Nvidia Unveils Jetson Nano 2GB, a Single Board Computer (zdnet.com) 35
Nvidia has debuted the Jetson Nano 2GB, a new developer kit for students and hobbyists with an interest in robotics. ZDNet reports: The Jetson Nano 2GB is geared towards robotics enthusiasts, students, and educators that want to enter the field of artificial intelligence (AI) and robotics. Nvidia says the entry-level Jetson Nano 2GB has been priced at $59 -- including online tutorials and certification -- to "make AI easily accessible for all." The Jetson Nano 2GB is a small package with a punch: not only supported by the Nvidia JetPack software development kit (SDK), the device also comes with Nvidia container runtime and a full Linux environment suitable for software development.
In addition, the Jetson Nano 2GB is powered by CUDA-X, a collection of libraries and tools designed to support AI-based features, data processing, machine learning (ML), and deployment. Nvidia says that this combination "allows developers to package their applications for Jetson with all its dependencies into a single container that is designed to work in any deployment." Free online training and certification are on offer, alongside open source projects, tutorials, and how-tos already contributed by thousands of Jetson developers. It's currently available for pre-order, but orders won't start shipping until the end of the month.
In addition, the Jetson Nano 2GB is powered by CUDA-X, a collection of libraries and tools designed to support AI-based features, data processing, machine learning (ML), and deployment. Nvidia says that this combination "allows developers to package their applications for Jetson with all its dependencies into a single container that is designed to work in any deployment." Free online training and certification are on offer, alongside open source projects, tutorials, and how-tos already contributed by thousands of Jetson developers. It's currently available for pre-order, but orders won't start shipping until the end of the month.
Re: (Score:1, Informative)
This repeated post proves you are unworthy. Please kill yourself as soon as possible so that the white race won't be marred by fucktards like yourself. SO much work to do.
No NVDLA support (Score:1)
no nVidia Deep Learning Accelerator support, so this won't be terribly useful for the stated ML goal.
Takyon says, (Score:1)
It will run emulators/games better than Raspberry Pi 4. The GPU is the only real selling point, and that 4x Cortex-A57 is getting old. A72 ranges from 16-50% better [tomshardware.com] at the same frequency which might not matter if the GPU is the bottleneck, and the Nintendo Switch also uses A57, at a lower frequency.
Cutting the RAM in half to 2 GiB isn't going to be good for some machine learning workloads. If you look at the RK3399Pro, that's an ARM chip with a 4 GiB memory limit, but the SoC's NPU can use a separate bank of up to 4 GiB.
Here's a paper that shows the Jetson Nano crushing Raspberry Pi 4 using a Convolutional Neural Network model:
10.1109/HORA49412.2020.9152915
As the dataset size increases, the Jetson TX2 ($400 with 8 GiB RAM) uses up to 6.5 GB in the test, so the Nano and Pi have to tap out. The Pi takes about 5-6x longer than the Nano. There's some problems with the paper but they're Turkish computer scientists so eh, whatever.
Cross-posted at faster-than-leet speeds from that other site, because it's a good comment.
Re: (Score:2)
no nVidia Deep Learning Accelerator support, so this won't be terribly useful for the stated ML goal.
But it'll be great for building little arcade machines. :-)
Comment removed (Score:3)
Indeed (Score:4, Interesting)
It's pretty amazing, isn't it. There's another board out for I think $25 that beats the Raspberry Pi 3B, with a compatible form factor. For $59 you get AI and shit. LOL
A year or so ago I needed a minimal computer to just run a web UI (web server) and do some IO, over WiFi. I found an appropriate board with integrated WiFi for $2. Two friggin dollars.
My latest project has a touchscreen and matching case I got as a combo for $13 on Banggood.
Re: (Score:2)
Re:Indeed (Score:4, Interesting)
eventually hardware would be free and we'd just pay for power and connectivity.
-jcr
I recently bought a pi zero w and a relay board from eBay to control some Christmas lights. The extension cords are easily the most expensive part of the whole project.
Close enough.
Re: (Score:2)
$59 is free by the standards of the 1980s. The comparator is thousands of dollars, not $0.
Re: (Score:2)
The $2 board is an ESP, but I messed up. The board is more like $9, the chip is $2 on singles.
The "higher specs that a pi but much cheaper" I don't remember the name of. It was featured on ZDNet about a week ago. It looks like it might be a re-branded Pine64 at a new lower price.
Re:SIXTY DOLLARS? (Score:4, Informative)
It doesn't really stack up against the maixduino, which gives you a dual core 64 bit RISC-V CPU, plus a MAIX AI accelerator at 576bit supporting convolution kernels and can do video object recognition at 60fps.
It also has an "APU" (Audio Processor) that supports 8 mics and has its own FFT, though really the purpose of the 8 mics is for using it with the AI module.
All IO are programmable.
Under $10 for just the module, and under $24 for the dev board with all the IO, a camera, and an LCD.
Yes, we are living in the future.
Re:SIXTY DOLLARS? (Score:4, Informative)
It doesn't really stack up against the maixduino, which gives you a dual core 64 bit RISC-V CPU
The Maixduino is a great board for its price, but has no HDMI, Ethernet or CUDA (the CPU is a dual core @600 Mhz against ARM A57 @1.4 GHz, I don't know how they stack against each other). You may well not need those features but it doesn't mean they're not worth the increased price.
