NVIDIA Announces Jetson AGX Orin With Ampere GPU + 12 x Arm Cortex-A78AE

[phoronix]

Phoronix: NVIDIA Announces Jetson AGX Orin With Ampere GPU + 12 x Arm Cortex-A78AE

NVIDIA used their virtual GTC event to announce Jetson AGX Orin as the latest addition to their Jetson family. With Jetson AGX Orin they are advertising it as "the world's smallest, most powerful and energy-efficient AI supercomputer" for small form factor and low-power environments like robotics and edge computing applications...

https://www.phoronix.com/scan.php?page=news_item&px=NVIDIA-Jetson-AGX-Orin

[pipe13]

Hmm... with 32 GB one can do serious damage. Look forward to your benchmarks upon release.

[bob l'eponge]

As usual with NVidia, the board will be 2x as powerful, but 2x as expensive compared to AGX Xavier. Since the Xavier was already prohibitively expensive, I can't imagine how many kidney you'll need to sell to buy this board.

[Aeder]

Are they going to release an updated Jetson Nano or at least update the stack to a newer ubuntu?

[danboid]

Are they going to release an updated Jetson Nano or at least update the stack to a newer ubuntu?

No, NVIDIA have announced that the Nano build of Jetpack is going to remain based upon 18.04, essentially dropping support after barely two years. I don't think I'll be buying another NVIDIA board unless they fully open source at least their ARM stack (yeah right).

[jjmcwill2003]

I feel like right now there are low cost, low performing SBCs like the Raspberry Pi 4, and then there are more expensive, high end workstation and server configs, like the HoneyComb LX2, the Gigabyte Mt Snow Workstation, and Cavium Thunder X2 systems. But really, there's nothing "mid range", no equivalent of a Ryzen 3600/5600 setup.

The NVidia Xavier NX and AGX come close to filling that mid-range gap, even though they aren't upgradable the way a standard ATX or EATX motherboard would be. (But then again neither is the HoneyComb LX2).

It's weird that despite all the hyper surrounding ARM, if I wanted to set up a box to tinker with / develop on, I have to choose between a "toy" SBC, a very expensive server/workstation system, or run Linux on top of an M1 Mac-mini via Docker or a VM. :-(

[Paradigm Shifter]

No, NVIDIA have announced that the Nano build of Jetpack is going to remain based upon 18.04, essentially dropping support after barely two years. I don't think I'll be buying another NVIDIA board unless they fully open source at least their ARM stack (yeah right).

Couldn't agree more.

nVidia's support (or lack thereof) for their ARM development community has made me say the same thing. As interested as I am in their GPU+ARM offerings, the prohibitive expense (before SARS-Cov-2 sent prices into the stratosphere, I could pop together a functional-but-not-amazing x86_64 box with nVidia GPU for the same sort of money demanded for the Xavier boards here) and absolutely appalling support just grinds my gears something chronic.

[piotrj3]

Nvidia Orin SOC supports hardware ENCODING of AV1.

That could be extremly impressive thing.

[brucethemoose]

I feel like right now there are low cost, low performing SBCs like the Raspberry Pi 4, and then there are more expensive, high end workstation and server configs, like the HoneyComb LX2, the Gigabyte Mt Snow Workstation, and Cavium Thunder X2 systems. But really, there's nothing "mid range", no equivalent of a Ryzen 3600/5600 setup.

The NVidia Xavier NX and AGX come close to filling that mid-range gap, even though they aren't upgradable the way a standard ATX or EATX motherboard would be. (But then again neither is the HoneyComb LX2).

It's weird that despite all the hyper surrounding ARM, if I wanted to set up a box to tinker with / develop on, I have to choose between a "toy" SBC, a very expensive server/workstation system, or run Linux on top of an M1 Mac-mini via Docker or a VM. :-(

With the caveat that the Nvidia SoCs cost as much as a server setup anyway.

Android phone level boards/boxes are more powerful than "toys," but it seems that they are rather uncommon and/or inaccessible.

Didn't Torvalds mention something about accessible hardware on the personal PC level being critical to taking off?