Thanks for the nice video Robin, would you recommend the new Jetson Xavier NX? I don’t know if it’s a developer’s kit as well or is just a module. What camara do you recommend for the unit you have shown? I want to do deep image recognition. Thanks 🙏
Jetson Xavier NX, I don't have one to try but it looks like they do come as a Dev Kit from NVIDIA too $399 www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-xavier-nx/ From the specs it looks pretty good. Xavier NX has 8GB RAM compare to 32GB in the AGX Xavier. A few less CUDA cores (384) than the AGX Xavier (512). Likely good value. The price difference is only $300 between the two. It is hard to know what you will need for your application. I would tend to buy the higher spec module and if it has more power than you need you can pick the lower spec for the next iteration of your project. Another module to consider is the Jetson Nano which is much lower spec but still packs a punch. I upgraded to AGX Xavier from Nano because I needed faster inference. Camera modules, USB are cheap and good and they work and you can get lots of different resolutions. The Raspberry Pi v2 camera module works with the Nano not tried to connect to with the Xavier, for Xavier you need something that supports MIPI CSI-2 which is Camera Serial Interface The RPi Camera v2 Its 8 mega pixel! Which turned out to be a pain and too much resolution and I lost cycles downsizing the image for my neural network on the Nano. There is a camera selector here and they have some pretty crazy 4K 4x camera setups which work with Xavier www.e-consystems.com/nvidia-cameras/jetson-agx-xavier-cameras/four-synchronized-4k-cameras.asp?CS_Processor=xavier
Loving your channel mate, very underrated imho. Subscribed. I would be very interested to watch some of your projects and I am sure most viewers would be interested too. I am currently learning (beginner) Python and ML. Wish me luck lol :)
nice little beast, I've never even considered the agx xavier for training, but It doesn't look too crazy to try. For the prize, for simple models, it can be a very affordable option.
I am using 2GB Jetson Nano, it is a cute device but it is easy to reach its limit. I was thinking about upgrading to Xavier NX, now I am more inclined to Xavier AGX. Still the 670 grams is not very attractive my pack back is already heavy enough 😅
The Jetson can act like an Ubuntu server and you can access over network connection using something like ssh. So yes you can treat the Jetson like a little server and connect with an M1 Mac no problem.
Prices for NVME SSD drives change a fair bit. Here is a good deal currently same brand and capacity as I used.. www.amazon.ca/dp/B07MH2P5ZD/ref=cm_sw_em_r_mt_dp_w7E2FbGRKEV87
I did some research on RAPIDS for SPARK and I noticed NVIDIA was hosting the software that made it work with JETSON devices on the commity dev page. Best of luck to you and your project, dude.
If it was just a normal PC with a GPU it would work but this is a arm Linux and NVIDIA built a bunch of tools around an Ubuntu distribution. So it might be possible but likely a fair bit of work figuring out how to get JetPack components to work in Manjaro.
Lol what is this comparison, a CPU and a GPU from 2016 to a modern GPU? Of course it's going to seem fast if your other hardware is trash. Developing on a Xavier is even more of a pain because of all the ARM incompatibilities for many programs. These devices aren't meant for training. They are meant for inference.
The laptop is 2019 MacBook Pro with a mobile intel CPU with the same TDP (power consumption) as the Xavier. These were manufactured in the same year and the CPU comparison is between these. The model training compared the same MacBook Pro with the Xavier and a server class GPU NVIDIA K80 that are still currently used in Cloud GPUs, the AWS P2.xlarge aws.amazon.com/ec2/instance-types/p2/ The K80 is slighty faster than the Xavier. Even though the K80 is a few years older. K80 is a server class GPU and has way more CUDA cores but I'd train on the Xavier before provisioning a P2.xlarge. The K80 buys very little and the Xavier has more GPU RAM if you need it. The K80 consumes 150W per GPU compared to 30W on the Xavier so Xavier is certainly the power efficient choice between these. The Xavier runs Ubuntu Linux (looks like 18.04) and the ARM support seems complete. I've not found it lacking anything for training or inference. The toolchain is pretty good. What comparison would you prefer?
