Topps Blog

Monday, 7 November 2022

Students Build DIY Supercomputer Out of Nvidia Jetson Nanos

For nearly as long as we’ve had supercomputers, we’ve also had people asking themselves, “How do I build myself one of those, except with a tenth of the budget and using just a fraction of the power?” Several teams of scientists have built “Beowulf clusters,” supercomputers that are actually clusters of commodity-grade hardware, sharing their own LAN. And remember all those PlayStation supercomputers? Now, a team of students at Southern Methodist University in Dallas has built a supercomputer by connecting 16 Nvidia Jetson Nano modules together, along with four power supplies, a network switch, some cooling fans, and roughly five dozen handmade wires. (Fact: All the best prototypes always have hand-soldered wires hanging out the back.)

According to Conner Ozenne, a senior computer science major and one of the leads on the project, “We chose to use Nvidia Jetson modules because no other small compute devices have onboard GPUs, which would let us tackle more AI and machine learning problems.”

‘Baby’ Supercomputer

Architecturally, the Jetson Nano is most similar to the Nintendo Switch, which runs on Nvidia’s Tegra X1 SoC, so we’ll use that as a point of comparison.

Students from the Southern Methodist University in Dallas built their DIY “baby supercomputer” from sixteen of these Jetson Nano modules. The students will be showing off their mini cluster at the SC22 supercomputing conference in Dallas.

The Switch and the Nano have the same theoretical maximum memory bandwidth (25.6 GB/s). They’ve also got the same quad-core Cortex-A57 SoC, but the Nano’s CPU is clocked considerably higher (1.43GHz versus 1.02GHz for the Switch when docked). As far as the two platform’s relative GPU power, however, the situation is reversed. The Maxwell-based Tegra X1 SoC inside the Switch offers 256 shader cores compared with just 128 on the Jetson Nano.

While this implies the Nano would be half the speed of the Switch in the same workload, the gap might not be quite that large. The Switch reportedly tops out at 768MHz in docked mode while the Jetson Nano has a maximum clock of up to 921MHz. Altogether, the “baby” supercomputer combines 64 Cortex-A57 cores, 64GB of RAM, and 2,048 Maxwell cores across 16 boards.

Nano Lives Up To Its Name

Let’s tackle the elephant in the room first. The objective specs of the SMU 16-board supercomputer are scarcely inspiring, considering that single-socket desktop systems now offer as many as 64 cores. Jetson Nano is really living up to the ‘nano’ part of its name here. Not only are the stats pretty pedestrian on their own, the entire cluster literally fits on a desk.

But all kidding aside, comparing the specs of a system like this to conventional PC hardware misses the point. The challenges associated with scaling workloads effectively across a large network of slow devices, with a relatively small amount of memory per device, are conceptually similar whether one is discussing true supercomputers or smaller-scale embedded device systems like this one.

NVIDIA’s Jetson Orin system-on-module

“We started this project to demonstrate the nuts and bolts of what goes into a computer cluster,” said Eric Godat, the team lead for research and data science in SMU’s IT organization. “The mini-cluster is an effective teaching tool for how all this stuff really works — it lets students experiment with stripping the wires, managing a parallel file system, reimaging cards, and deploying cluster software.”

Price vs. Performance

Any given AI workload would likely run better on the GTX 980 (2,048 cores on one chip) as opposed to 16 Jetson Nano GPUs across 16 boards, but the latter is a much better, if still simplistic, simulation of some of the scaling challenges full-scale supercomputing engineers deal with on the job.

Nvidia’s blog post references the idea of upgrading the current 16-board system with Jetson Orin Nano hardware. The performance boost from any such jump would be considerable. As we’ve previously detailed, Orin Nano offers six Cortex A-78AE CPU cores at 1.5GHz and 512 Ampere GPU cores with 16 tensor cores. Jetson Nano is a comparative shrimp with its 4x Cortex-A57 CPUs and 128 Maxwell cores. Orin Nano is more expensive than Jetson Nano, however, at $199 versus $129.

NVIDIA’s Jetson Orin Nano system-on-module. This is the chip NVIDIA suggests for an upgrade to the SMU “baby supercomputer.”

