Intel offers a glimpse into its future – The main innovations of IDF

Traditionally, we know the Intel Developer Forum, or IDF for short, as a three-day nerd party. Intel previously announced its latest processors at IDF, showcased future products and technology the company is working on, and took the opportunity to showcase its latest initiatives. However, in the last few years we have seen a change; slowly but surely, the IDF has come to live up to its name and focus not so much on the press as on the developers. That means, among other things, a new slogan: “What will you make?” and a clear emphasis on the community of creators. There is of course nothing wrong with that, but the in-depth backgrounds about processor architectures are compromised in this way.

That does not alter the fact that Intel also introduced a large number of products, technologies and initiatives during IDF16. In doing so, the manufacturer did not focus on its ‘old’ product lines, but on the new, innovative and maker products. In doing so, the company actually sends out two signals; it is trying to transform itself from chip manufacturer to facilitator of new technology and on the other hand it seems a clear signal that the client PC, the desktop and laptop, is becoming less and less important. We list the announcements and noteworthy developments. There may of course be some overlap with previous news items.

Kabby Lake

We’ll start with the elephant in the room: Kaby Lake. Skylake’s successor was not presented at IDF16, although CEO Brian Krzanich spoke briefly about it during the keynote. However, Intel didn’t show much more than a short demo of the built-in graphics processor. Overwatch was played on a Dell XPS 13 laptop with ulv-i5 and we got to see how the laptop could not only play but also edit 4k content. Now, 4k video playback isn’t an earth-shattering feature, but the laptop had surprisingly little trouble with it. So was editing 4k video, which could be done almost in real time on the new platform. Because all of Kaby Lake’s performance numbers are still under wraps, Krzanich didn’t comment on Overwatch’s resolution and quality settings.

Project Alloy

We have already reported on Project Alloy and Intel does not want to disclose much more than the information from our news item. Project Alloy is an open hardware design for a VR headset, which means that any manufacturer can take Intel ‘s reference design and adapt it where desired. That would mean we could see some Alloy-based VR headsets in the future, but little is known about the specs.

It is clear that it is a self-contained HMD , so you would not need a PC to use the VR headset. Namely, that PC is integrated into the headset, with Intel placing an unnamed series Skylake processor in the headset, complete with memory and storage. There are batteries in the band at the back of the head to provide energy to Alloy. So you can move freely without being hindered by a cable to a desktop PC or being condemned to walk around with a backpack PC.

That walking around has a potential downside, of course; you can run into all kinds of objects and people. Where with an HTC Vive or Oculus you are bound to a specific room in which you have placed the sensors, with Alloy you have complete freedom. To avoid objects and people, the Alloy HMD would therefore not be so much a VR headset, but a mixed reality headset. You can see real objects and people in your virtual world, and even manipulate the virtual world with objects from the physical world. This is made possible by two RealSense cameras that are in the headset and provide depth information about your surroundings. This also allows you, for example, to see your own hands and use them as ‘controllers’ in the VR world. The Alloy headset has two additional RGB cameras

With a PC integrated into the HMD, the Project Alloy headset is most similar to Microsoft’s Hololens. That HMD also has computing power built into the headset and is not dependent on an external computer. The question is therefore to what extent Project Alloy should be seen as an Oculus or Vive competitor and perhaps more of a Hololens alternative. The fact that the hmd was developed together with Microsoft and that Windows 10 will receive an update next year for Holographic, the holographic shell of the operating system, seem to be indications of this. Nevertheless, Intel employees believe that the Alloy HMD is indeed powerful enough for games, thanks to the Skylake processor with Iris GPU. The performance should then be somewhere between that of a headset for phones and the performance of a high-end headset like the Vive or Oculus.

Incidentally, engineers at Intel are not only working on Project Alloy, existing VR headsets would also be made wireless or less dependent on PCs. For example, we came across an Oculus Rift with a WiGig adapter built-in. It was a prototype of course; for ease of hacking, the second development kit was used and the housing was made with a 3d printer. A battery had to provide the punch, so that the headset could work wirelessly. With a compact battery, the headset could run for over an hour, with the video for the Oculus screens being sent wirelessly from a PC to the headset via the WiGig adapter at 60GHz.

Another option to do something about the cable mess could be to replace a Vive’s USB, video and power cable with a single thunderbolt cable. After all, USB 3.1, display port and power can be sent over Thunderbolt 3 cables, everything that a Vive needs, for example.

