Meet the Jetsons

For all your Raspberry Pi's, BeagleBoard's, Parallella's, ORDROID'S, TV boxes et al.
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#1 Meet the Jetsons

Post by Dirk Broer »

Having gone from the creditcard-sized SBC's by beginning to buy Odroid-N2's one can't help looking into the Jetsons as well.
Image
It now comes clear why those future Jetsons have a green rocket ship: the were sponsored by Nvidia all the time, perhaps all their gadgets were as well -Nvidia Rosie household robot?

The cheapest Nvidia to play around with is the 2GB model of the Jetson Nano [missed chance for Nvidia: should have been called 'Elroy', but perhaps Hanna-Barbera would not concede].
The 2GB Jetson Nano is a stepped-down version of the $99 4GB model that is offered for prices as low as $59 and that is powered by a (relatively low-cost) 5V 3A USB-C adapter, such as you use with the Raspberry Pi 4. What do you get for those $59 (USA) / £66.99 (UK) [-pure theft, that price: $59 is £44.70] / €89.99 (EU) [-again pure theft, that price: $59 is €49.85]
  • A quad-core Arm Cortex-A57 CPU @1430 MHz,
  • A NVIDIA Maxwell GPU with 128 CUDA cores,
  • 2GB 64-bit LPDDR4, 1600MHz / 25.6 GB/s,
  • One USB 3.0 Type-A, 2 USB 2.0 Type-A, 1 USB 2.0 Micro-B,
  • A 40 GPIO header, 12-pin power/UART header,
  • 4-pin fan header,
  • A Gigabit Ethernet RJ45 jack,
  • A full-size HDMI 2.0 port.
And, depending on where you live, you might also get a 802.11ac wireless adapter and extension cable† (†= 'Not initially available in all regions')

The Nano 2GB kit just comes with the board and a quick start guide. What you don't get is a PSU or a SD card -you need that, the 2GB version has no onboard eMMC-Jetpack SDK and the Nvidia Deepstream SDK-, and you might need additional cooling.
The passive cooling block has mounting holes for a 40x40mm fan, but those who really want to run things cool will go for the Nvidia Nano version of the 52Pi ICE Tower Cooling Fan -or mount a 40mm to 120mm Vertical Fan Adapter Converter...

The people who want to blow $30 on cooling, instead of a Waveshare fan of a mere $5, can of course also go for the $99 model and a 5V/4A PSU.
  • A quad-core Arm Cortex-A57 CPU @1430 MHz,
  • A NVIDIA Maxwell GPU with 128 CUDA cores,
  • 4 GB 64-bit LPDDR4, 1600MHz / 25.6 GB/s,
  • Four USB 3.0, USB 2.0 Micro-B,
  • A 40 GPIO header, 12-pin power/UART header,
  • 4-pin fan header,
  • A Gigabit Ethernet RJ45 jack,
  • A full-size HDMI 2.0 port and display port 1.4
The new B02 baseboard can also be upgraded to take the Xavier NX module instead of the Nano module. This costs four times as much as the 4GB Nano module and also uses more power,
but you do get more performance for that price:
  • A six-core Nvidia Carmel CPU @1200-1900 MHz,
  • A 384-CUDA core NVIDIA Volta GPU with 48 Tensor Cores,
  • 8 GB 128-bit LPDDR4x @ 1600 MHz/51.2GB/s,
  • Four USB 3.0, USB 2.0 Micro-B,
  • A 40 GPIO header, 12-pin power/UART header,
  • 4-pin fan header,
  • A Gigabit Ethernet RJ45 jack,
  • A full-size HDMI 2.0 port and display port 1.4
And, if you buy a separate Jetson Nano 4GB or Jetson Xavier NX module, you'll also get
  • 16 GB of eMMC storage
This might explain why they are so f@#$%ing expensive, much more so than the development kit, that has a somewhat castrated module and an additional BaseBoard. In the case of the Jetson Xavier XN it even includes a much more sophisticated (M.2 NVMe capable) BaseBoard and the power supply!
Image
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#2 Re: Meet the Jetsons

Post by Dirk Broer »

If you fancied the seven-compute-module3 Turing Pi and were thrilled by the four-compute-module4 Turing Pi2, you'll go bonkers for the Jetson Mate,
Image
that can take up to four Jetson Nano and/or Jetson Xavier NX modules. For the price of the most elaborate model, that needs four full Xavier NX modules plus the carrier board you can also buy a pretty decent Threadripper system, though.
Image
Miklos M
Butterfly Whisperer / Milkweeder
Butterfly Whisperer / Milkweeder
Posts: 1452
Joined: Fri Aug 14, 2020 3:39 pm
Location: Florida, USA

#3 Re: Meet the Jetsons

Post by Miklos M »

Dirk Broer wrote: Thu Dec 17, 2020 9:33 pm If you fancied the seven-compute-module3 Turing Pi and were thrilled by the four-compute-module4 Turing Pi2, you'll go bonkers for the Jetson Mate,
Image
that can take up to four Jetson Nano and/or Jetson Xavier NX modules. For the price of the most elaborate model, that needs four full Xavier NX modules plus the carrier board you can also buy a pretty decent Threadripper system, though.
Sounds like this would be able to crunch very fast. Does it? :idea:
Image
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#4 Re: Meet the Jetsons

Post by Dirk Broer »

Very fast? That depends. Four Jetson Nano's would be 4x4=16 Cortex-A57 cores, able to crunch anything that BOINC offers, provided that it is made for ARM.
The CUDA cores would be without work, unless someone makes an app that can run CUDA/ARM -and as the projects don't cater that themselves, you'd have to make your own app, use the anonymous platform and pray that you get work.

It is even more painful when you'd have four Xavier NX modules: 4x6=24 cores, each better than the Cortex-A57 (more Cortex-A73 level), and an astonishing number of unused CUDA and TENSOR cores -those Xaviers have Volta architectures GPUs.

The various BOINC projects -or the volunteers themselves- should deliver the apps, the needed libraries are already mostly in the JetPack, Deepstream and/or other Nvidia developer software that is specially aimed at the Jetson family. If we -or the projects- can unlock the potential of the Jetson family -heck, of all ARM based SBCs with GPUs, NPUs or TPUs-, they'd be stellar performers in the ARM department.
Image
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#5 Re: Meet the Jetsons

Post by Dirk Broer »

Nvidia Jetsons for the future.

What should Nvidia do to make their ARM products more competitive? There are several strategies, but let's start with a sore point on my part: the price.
The 2GB Nano comes at a price halfway the 4GB Pi 4 and the 8GB Pi4 here, while the 4GB Nano is more expensive than the 8GB Pi 4 and even more expensive than the 4GB Odroid-N2+!
The next model, the Xavier NX comes at a whopping 500+ Euro's.....That's five Raspberry Pi 4 8GB's, or almost ten Raspberry Pi 4 4GB's -or four Odroid-N2+'s, including eMMC and power adapter.

My proposals for Nvidia:
  • Lower the price of the 2GB Nano to $35, the 4GB Nano to $55, so they can compete with the Raspberries. The USP (Unique Selling Point) might be the better GPU. CPU-wise the Raspberry Pi 4's are one generation ahead already.
  • Bring out a 8GB Nano. This to prevent messages in BOINC like this for my Jetson-Nano2GB:
    Rosetta@home 4/9/2021 4:00:11 PM Message from server: Rosetta needs 6675.72 MB RAM but only 1879.68 MB is available for use.
    Even the 4GB would get this from Ralph and Rosetta, so a 8GB would be very nice, not only to keep up with the Raspberry Pi's, but to do meaningful BOINC work.
    Amicable Numbers 4/11/2021 12:25:21 AM Message from server: Amicable Numbers up to 10^21 needs 2861.02 MB RAM but only 1879.68 MB is available for use.
    Amicable Numbers should be covered with the 4GB model though...
  • Bring out more versions of the Xavier NX, and lower the price. At this moment we only have the far-too-expensive 8GB model, make a 6GB and a 12 GB (16 GB?) model too, and make it a bit more affordable.
  • Bring OpenCL support for your GPUs to the Linux/ARM platform. CUDA works at least but for BOINC (and other applications) to really get things into gear, OpenCL support for the GPU is needed.
Edit: Rosetta have lowered their (unintended) RAM and diskspace limits. Someone seems to have mislaid a comma in the code, or something likewise.
Image
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#6 Re: Meet the Jetsons