Orin Nano’s performance improvement ought to be much larger than the increase in price, but we hope Nvidia brings a still lower-cost Orin to market in this space. A $129 Orin Nano with 256 Ampere cores and, say, eight tensor cores would still be a huge upgrade.

At the same time, Nvidia has little reason to cut prices. Right now, the Jetson Nano really only competes with itself. While there are some other ARM-based boards that are compatible with accelerators, the Jetson Nano’s GPU is the only product in its price class and of its type.

The students will be showing off their mini cluster at the SC22 supercomputing conference in Dallas. This year, SC22 runs Nov. 13-18.

Now Read:

Crewless Black Hawk Helicopter Completes First Round of Autonomous Resupply Missions

(Photo: Sikorsky) Resupply missions are vital to the success of nearly any combat scenario. They’re also dangerous, placing drivers and pilots in harm’s way as they navigate the battlefield and perilous airspace. But what if supply vessels could undergo these missions on their own, no human operators required?

If the US Army’s most recent autonomous flight experiment is anything to go by, this might occur sooner than we think. Last year, the Army partnered with Sikorsky (a Lockheed Martin subsidiary) and the Pentagon’s Defense Advanced Research Projects Agency (DARPA) to retrofit a Black Hawk helicopter with autonomous flight technology. February saw the helicopter complete its first successful test flight without a pilot. The test flight didn’t simulate any life-saving operations, however, meaning its real-life practicality was still unconfirmed.

In mid-October, the Army successfully completed a series of autonomous resupply and rescue missions affirming the Black Hawk’s value in dangerous conditions. One was a long endurance medical resupply mission, in which the helicopter independently flew 83 miles over Yuma, Arizona while carrying 500 pounds of real and simulated human blood supply. During a portion of the 50-minute flight, the Black Hawk descended to 200 feet above ground to mask its signature, meanwhile maintaining 100 knots (115 miles per hour) airspeed.

(Photos: Sikorsky)

A second test mission required the aircraft to lift off with a 2,600-pound payload attached to the end of a 40-foot sling. At the end of the 30-minute, 100-knot flight, a ground operator used radio and a tablet to direct the Black Hawk to lower its payload to the ground. The operator then commanded the Black Hawk to pick up a simulated casualty nearby. Crews loaded a manikin patient into the aircraft’s cabin, after which the operator initiated a short autonomous flight to a field hospital. A built-in health monitoring system relayed the manikin’s vitals to medical personnel on the ground during transport.

The Black Hawk’s autonomous flight capabilities are made possible via Matrix, Sikorsky’s “drop-in, removable kit” that can be retrofitted to existing helicopters and winged aircraft. Matrix makes the aircraft pilot-optional, allowing for autonomous flight in a pinch (such as in dangerous battlespace or limited-visibility environments) and piloted missions under safer circumstances. Despite compatibility with ground-based operators, Matrix allows the aircraft’s flight computer to create a flight plan all on its own using destination location, distance, topography, and other high-level information.

“During PC22 Technology Gateway, we showed how the optionally piloted Black Hawk helicopter can be flown by human pilots, who then land the aircraft and simply flip a switch to activate flight with zero pilots,” said Igor Cherepinsky, Sikorsky Innovations director, in a statement. “We believe Matrix Technology is ready now for transition to the Army as they look to modernize the enduring helicopter fleet.”

Now Read:

Researchers Build Creepy Drones That Can ‘See’ Other Wi-Fi Devices Through Walls

(Photo: University of Waterloo) The twenty-first century might be chock-full of unprecedented privacy violations, but at least we’re safe in our own homes, right? Wrong. For some reason, scientists are working on technology that helps drones “see through walls,” and if previous groundbreaking tech is anything to go by, it’s only a matter of time until it ends up in the wrong hands.

A pair of researchers at the University of Waterloo and the University of Illinois have created what they call “Wi-Peep,” a system that uses the data in Wi-Fi networks to locate devices behind closed doors. Wi-Peep itself consists of a low-cost ESP8266 Wi-Fi microchip, an ESP32 Wi-Fi and Bluetooth microcontroller, a microSD module, and a voltage regulator. The researchers attached these to a lightweight drone—an unnecessary but convenient addition that enhances their creepy project’s mobility.