The only drawback is that an active cable with copper wires cannot be longer than two meters. If it is longer, the signals will be too distorted. However, Intel is working on an extension of Thunderbolt 3, where the data cables in the Thunderbolt 3 cable are optical. They can bridge longer distances of up to five or ten metres. The power wires must of course remain made of copper, but they are not affected by the extra length. That standard isn’t there yet, but we did find a company that has developed an all-optical Thunderbolt 3 cable. In fact, the company only makes the connectors that convert the electrical signal into an optical signal. The prototype cable the company showed was five meters long and only suitable for data; power cannot pass through due to the lack of copper wires.

Project Euclid

Intel is fully committed to the Projects, besides Alloy Euclid was also announced at this IDF. We’re all familiar with Intel’s RealSense camera: a kind of Kinect with a regular color camera, an infrared camera, and an infrared laser projector. The IR camera picks up reflected laser light from the projector and builds a 3D point cloud of the environment that can be overlaid on the images from the regular camera. This gives you a 3D image of your surroundings that is very precise, enough to distinguish hand and finger movements, and in combination with Windows Hello, precise enough to recognize your face to log into Windows. There are basically two versions of the Realsense camera: one with a short range, for laptops with Windows Hello and gesture recognition, among other things, and a version with a longer range, which does not have a, only has two IR cameras. These first generation RealSense cameras are known as F200 and R200 respectively.

With the Camera 400 series, a new, improved camera has been added to the RealSense series. The 400 series should replace the F200 and R200, because it is suitable for both indoor and outdoor use, i.e. short and long distances. Not only the range has been improved, but the number of 3D points that the camera can observe per second has also doubled. The precision would be much better that way. In addition, the 400 series is a lot smaller and, above all, much thinner than its predecessor, so that the number of devices in which the RealSense camera can be accommodated is much larger. Thin-bezel laptops, tablets and even phones can now be equipped with the camera.

It is somewhat unclear whether the 400 series is also used in Project Euclid, but Intel has managed to accommodate a complete platform in Euclid in addition to a RealSense camera. The result is a module with a 3D camera and an Atom system the size of a Mars. The platform was developed as a vision platform for robots, so that developers can equip a robot with a RealSense camera for 3D perception without much effort. An onboard battery supplies the energy, and Euclid runs Ubuntu and Robot OS, or ROS for short, by default. Onboard WiFi can provide communication with other robot components. Intel also introduced a second RealSense developer kit for robots: the RealSense Robotic Development Kit. It is based on a separate RealSense R200 module and a PCB the size of a Raspberry Pi with an Atom x5-Z8350 running Ubuntu. That kit should cost three hundred euros and be available next month; what Project Euclid will cost is not yet known.

Project Aero

The price of Project Aero is well known; it will cost $399. For that money you get a complete drone operating system in the form of a pcb with an Atom x7-Z8700 quad core running Yocto-Linux and has 4GB lpddr3 memory and 16GB emmc. There is an interface for a regular and a RealSense R200 camera, where the 3d camera must provide obstacle detection and avoidance. Built-in WiFi-AC provides connectivity and of course there are plenty of I/O headers to communicate with a drone’s other hardware. Those who find self-construction too much trouble can also wait until the end of this year. Intel will then release a Ready-to-Fly kit based on the Aero, complete with frame, RealSense camera, motors and even a remote control. The price of that kit is not yet known.

Project Joule

Those who have not yet got the idea that Intel is increasingly focusing on the ‘maker’, as witnessed by Aero and Euclid, and only casually showing a bit of Kaby Lake, may be convinced by the last developer kit we mention: Joule, which we have for the convenience of giving Project status as well. No kidding, Joule is a maker kit and also meant for iot development. You have to think of it a bit like Edison on steroids. Edison was shown at the IDF in 2014 and had two Atom Silvermont cores. Joule is available in two versions, the 550X and the 570X, both equipped with four Atom Goldmont cores, a T5500 and a T5700 respectively. The memory and storage also differ, which amount to 3 and 4GB lpddr4, combined with 8 and 16GB emmc memory. Wireless ac and bluetooth 4.1 are available, as well as a choice of i/o. Nothing special you would say, until you see the dimensions of the modules; those are only 48 by 24 by 3.5mm. Slightly more elongated than Edison, but less than 5 by 2.5 cm in size.