@@robingrosset6941 The OP is forgetting that Xavier also contains Tensor cores and various other cores (visual and deep learning accelerators) which did not come with Volta architecture in 2016. Regardless of their name change, the CUDA principle remains the same and the lithography remains the same as Turing, the last RTX architecture, and Volta had the same 12nm fabrication process. Perhaps the OP is missing the application of the board and mistakes this as a graphics card built for gaming.
32GB of GPU RAM is huge on such as small device. You won't find that much VRAM on consumer graphics cards for regular PC's. I am drooling over that 32GB RAM for 3D mapping purposes. I'd like to build a mobile 3D scanning solution based on it. Right now I have something working on a smartphone, but this would be much more powerful and allow high resolution.
@@robingrosset6941 I think what he means by incompatibilities for certain programs would be the non-open source IDEs like Jetbrains IDEs such as Android Studio, etc. I wish Android Studio worked on ARM, but it doesn't as far as I know.
Thanks for the nice video Robin, would you recommend the new Jetson Xavier NX? I don’t know if it’s a developer’s kit as well or is just a module. What camara do you recommend for the unit you have shown? I want to do deep image recognition. Thanks 🙏
Jetson Xavier NX, I don't have one to try but it looks like they do come as a Dev Kit from NVIDIA too $399
www.nvidia.com/en-us/autonomous-machines/embedded-systems/jetson-xavier-nx/
From the specs it looks pretty good. Xavier NX has 8GB RAM compare to 32GB in the AGX Xavier. A few less CUDA cores (384) than the AGX Xavier (512). Likely good value. The price difference is only $300 between the two. It is hard to know what you will need for your application. I would tend to buy the higher spec module and if it has more power than you need you can pick the lower spec for the next iteration of your project. Another module to consider is the Jetson Nano which is much lower spec but still packs a punch. I upgraded to AGX Xavier from Nano because I needed faster inference.
Camera modules, USB are cheap and good and they work and you can get lots of different resolutions. The Raspberry Pi v2 camera module works with the Nano not tried to connect to with the Xavier, for Xavier you need something that supports MIPI CSI-2 which is Camera Serial Interface
The RPi Camera v2 Its 8 mega pixel! Which turned out to be a pain and too much resolution and I lost cycles downsizing the image for my neural network on the Nano.
There is a camera selector here and they have some pretty crazy 4K 4x camera setups which work with Xavier
www.e-consystems.com/nvidia-cameras/jetson-agx-xavier-cameras/four-synchronized-4k-cameras.asp?CS_Processor=xavier
Thank you for the great Xavier overview!
Loving your channel mate, very underrated imho. Subscribed. I would be very interested to watch some of your projects and I am sure most viewers would be interested too. I am currently learning (beginner) Python and ML. Wish me luck lol :)
nice little beast, I've never even considered the agx xavier for training, but It doesn't look too crazy to try. For the prize, for simple models, it can be a very affordable option.
It is!
I am using 2GB Jetson Nano, it is a cute device but it is easy to reach its limit. I was thinking about upgrading to Xavier NX, now I am more inclined to Xavier AGX. Still the 670 grams is not very attractive my pack back is already heavy enough 😅
Good video 👍 n well narration.
Staying safe pal .
Thanks, you too!
Is it possible some how that macbook Air M1 can be use an monitor for Jetson board ?
Please reply I am new to this field
The Jetson can act like an Ubuntu server and you can access over network connection using something like ssh. So yes you can treat the Jetson like a little server and connect with an M1 Mac no problem.
@@robingrosset6941
Thank you so much for suggestions and for prompt reply.