Using what the researchers refer to as a “location-revealing privacy attack,” Wi-Peep exploits security gaps in the local access network protocol known as IEEE 802.11. Because wireless devices are typically made to accept connection requests (known as “contact attempts”) from other devices in close proximity, Wi-Peep can use Time-of-Flight (ToF) signals to ping a device it has no business attempting contact with. Wi-Peep can then use the resulting feedback to approximate the distance between itself and the target device.

(Image: Ali Abedi and Deepak Vasisht)

How does Wi-Peep know what type of device it’s “seeing?” A device’s attributes are discoverable through its media access control (MAC) address, which is used to uniquely identify each device on a network. By gaining hold of a target device’s MAC address, Wi-Peep can help confirm whether an item is indeed the smart TV, tablet, smart home assistant, or another device it’s suspected to be.

The researchers, who specialize in computer science and the Internet of Things (IoT), are well aware that Wi-Peep and similar technologies can be wielded with malicious intent. In a presentation at MobiCom 2022, they expressed concern that bad actors could use Wi-Peep to track the locations of security guards in a bank or of unwitting families inside a residence.

But devices like Wi-Peep could have even broader privacy implications. Police officers—some of whom have made it clear they’re not above falsifying warrant information to invade people’s homes—could use the tech to conduct home and vehicle searches without people’s knowledge or consent. Insurance companies could use it to sneakily verify applications and claims, abusive exes could use it to control or spy on their former partners, and burglars could locate valuable electronics before strategizing a break-in.

Wi-Peep is a demonstration of the dire need for improved network protocol protections, as the researchers noted at MobiCom 2022. Whatever follows IEEE 802.11—a protocol known for its data interception loopholes—should be tougher to compromise than its predecessor. That is, unless we want people monitoring our home devices from the outside.

Now Read:

Samsung Unveils First 8K Ultrawide Gaming Monitor With DisplayPort 2.1

In the wake of yesterday’s AMD RDNA3 GPU reveal, the era of 8K gaming is suddenly upon us. Sure, it existed before, theoretically. But nobody was really gaming at 8K as you needed a television to do it. Plus, frame rates were pretty borderline, as we recall. All that’s changed now that AMD’s newest GPUs have a DisplayPort 2.1 interface. The previous DisplayPort 1.4a version was limited to a peasant’s refresh rate of 60Hz at 8K. The new version bumps that up to a satisfying 165Hz. And according to AMD, its new GPUs can actually achieve that frame rate in some games at that resolution. In light of this advancement, Samsung has announced the world’s first 8K UltraWide gaming monitor.

Now, some of you might recall Nvidia tried to make 8K gaming a thing when it launched the RTX 3090 two years ago. Nobody really bought it at the time, but it made for a good marketing slogan since AMD never brought it up. That is despite the fact that its 6000-series GPUs were just as capable as Nvidia’s Ampere cards. However, Nvidia has ditched that angle with its RTX 4090 GPUs, focusing instead on improvements at 4K with its Deep Learning Super Sampling (DLSS). The reason is simple: Nobody games at 8K. Samsung is hoping to change that with its new Odyssey Neo G9 49-inch UltraWide monitor.

Odyssey-G9

According to PC Gamer, it’s not going to be a true 8K panel. Instead, it’ll only offer that resolution horizontally, not vertically. That will endow it with a resolution of 7680 x 2160. If it were a true 8K display, that would be bumped to 7680 x 4320. Sure, it’ll still probably make any GPU cry out in agony, but buyer beware. Samsung’s current version of the Neo G9 sports a 5,120 x 1,440 resolution with a 240Hz refresh rate, so it already requires a beefy GPU for AAA gaming. The new 8K version will pretty much require a new AMD GPU, though we’re sure the RTX 4090 and upcoming 4080 will probably handle it reasonably well, but they’ll be limited to 135Hz at ultra-wide resolution.

