creation and advancement have enabled this revolution in display quality. Over the years, we’ve seen GPUs used to accelerate the creation of computerized images via a number of different methods including manipulating bitmaps, generating polygons, programmable shaders, and, most recently, calculating how rays of light bounce off of images in a scene to create realistic shadows and reflections—a technique referred to as ray tracing.

Ray tracing isn’t a new phenomenon—indeed, some of the earliest personal computers, such as the Amiga, were famous for being able to generate what—at the time—felt like very realistic looking images made entirely on a PC via ray tracing. Back then, however, it could often take hours to complete a single image because of the enormous amount of computing power necessary to create the scene. Today, we’re starting to see the first implementations of real-time ray tracing, where GPUs are able to generate extremely complex images at the fast frame rates necessary for compelling game play.

Nvidia kicked off the real-time, PC-based ray tracing movement with the debut of their Turing GPU architecture and the RTX 2000 series graphics cards based on those GPUs last year. Now the company is working to push the momentum forward with their second-generation desktop graphics cards, the RTX Super line, including the RTX Super 2060, RTX Super 2070, and RTX Super 2080. All three cards offer performance improvements in both ray tracing and traditional graphics acceleration. At the high end ($999), the RTX 2080 TI remains as the highest performing card in the Nvidia line, while at the low end ($349), the original RTX 2060 remains as the lowest priced option. In between, the original 2070 and 2080 are being replaced by their Super versions (but at the same $499 and $699 prices), while the Super 2060 at $399, ups the onboard graphics memory to 8 GB and nearly matches the performance of the original RTX 2070. As a bonus, all three RTX Super cards come bundled with two games that support real-time ray tracing: Control and Wolfenstein: Youngblood.

Nvidia faced some criticism (and, reportedly, saw somewhat muted sales) after the launch of the first generation RTX cards because of the limited support for real-time ray tracing in many popular PC gaming titles. Since then, the major gaming engines, including Unreal and Unity announced support for ray tracing, as well as Microsoft’s Direct X Ray Tracing (DXR) API, and several AAA gaming titles, including Cyberpunk 2077 and Call of Duty: Modern Warfare. In addition, other games, such as Quake II RTX and Bloodhound have also announced support for accelerated ray tracing hardware.

On top of this, recent announcements from both Microsoft (Project Scarlett) and Sony (PlayStation V) made it clear that the next generation of game consoles (expected in 2020) will incorporate hardware-based support of real-time ray tracing as well. Interestingly, both of those devices will be powered by AMD-designed GPUs, strongly suggesting that AMD will be bringing real-time ray tracing hardware technology to future generations of their Radeon line of desktop and laptop GPUs.

As the market has demonstrated, not everybody currently feels the need to purchase GPUs with dedicated ray tracing accelerated hardware. Many gamers focus on purchasing desktop graphics cards (or gaming laptops) that can play the current titles they’re interested in at the fastest possible frame rates and the highest possible screen resolutions at price points they can afford. For those gamers who are thinking ahead, however, it’s clear that there’s a great deal of momentum starting to build around real-time ray tracing. In addition to the previous examples, both Nvidia and AMD have announced software-based support of ray tracing in the latest drivers for their existing GPUs, which will likely encourage more game developers to add support for the technology in their next generation games. While the software-based solutions won’t run as fast, nor provide the same level of image quality for ray traced effects as hardware accelerated solutions, they will at least make people more aware of the kind of graphics enhancements that ray tracing can provide.

The evolution of computer graphics is still clearly moving ahead and, as a long-time industry -watcher, it’s great to see the once far-off concept of real-time ray tracing finally come to life.

OK, let’s get this part out of the way first. In the right conditions, 5G is fast—really fast. Like 1.8 Gbps download speed fast. To put that into perspective, we’re talking 5-10x faster than even the fastest home WiFi, and more than 50x faster than a lot of the typical 25-35 Mbps download speeds most people experience with their day-to-day 4G LTE connections.

