NVIDIA Jetson AGX Orin Dev Kit - Unboxing, Overview, and Quick Teardown
Vložit
- čas přidán 25. 07. 2024
- NVIDIA's latest Jetson dev kit is built for the new Orin module which has a 12 core Arm Cortex-A78 CPU, and a 2048 NVIDIA Ampere GPU. Once you add 32GB of RAM, PCIe 4.0, and 10GbE, you are looking at a serious bit of hardware!
Let Me Explain T-shirt: teespring.com/gary-explains-l...
Twitter: / garyexplains
Instagram: / garyexplains
#garyexplains
2:28 now THAT's impressive! Good job, Gary, made my day :D
Can’t wait to see benchmarks of some heavy ML models to see how it compares with other Nvidia cards and Jetsons
love to see how fast this thing learn simple tasks (say sort the dogs from the cats or something like that) in a perceptron network. many thanks in advance.
You’re a nice guy, thanks!
In your full review can you show how to use the MIPI CSI Camera conectors and what the lanes mean, can I connect 16 cameras?
“I wanna find out more”
Nice video! Now I'm not owning the most expensive one anymore. But I don't think I need to the AGX Xavier already has 32 GB (at least my model) and 8 Cores are enough for me at the moment.
Btw the AGX Xavier already has a similar case, so this is not the first Jetson Device which has a nice case from nvidia.
The specification is much better than I was expecting. I assume its OS will be based on Ubuntu Linux as per the Jetson Nano. I wonder how its performance will compare to an Apple M1 Mac?
Nvidia's Orin Dev Kit is the same retail price as the Apple Mac Studio, and easier to disassemble, upgrade or repair. May the Best Benchmarks win :-)
Yes that would be interesting.
I wonder what version of TF and Torch they support and which version of Python. I have AGX Xavier and in the docs they still mention only python3.6. Also do they have proper nvidia-smi or is it still tegrastats? I like running nvtop on regular GPUs. The closer dev environment to normal GPUs the better.
Can you install PopOS on this by chance?
is it possible for you to compare the AGX Orin's machine learning capabilities to apples M1
Have you taken a brake from speed test g channel or just lots of other projects at the moment that need to be done first?
Can I ran Nvidia's Isaac Sim, IsaacGym, and Orbit simulations directly on this processing unit ?
For learning distributed computing should I go with Jetsons or a couple of raspberries?
Not sure if cuda is gonna be something I use
I have videos on how to built a cluster with Raspberry Pi boards and another video with Jetson Nanos. Either would be a good choice. It depends on your budget.
Go with the jetsons. You can do distributing computing on the CPU cores and the GPU cores easily with python and cupy.
Are you going to run some games emulation?
PLEASE DO ANTUTU BECHMARKS AND AND TRY BATOCERA LINUX ON IT
Will we see benchmarks? I'd like a ARM PC. I already use Linux as my daily driver anyway.
Can it run Crysis ?
Why are you no longer mantaining the speedtest g channel?
Yikes, that was a 16x slot - so you could throw a 30 series card on it?
If you did not need to sell your kidneys for this thing it would have been awesome....
What are your thoughts on competition in the AI accelerator market why only Nvidia is dominating this market ??
CUDA , lots of api’s use the CUDA cores. This software is not available on AMD or Intel.
@@elephantwalkersmith1533 So what's the solution why don't intel and AMD develope their software stacks ??
Can it run Python. Python is important for AI ...
LOL
Dear god, it costs more on Arrow than a 3090!
Ooof the price on this thing is almost the price of a top end GPU.
$2,000? eBay has it at $9,999. What’s up?
I like Nvidia GPUs, and their Jetson SBC's are fun and great, and I have a few of them myself. However now their prices are just ridiculous, and there are now cheaper alternatives at a fraction of the prices that Nvidia is asking. The Jetson Xavier NX was $399 USD, and went up to over $2000 for the same board. The Jetson Nano 2 GB version was initially $59, and is now around $300 for the same board on Amazon. The Jetson Nano original with 4 GB of RAM is now around the $400 mark on Amazon. This new Orin board starts around the $1900 USD mark. The Jetson Xavier AGX was initially $699, and now its around $3600 USD for that board. Nvidia are pricing themselves out of the market for many. Especially since you can find other alternative boards, which can do roughly the same as many of these Nvidia SBCs, only at a fraction of the price. Gary you should create a video that talks about alternative boards, and mention their prices as well.
NVIDIA's retail prices are still the same. Don't confuse scalper's prices with the right prices. There is certainly a stock crisis, but that is a larger industry wide issue.
@@GaryExplains I don't think Nvidia will ever make the NX, or original AGX SBCs ever again, because even their new Orin boards cost a lot more, and Nvidia makes much more money for actually selling less units. Ask yourself this why would Nvidia even want to continue to make and sell their older SBCs? Sure scalping is out in the market, but Nvidia had a number of chances to control scalpers. But chose not to do a damn thing.
@@sgodsellify Nvidia will respond to market. Edge compute is still an emerging market, and the competition is slim for inference, high speed sensor IO, and general computer vision. Who makes competitive SBC’s or SOM’s in this industry segment that can also be brought to scale on such a large MTBF?
NVIDIA will still make these. It has commercial clients who make systems based on these boards and modules and they need a supply of them. If they just stop making them then nobody will use NVIDIA's stuff ever again. According to the Jetson Product Lifecycle page the Jetson Xavier NX will be in production until Jan 2026.
I have an old Jetson TK1; it's hard to find recent software for, but a lot of the current tools still support it if you're careful. The TK1 is woefully underpowered for ML by todays standards, though.
(I have an orphanage for old development systems (PPC, Snapdragon, etc.) and I will say that Nvidia at least acknowledges the TK1 and has the latest (2015?) downloads for it still available, which cannot be said for most of these old boards I have.
Why are you not uploading vidoes on speed test G channel ? Countless phones have come with Snapdragon 8 Gen 1, how much longer man ? Seems, that channel is dead now.
The channel didn't do as well as I expected and there isn't the level of interest that I anticipated. It isn't dead but I am rethinking my strategy.