In addition, you now have a real Gen9 GPU at your disposal, so that you no longer have to develop headlessly and can also use that GPU power for visual tasks. The small computer has 18 execution units for graphic work, and the soc has a separate ASIC for depth vision on board. It’s up to the iot visionaries and makers to decide what the Joule modules are good for. Intel mentions robots, VR and AR, drones and computer vision as possibilities. For the latter there is support for RealSense cameras and the modules come with a development board to easily connect all the i/o. With a price of $ 369 for the 570X, it is clear that you can expect a bit more from the Joule than from a Raspberry Pi of a few tens.

10nm and beyond

Once the heart of the IDFs, now hidden in a presentation on Altera’s latest FPGAs, are Mark Bohr’s presentations. Traditionally, Bohr, an architecture and lithography specialist, can tell you about the latest techniques and state-of-the-art developments. Fortunately, we were able to learn something about Intel’s future 10nm designs. Everything seems to be on track and Intel insists it still has a significant lead over the competition. That seems strange, since Intel now produces on 14nm and various other foundriesdo that too. Now the node, as 22nm, 14nm or 10nm is called, is no longer an indication of the actual transistor size; it’s really just a marketing term. Nevertheless, the transistor size still shrinks with every smaller node, and one of the dimensions that is indicative is the gate pitch , or the distance between two gates.

With the 28nm and 20nm (or 22nm, if you will) process, the gate pitches between the foundries were still fairly equal, but with the 14/16nm node, many foundries actually got stuck and only a few other dimensions were added. got a little smaller. Intel says it actually made the gate pitch about 0.76 times smaller, from 22 to 14nm and would match that to 10nm. The competition would not be able to keep up, reason for Intel to claim to be a generation ahead.

That the 10nm node scales well is confirmed by the company with another measurement: the transistor surface is always scaled 0.46 times in the last nodes, but the 14nm and 10nm nodes would scale slightly better than that. In this respect, the company lagged behind the competition, but again Intel claims a generation ahead of the competition for the 10nm node. That’s all nice for Intel, but the good news for us consumers is that both 10nm and 7nm transistors continue to get cheaper. Not only that, the switching speed and energy consumption are still increasing and decreasing respectively. That means cheaper, faster and more efficient transistors.

It is now known that Intel no longer maintains its tick-tock model, but according to Bohr there are certainly advantages to Kaby Lake’s improved process. Intel calls this a 14Plus process, in which the lithography has been improved. In practice, this mainly means higher fins on the finfets, which leads to better performance. According to Intel’s promise, that would even yield up to twelve percent performance gain. Such improvements are also planned for the 10nm node. Intel even speaks of three ‘waves’ of 10nm products: a 10nm process, followed by 10+ and 10++. The 10nm process would also allow more designs, thanks to additional components that can be etched, including analog components. Moreover, the chips could be easily tunable for high performance or very economical applications. To this end, it is also possible

5g, cars and data centers

It’s been clear for a few years now that Intel is reinventing itself. Of course, the company continues to make good processors for laptops and desktops, but the consumer market is becoming less and less important. We see that reflected in the topics at the IDF, and in reorganizations and shifted centers of gravity. One of the focus areas that Intel is responding to with increasing flexibility is the data center. Data centers will become increasingly important in the coming years, and will have to deal with more and more data, information and calculations. The internet of things, a buzzword and unclear term for many, would make a strong contribution to this. Think of many more connected devices from wearables, cars, sensors and cities. The first category is only a minuscule part of it. For the consumer, the IoT may not mean much; after all, what kind of smart equipment should you have with you? However, the IoT must encompass much more, from home automation systems and in office buildings to entire cities.

Another big change that is coming not only according to Intel, but also according to many automakers and technology companies, is the smart car. Tesla is already a tangible example of this, but in the near future cars would be fully autonomous and collect huge amounts of sensor data. This data is partly processed locally in the cars, but partly also shared with other traffic, with users and with cities, so that traffic can be regulated.

A fast wireless network is indispensable for such applications, which is why the development of 5G networks is being accelerated. 5g would not simply be the successor to 4g for our mobile data traffic, but a fully software-manageable collection of network protocols that can use all currently available wireless networks and frequencies as needed. In addition to an enormous bandwidth, in the order of gigabits, the latency would also be much lower: less than 2ms. The latter is extremely important for autonomous cars. Cars coming around a corner need to know without delay what is happening around the corner to avoid accidents.