The new DisplayPort 2.1 interface allows for some truly breathtaking data transfer speeds. It supports 4K at 480Hz, 8K at 165Hz, and 1440p at 900Hz. That level of throughput should see the interface enjoy a long gaming life cycle. According to AMD, although Samsung will be the first company with a DisplayPort 2.1 monitor, more will follow. LG, Asus, Dell, and Acer are reportedly working on new gaming monitors set for release in early 2023.

If you want to run 8K at 165Hz, you’ll need a DisplayPort 2.1 compatible GPU like one of the new Radeons. (Image: AMD)

If there’s one takeaway from this move to 8K, it’s that high-end gaming is now quite an expensive hobby. The current Neo G9 sells for $2,300, though you can find it on sale for $1,700. Add a bit to that price for the 8K version and throw in a $1,000+ GPU and you’re looking at one expensive gaming setup. Not to mention the obvious fact of needing a pricey CPU, NVME storage, and so forth. A lot of people lauded AMD’s relatively “affordable” pricing for its RDNA3 GPUs, but that’s only in comparison to Nvidia’s pricing. A $1,000 for a GPU used to be the price of Nvidia’s outlandish Titan GPUs. Now, it’s considered a bargain in comparison to the RTX 4090. That’s kind of ridiculous on its face.

Regardless, we welcome our new 8K overlords. Though we’re still content to game at 1440p and 120Hz, we’re sure one day we’ll step up to 4K, then eventually 8K when that level of performance is more reasonably priced. Samsung hasn’t released pricing for its 8K Neo G9, but you can bet it’ll cost a pretty penny when it launches next year.

Now Read:

One Year Later, Barely Anyone Is Using Windows 11

It’s been a year since Microsoft began rolling out Windows 11, its first new version of Windows since the 2015 launch of Windows 10. With Windows 11, Microsoft addressed several shortcomings of Windows 10 and revamped the interface to make it more modern. However, it also dropped some features users had come to rely upon. While Windows 11 is a free upgrade for Windows 10 users, not everyone can install it. The latest report from Statcounter says that just 15 percent of Windows installs are on the latest version, which is far short of Windows 10’s performance at this juncture.

Statcounter’s numbers are not official, but Microsoft doesn’t provide exact numbers on a monthly basis. And if Microsoft ever does provide numbers, you can bet it’ll only happen when the news is good. With Windows 11, the news isn’t great. Windows 10 is still far and away the leader among Microsoft’s operating systems with a 71 percent market share. Windows 11 is a distant second with 15.45 percent, and Windows 7 still has a dedicated user base at 9.62 percent.

Microsoft first released Windows 11 in a very limited fashion, which accounts for almost zero market share in early 2022. However, even with wide availability, its average monthly growth is under one percent, based on Statcounter figures. To make matters worse for Microsoft, Windows 10 has proven very stable in the face of the new release. Its market share has hovered between 70 and 73 percent for most of the year, losing market share even more slowly than Windows 11 gains — perhaps most Windows 11 upgrades are coming from former Windows 7 and 8 users.

At this rate, it would take years for Windows 11 to overtake Windows 10. This was not the case when Windows 10 launched. After a year on the market, Windows 10 had roughly twice the Windows market share as Windows 11 does at the same point. Microsoft requires features like a newer CPU and a TPM 2.0 security module to run Windows 11. So, perhaps the much slower adoption rate won’t bother the company too much.

Those who wish to remain on Windows 10 can safely do so for another three years. In October 2025, Microsoft will stop updating that OS, forcing anyone who cares about security to get on a newer version.

Now read:

SoCal Hyperloop Tube Gets Dismantled, Turned Into a Parking Lot

As Elon Musk spends more time being a troll on his newly acquired social network, some projects are apparently falling by the wayside. SpaceX has reportedly demolished the Hyperloop test track near its offices in Hawthorne, California. In its place, the company has erected a parking lot, which seems like a fitting end for something Musk once hailed as the “fifth mode of transportation.” Now it’s just a place to park one of the other four.

The hyperloop concept was popularized in 2013 when Musk released the “Hyperloop Alpha” white paper, inviting companies to build on the idea. The gist is that you build a tunnel, depressurize it, and run pods through it at extremely high speed. Allegedly, this sci-fi mode of transportation could reach speeds as high as 760 miles per hour (1,223 kph). Musk founded The Boring Company to work on building hyperloop technology, but almost a decade later, there are no operational hyperloops, and the largest prototype (above) is now gone.