Hi, robin! Thanks for your nice presentation. You discuss SSD in the video. Can you please share the link and price information. Thanks
Prices for NVME SSD drives change a fair bit. Here is a good deal currently same brand and capacity as I used.. www.amazon.ca/dp/B07MH2P5ZD/ref=cm_sw_em_r_mt_dp_w7E2FbGRKEV87
@@robingrosset6941 thank U so much
Thank you much for the video. Can you run RAPIDS on the Xavier?
I did some research on RAPIDS for SPARK and I noticed NVIDIA was hosting the software that made it work with JETSON devices on the commity dev page. Best of luck to you and your project, dude.
Where did you get that Macbook skin? Looks so nice
ARM CPU's are looking better than ever these days. I'd imagine there would be a market for a proper desktop version of these CPU's with higher TDP's.
Yes for sure. I just got a Mac Mini M1 and it’s really impressive. Can’t wait to see the M2 and higher TDP Arm based CPUs .
@@robingrosset6941 can you make a video, comparing the Jetson Xavier with the M1 neural engine i think it will be interesting comparison.
can't find these easily on aliexpress :(
Thanks!!!! ☺️☺️☺️☺️☺️😘❤️❤️❤️❤️
No problem!!
Does this work with Manjaro?
If it was just a normal PC with a GPU it would work but this is a arm Linux and NVIDIA built a bunch of tools around an Ubuntu distribution. So it might be possible but likely a fair bit of work figuring out how to get JetPack components to work in Manjaro.
15 fps video? but good quality
Thanks for that. I had not noticed that the frame rate was set to automatic, not sure why it chose 15. I will up it to 24 or 30 next time.
Lol what is this comparison, a CPU and a GPU from 2016 to a modern GPU? Of course it's going to seem fast if your other hardware is trash. Developing on a Xavier is even more of a pain because of all the ARM incompatibilities for many programs.
These devices aren't meant for training. They are meant for inference.
The laptop is 2019 MacBook Pro with a mobile intel CPU with the same TDP (power consumption) as the Xavier. These were manufactured in the same year and the CPU comparison is between these.
The model training compared the same MacBook Pro with the Xavier and a server class GPU NVIDIA K80 that are still currently used in Cloud GPUs, the AWS P2.xlarge
aws.amazon.com/ec2/instance-types/p2/
The K80 is slighty faster than the Xavier. Even though the K80 is a few years older. K80 is a server class GPU and has way more CUDA cores but I'd train on the Xavier before provisioning a P2.xlarge. The K80 buys very little and the Xavier has more GPU RAM if you need it. The K80 consumes 150W per GPU compared to 30W on the Xavier so Xavier is certainly the power efficient choice between these.
The Xavier runs Ubuntu Linux (looks like 18.04) and the ARM support seems complete. I've not found it lacking anything for training or inference. The toolchain is pretty good.
What comparison would you prefer?
@@robingrosset6941
The OP is forgetting that Xavier also contains Tensor cores and various other cores (visual and deep learning accelerators) which did not come with Volta architecture in 2016. Regardless of their name change, the CUDA principle remains the same and the lithography remains the same as Turing, the last RTX architecture, and Volta had the same 12nm fabrication process. Perhaps the OP is missing the application of the board and mistakes this as a graphics card built for gaming.
32GB of GPU RAM is huge on such as small device. You won't find that much VRAM on consumer graphics cards for regular PC's. I am drooling over that 32GB RAM for 3D mapping purposes. I'd like to build a mobile 3D scanning solution based on it. Right now I have something working on a smartphone, but this would be much more powerful and allow high resolution.
@@robingrosset6941 I think what he means by incompatibilities for certain programs would be the non-open source IDEs like Jetbrains IDEs such as Android Studio, etc. I wish Android Studio worked on ARM, but it doesn't as far as I know.
You can train models just fine on it.
forums.developer.nvidia.com/t/can-i-use-jetson-agx-xavier-for-traning-purposes/77077