Post by Dirk Broer »

nVidia brought out a 16GB Xavier NX module! And for those with a smaller budget who want to upgrade their 2GB or 4GB Jetson Nano developer kit there is also a TX2-NX module. Those of us who have enough money might also want to upgrade their Jetson AGX Xavier with a 64GB module, or wait for the AGX Orin.
Image
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#7 Re: Meet the Jetsons

Post by Dirk Broer »

Nvidia now also has Jetson Orin NX 8GB and Orin NX 16GB Modules, that will fit into the Jetson Nano and Jetson Xavier NX carrier boards -and into the Turing Pi 2....
Image
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#8 Re: Meet the Jetsons

Post by Dirk Broer »

The Jetson Mate and the Turing Pi v2 compared:

Boards
Image

Modules
Image
Click on pictures if you want to see a bigger version.

As you can see the Turing Pi v2 has some advantages, such as the standard form factor (any iTX case will do), the standard ATX power connector, the more versatile SOC/SOM options and the ease of connection extra storage.
The Turing PI v2 gives you the opportunity to make a 32-core cruncher with up to 128 GB of RAM (with four Turing RK-1 modules) or with 32 ARM Cortex-A78 CPUs (with four Jetson Orin NX 8-core/16 GB modules), while the Jetson Mate only has the latter choice -still a very potent 32-core 64 GB cruncher with four Nvidia Ampere GPUs.

If you want to make a multi-core ARM cruncher that uses the utmost minimum of Watts you can choose for four Jetson Nano modules, and set them to use only 5 Watt each, so you have 16 Cortex-A57 cores running for a mere 20 Watt TDP. If your power bill can stand 25 Watt you can up that to 16 Cortex-A72 cores (four Raspberry Pi CM4's) in the case of the Turing Pi v2. Even four RK-1 modules -if and when they arrive, and are as good as the claims have them- will only use 28 Watt -and that's 16 Cortex-A76 cores PLUS 16 Cortex-A55 cores, and four Mali-G610 GPUs with of as yet mainly unknown capabilities.
Image
User avatar
Dirk Broer
Corsair
Corsair
Posts: 1962
Joined: Thu Feb 20, 2014 11:24 pm
Location: Leiden, South Holland, Netherlands
Contact:

#9 Re: Meet the Jetsons

Post by Dirk Broer »

For those of us with interest in nVidias offerings in the SBC-field: they have discontinued their old Jetson Nano, Jetson TX2 NX, Jetson Xavier NX and Jetson AGX Xavier boards. The good news: They are replacing them January 2023 with Jetson Orin offerings. The one with most interest for the budget-wise cruncher are the Jetson Orin Nanos, so let's take a closer look, and compare them with the old offerings :
Image (click to make bigger)
You might need a new, separate carrier board when you want to boot the new Jetson Orin Nano modules: not all Jetson carrier boards feature the needed M.2 M-key for a NVMe SSD.
Those that have them deep pockets and were able to afford themselves a Jetson Xavier NX in the past can now rejoice a 2nd time: their carrier boards are able to work with the new Orin Nanos, but also with the Orin NX modules!
Image (click to make bigger)
Now that is some serious ARM computing power, we might even forget about the RK3588 vapourware.....
Image
Post Reply Previous topicNext topic

Return to “Single-board Computers”