As news of the impending closure spread earlier this year, some of the engineers who took part in the construction got together to mourn the loss. Erik Wright, the owner of Precision Construction Services, noted that his firm’s work on the hyperloop track led to many more projects. “The Hyperloop is quite a badge of honor for us,” Wright told Bloomberg. The tunnel was about a mile long and 12 feet across (1.6 kilometers and 3.6 meters), consisting of steel, concrete, and aluminum. All these materials expand and contract at different rates, making it difficult to meet the required precision of 0.04 inches.

A version of Musk’s Hyperloop alpha from 2013.

The test track wasn’t getting a lot of use even before it was closed down. Since 2017, the track was used for student demonstrations of hyperloop technology. Musk says he still wants to build a hyperloop, but the Boring Company appears to be heading in other directions (including selling a cologne that smells like burnt hair). Musk’s tunnel-excavating firm has been working on a series of tunnels under Las Vegas, but these are not depressurized tubes for ultra-fast transport. Rather, they are just tunnels for Tesla’s electric vehicles to drive through.

It’s possible Musk is preparing to move more of his operations to Texas, which already hosts Tesla’s headquarters in Austin, as well as the SpaceX testing and development facility in Boca Chica. Musk, whose net worth is about $200 billion, has said he moved to Texas to save money on taxes. Maybe the dream of a hyperloop isn’t dead, but it’s definitely on the back burner.

Now read:

Friday, 4 November 2022

Matter Smart Home Standard Officially Launches With 190 Devices Supported

Most smart homes are still surprisingly dumb, but there’s a new technology that could change that. The Matter standard, a collaboration between more than 300 companies, was finalized last month, and now the first devices are ready to make Matter a reality. The Connectivity Standards Alliance (CSA) that oversees Matter held an event to talk about the lights, thermostats, sensors, and other smart home gadgets that you can now get with Matter support.

The biggest issue in setting up smart home devices is that most manufacturers have their own apps, and not all of them connect to the same platforms. For example, you can get light bulbs that will work fine with Alexa, but they might not talk to Google Assistant, SmartThings, or HomeKit. Matter is supposed to fix that by pulling together Wi-Fi, Bluetooth, and Thread connectivity under one umbrella. That means a Matter-certified device will be able to talk to all your Matter-enabled platforms. You get to choose the apps (or apps) you want to use to manage them.

According to the CSA, there are more than 190 devices already certified for Matter. Some of them are already available, but most will launch in the coming months. If you’ve got a Philips Hue bridge, that’s one of the first devices with Matter certification. An update will roll out soon that upgrades the bridge and its connected bulbs to work with Matter. If you’ve got an Amazon Echo device, you’ll also have Matter support in your home soon. Amazon says it’s adding support for Matter to 17 Echo devices this year, but only for Wi-Fi at first. Next year, Amazon will unlock Thread support for Matter on select Echo devices, as well as its Eero mesh routers. We also know that Google is updating most of its Nest and Google Home hardware with Matter support.

The Philips Hue bridge will be one of the first Matter controllers following a software update.

Since the matter 1.0 standard was released, 20 new companies have joined the project, and that number could expand as more device types become available. Right now, Matter works on simpler smart home gadgets like light bulbs and smart locks. The CSA already confirmed that next year will bring product categories like cameras and robot vacuums. At the event, CSA president Tobin Richardson confirmed Matter will also come to garage door controllers, air quality sensors, smoke and carbon monoxide detectors, and motion sensors.

While it’s nice to see the CSA managed almost 200 Matter device certifications out of the gate, that’s still a tiny sliver of the smart home ecosystem. It will take time for Matter-certified devices to displace older, cheaper gadgets that don’t speak Matter’s language, so you’ll have to keep an eye out for the Matter logo if you want to take advantage. The CSA also has a site that lists Matter-certified devices — however, it’s mostly light bulbs right now and is missing many of the products mentioned at the event.

Now read: