For as long as we’ve had supercomputers, we’ve also had people asking themselves, ‘How can I build myself one of these? thoseExcept for a tenth of the budget, using a fraction of the energy? Several teams of scientists have built “Beowulf clusters,” which are supercomputers that are actually groups of commodity-class devices, that share their own LAN. Remember all the PlayStation supercomputers? Now, a team of students at the University of Southern Methodist of Dallas built a supercomputer by connecting 16 Nvidia Jetson Nano units together, along with four power supplies, a network switch, some cooling fans, and roughly fifteen handcrafted wires.(Fact: All the better prototypes contain Always have hand-welded wires dangling from the back.)
According to Koner Ozen, Senior Computer Science Specialist and one of the project leaders, “We chose to use Nvidia Jetson modules because no other small computing devices have GPUs on board, which will allow us to tackle more AI and machine learning issues.”
Giant computer “Baby”
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 comparison point.
Switch and Nano have the same maximum theoretical memory bandwidth (25.6 GB/s). They also have the same quad-core Cortex-A57 SoC, but the Nano CPU clocks in at a much higher (1.43GHz vs. 1.02GHz for the switch when docked). Regarding the relative GPU power of the two platforms, the situation is reversed. The Maxwell-based Tegra X1 SoC inside the Switch offers 256 shader cores compared to just 128 on the Jetson Nano.
While this means that the Nano will be half the speed of the transformer at the same workload, the gap may not be very large. The switch is said to clock at 768MHz in docked mode while the Jetson Nano clocks in at 921MHz. In all, the “mini” supercomputer combines 64 Cortex-A57 cores, 64 GB of RAM and 2,048 Maxwell cores across 16 boards.
Nano lives up to its name
Let’s deal with the elephant in the room first. The objective specifications of the 16-board SMU supercomputer are rarely inspiring, considering that single-socket desktop systems now offer up to 64 cores. Jetson Nano really lives up to the “nano” part of its name here. Not only are the stats pretty pedestrian on their own, the entire set literally fits into the office.
But no joke, comparing the specs of a system like this to traditional computers misses the point. The challenges associated with effectively scaling workloads 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 small compact hardware systems like this.
“We started this project to illustrate the weaknesses and the screws of what goes into the computer suite,” said Eric Juddat, Head of Research and Data Science at SMU’s IT Enterprise. “The mini-collection is a powerful teaching tool for how all of this really works – it allows students to experiment with stripping wires, managing a parallel file system, reimagining cards, and publishing the deck’s program.”
Price vs. performance
Any given AI workload is likely to perform better on a GTX 980 (2048 cores on a single chip) versus 16 Jetson Nano GPUs across 16 boards, but the latter is a much better, albeit still simplified, emulation of some full scaling challenges – Supercomputing engineers on a scale deal with them on the job.
nvidia Blog post Refers to the idea of upgrading the existing 16-board system with Jetson Orin Nano hardware. The performance boost of any would be a huge leap. As mentioned earlier, Orin Nano offers six Cortex A-78AE CPU cores clocked at 1.5GHz and 512A with 16 threaded cores. The Jetson Nano is a comparable shrimp with 4x Cortex-A57 CPUs and 128 cores from Maxwell. The Orin Nano is more expensive than the Jetson Nano, priced at $199 versus $129.
The performance improvement of the Orin Nano should be much greater than the increase in the price, but we hope that Nvidia will bring the lower-cost Orin price to market in this area. The $129 Orin Nano with 256 amp cores and, say, eight motorized cores, would still be a huge upgrade.
At the same time, Nvidia has no good reason to cut prices. For now, the Jetson Nano is only competing with itself. While there are some other ARM-based motherboards that are compatible with accelerators, the Jetson Nano GPU is the only product in its price category and kind.
Students will show off their small group at the SC22 Supercomputing Conference in Dallas. this year, SC22 It runs from November 13-18.
#Students #build #DIY #supercomputer #Nvidia #Jetson #Nanos