The catch is, however, that the “right conditions” are rarely going to be available. At AT&T’s recent Shape Expo event on the Warner Bros. studio lot in Burbank CA, I did actually see just over 1.8 Gbps on a speed test using Samsung’s brand new S10 5G phone when I stood 75 feet away from a tiny cell tower installed as part of a new 5G network on the lot and pointed the phone directly at it. Impressive, to be sure.

However, when I turned away and walked another 50 feet from the tower and held the phone in my hand as you normally would (and not in direct sight of the special 5G antenna that was part of the network), the speed dropped to just under 150 Mbps because the connection switched over to LTE. Now, that’s still nothing to shake a stick at, but it’s more than 10x slower than the fastest connection. This succinctly highlights some of the challenges that 5G early adopters will likely face.

To understand the dilemma, you need to know a bit more about how 5G networks work. First, the good news is that 5G builds on top of existing 4G LTE networks, and whenever 5G signals aren’t there, smartphones and other devices with cellular modem connections (such as wireless broadband access points—often nicknamed “pucks” because they look a bit like hockey pucks) fall back to 4G. Plus, as my experiment showed, it’s often a very good 4G connection, because any phone with 5G also typically has the most modern 4G modems. Similarly, locations that have 5G networks usually have the most current 4G technology installed as part of the network as well. Together, that combination typically means that you’ll get the best 4G network connection you can—to put it numerically, it can be as much as 5x faster than the typical LTE speeds many people experience today.

Within the 5G world, there are two basic types of connections that leverage two different types of radio frequencies to deliver the signals from cellular networks to devices: millimeter wave and what’s termed “sub-6”—short for sub, or below, 6 GHz. Millimeter wave signals (so named because their wavelengths are about 1 millimeter in length) are extremely fast, but they don’t travel far and demand a direct line-of-sight connection. Like Verizon and T-Mobile, AT&T’s initial implementation of 5G networks use millimeter wave technology and the new Samsung S10 5G supports that as well.

So, back to my original test, I was only able to see the crazy-fast 1.8 Gbps download speeds when the phone was within the short range and direct line-of-sight of the 5G tower, which was transmitting millimeter waves at 39 GHz (which happens to be one of the frequency bands that AT&T controls). As soon as I moved a bit away and that connection was lost, both the phone and network connection fell back to 4G LTE—albeit the latest LTE Advanced Pro version of 4G (which AT&T confusingly calls 5Ge, or 5G Evolution). In other words, to really enjoy the full benefits of 5G speed and millimeter wave technology, carriers like AT&T are going to have to install a lot (!) of 5G millimeter wave-capable technology. Thankfully, 5G-specific antennas can be added to existing 4G towers and 5G smalls cells take up much less space than typical cellular network infrastructure components, but there’s still going to have to be a lot more independent 5G cell sites to fully leverage 5G.

Later down the road for AT&T, Verizon and T-Mobile (but in the forthcoming initial implementation of 5G from Sprint), you’ll be able to access the “other” kind of 5G frequencies, collectively referred to as “sub-6”. The sub-6 frequencies can all travel farther than millimeter wave and don’t require line-of-sight, so they can work in a lot more places (including inside buildings). However, they’re also much slower than millimeter wave. As a result, the “sub-6” 5G options will enable much wider coverage but won’t really be significantly faster than many 4G LTE networks. (FYI, all existing 4G radio connections occur below 6 GHz as well, in fact, below 3 Ghz, but they use different methods for connections and different types of radio frequency modulations than 5G.) Practically speaking, this means it will be easier to build out better coverage networks with “sub-6” 5G, but at the expense of speed. It’s a classic engineering tradeoff.

Of course, there’s more to 5G than just speed and some of that potential for future 5G applications was also on display at the AT&T Shape Event. Most notably, reductions in latency, or lag time, can start to enable much better, and more compelling implementations of cloud-based gaming over mobile network connections. Nvidia, for example, showed off a lag-free 5G connected version of its GeForce Now cloud gaming service, which allows you to have a high-end desktop gaming experience powered by Nvidia graphics chips even on older PCs or laptops. In addition, several vendors started talking about delivering higher-quality video and graphics to AR and VR headsets courtesy of future 5G products.

There’s no question that 5G can and will make a large impact on many markets over time. But as these real-world experiences demonstrate, it’s a complicated story that’s going to take several years to really show off its full potential.

The range of choices that enterprises have when it comes to both locations and methods for running applications and other critical workloads continues to expand at a dizzying rate. From public cloud service providers like Amazon’s AWS and Microsoft’s Azure, to on-premise private cloud data centers, as well as traditional legacy applications, to containerized, orchestrated microservices, the range of computing options available to today’s businesses is vast.

As interesting as some of the new solutions may be, however, the selection of one versus another has often been a binary choice that necessitated complicated and expensive migrations from one location or application type to another. In fact, there are many organizations that have investigated making these kinds of transitions, but then stopped, either before they began or shortly after having started, once they realized how challenging and/or costly these efforts were going to be.

Thankfully, a variety of tech vendors have recognized that businesses are looking for more flexibility when it comes to options for modernizing their IT environments. The latest effort comes via an extension of the partnership between HPE and Google Cloud, which was first detailed at Google’s Cloud Next event in April. Combining a variety of different HPE products and services with Google Cloud’s expertise in containerized applications and the multi-cloud transportability enabled by Google’s Anthos, the two companies just announced what they call a hybrid cloud for containers.

Basically, the new service allows companies to create modern, containerized workloads either in the cloud or leveraging cloud software technologies on premise, then run those apps locally on HPE servers and storage solutions but manage them and run analytics on them in the cloud via Google Cloud. In addition, thanks to Anthos’ ability to work across multiple cloud providers, those workloads could be run on AWS or Azure (in addition to Google Cloud), or even get moved back into a business’ own on-premise data center or into a co-location facility they rent as needs and requirements change. In the third quarter of this year, HPE will also be adding support for its Cloud Volumes service, which provides a consistent storage platform that can be connected to any of the public cloud services and avoids the challenges and costs of migrating that data across different service providers.

On top of all this, HPE is going to make this offering part of their GreenLake, pay-as-you-go service consumption model. With GreenLake, companies only pay for whatever services they use—similarly to how cloud computing providers offer infrastructure as a service (IaaS). However, HPE extends what companies like Amazon do by providing a significantly wider range of partners and products that can be put together to create a finished solution. So, rather than having to simply use whatever tools someone like Amazon might provide, HPE’s GreenLake offerings can leverage existing software licenses or other legacy applications that a business may have or may already use. Ultimately, it comes down to a question of choice, with HPE focused on giving companies as much flexibility as possible.

The GreenLake offerings, which HPE rebranded about 18 months ago, are apparently the fastest growing product the company has—the partner channel portion of the business grew 275% over the last year according to the company (though obviously from a tiny base). They’ve become so important, in fact, that HPE is expected to extend GreenLake into a significantly wider range of service offerings over the next several years. In fact, in the slide describing the new HPE/Google Cloud offering, HPE used the phrase “everything as a service,” implying a very aggressive move into a more customer experience-focused set of products.

What’s particularly interesting about this latest offering from the two companies is that it’s indicative of a larger trend in IT to move away from capital-intensive hardware purchases towards a longer-term, and theoretically stickier, business model based on operating expenses. More importantly, the idea also reflects the growing expectations that IT suppliers need to become true solution providers and offer up complete experiences that businesses can easily integrate into their organizations. It’s an idea that’s been talked about for a long time now (and one that isn’t going to happen overnight), but this latest announcement from HPE and Google clearly highlights that trends seem to be moving more quickly in that direction.

From a technology perspective, the news also provides yet more evidence that for the vast majority of businesses, the future of the cloud is a hybrid one that can leverage both on-premise (or co-located) computing resources and elements of the public cloud. Companies need the flexibility to have capabilities in both worlds, to have additional choices in who manages those resources and how they’re paid for, and to have the ability to easily move back and forth between them as needs evolve. Hybrid cloud options are really the only ones that can meet this range of needs.

Overcoming the complexity of modern IT still remains a significant challenge for many organizations, but options that can increase flexibility and choice are clearly going to be important tools moving forward.