Re: (Score:2)
800 Mhz is what people run the maix at, usually.
Calculate what you pay for each unit of performance, however you calculate it.
For an AI accelerator, saying "CUDA" is meaningless; that is one of the things you'd use when you didn't have the hardware accelerator. It doesn't add much of anything.
Jetson Nano is just really expensive per performance unit, whichever unit you pick. And for AI systems, that is what matters.
Also, don't forget to price the cheapest CPU that is the speed you want, plus the $9 module
Entry-Level version (Score:5, Informative)
The post wasn't clear in stating that Jetson Nano is an already established platform, with a 4GB version [nvidia.com] that has been available for awhile at $99. This new model is a cheaper version ($59) with less memory and IO ports.
Imagine (Score:3)
Re: (Score:2)
A beowulf cluster of these!
Funny you should say that. A friend and I had a long conversation last week about what a physical form factor might look like for an expandable cluster of the Jetson modules. (All the I/O differences between the Jetson Nano 2 GB and Jetson Nano 4 GB are on the carrier board.) I was planning on doing some thermal modeling this weekend.
The tricky part is connectivity between nodes sufficient to make it a dense enough cluster to be worth the trouble, while maintaining a "corner of your desk" form factor. T
3A USB power connector? (Score:2)
What would it cost them to put a proper power connector on these things?
Same goes for Raspberry PI, etc. Let me use a proper power supply if I want to instead of a guaranteed-to-run-hot-and-fail-in-a-couple-of-months wall wart.
Even a couple of nice solder pads on the PCB would do and that'd be completely free for them to implement.
Raspberry Pi (Score:5, Informative)
GPIO Pins 2 (+5V) and 6 (GND). No protection, but moot if you are hooking it up to a proper power supply, right?
Re: (Score:3)
moot if you are hooking it up to a proper power supply, right?
Right.
I get that the USB C connector makes this thing a lot more idiot proof (and I'm actually OK with that) but if I'm building something that needs reliability then I want something better than a phone-charger to power it. 3A is a lot of power and phone chargers are compromised by trying to make them as tiny and fanless as possible (they only have to run for as long as it takes to charge a phone).
GPIO pins aren't the best connector for 3A but at least I can solder wires to them. Let's hope the PCB traces
Re: (Score:3)
The 4GB, $99 version was said to have a "DC barrel connector".
Bare in mind, it's a pre-production devkit. The actual Jetson module mounted on the board is presumably the bit hidden beneath that massive heatsink.
Re: (Score:3)
I get that the USB C connector makes this thing a lot more idiot proof
Not sure about the Jetson, in the Raspi case it does not. Raspi 4 has no power fuse or other protection on the USB side. In fact there is no difference between feeding it via the pins and the USB. Ditto for Raspi zero. Raspi 1-3 inclusive are different - they have protection on the USB connector, same as most proper USB devices.
Re:Raspberry Pi (Score:4, Informative)
GPIO pins aren't the best connector for 3A but at least I can solder wires to them.
You can easily do 4A through them. Not that it's relevant since it draws less than 2A during a stress test. The 3A supply rating assumes you've loaded the Raspberry Pi with the largest screen officially sold by the foundation, included a mouse and keyboard and also connected a USB thumb drive. Otherwise you get no where near the 3A rating of the powersupply. Many accessories support external power as well.
Also what are these traces things you speak of. These devices have ground and power planes :-)
Jetson Nano 2GB? (Score:2)
Jane! Stop this crazy thing!!
My Jetson Nano is collecting dust (Score:5, Interesting)
I was very enthousiastic about the Jetson Nano family as I wanted to get started with machine learning projects. But in practise I found that too many of the tutorials and much of the code that I found online didn't work on it.
I fell back on the Raspberry Pi 4. It made me really appreciate what the value of the Raspberry Pi devices really is: because everybody is using them, a lot of paths have been 'cut through the jungle' already. I knew that was valuable, but underestimated how valuable.
In the very end I shifted tactics yet again, and now do all machine learning projects using TensorFlow JS. By making my projects browser based it opened the door to all kinds of nice form factors (the Jetson Nano is very bulky), and I can re-use my code for purely online projects. My hardware platform of choice is now an $6 ESP32 that acts as a wifi hotspot webserver and can interface with hardware and sensors. To that I can connect any mobile phone or tablet frontend to do the heavy processing. This setup already comes with built-in camera, battery and touch screen too. Most importantly, web development is much more standardised, so I don't have to worry about drivers, software version compatibility, and all that jazz. Now I just fly over the jungle.
To me it now feels like the educational development boards, such as the Raspberry Pi and Jetson Nano, blinded me from seeing this pathway earlier.
Re: (Score:3)
Raspi is the right choice, but not for hardware reasons. As a SOC it sucks. The old versions are ridiculously underpowered compared to an Allwinner based SOC from the same time (f.e. Banana Pi). The newer versions are finally OK on performance, but run way too hot and the reliability of the SD card slot has gone totally down the drain.
What it does not deliver as hardware, it delivers as community and most importantly support. Most ARM SOCs on the market have support for 2-
Certification? (Score:2)
Still waiting for A76, A77-based SBCs (Score:2)
Great, $59 SBCs for maker projects are fun and all, but some of us would much rather have an A76 or A77-based SBC for tinkering around with Linux on ARM using a fairly-modern core. A72 be damned, there's too much of that out there already.
But no, instead we get Jetson Nano 2. Oh well.