I order this and expect it this week, but I may have gotten in over my head!! I expected to connect my Raspberry Pi camera, but as far as I can see there is no compatible connector! How can I connect the ribbon cable w/o buying an expensive base board?
I do not know, there is no CSI connector on the AGX Xavier. You can ask this question on the official NVIDIA Jetson AGX Xavier forum where a large group of developers and NVIDIA engineers share their experience.
The AGX Xavier runs from an internal eMMC. It does not have a SD card image. You can download the flashing tool SDK Manager from NVIDIA to accomplish the task. Thanks for watching!
Can u pls make a Jetson nano video guide too ? I keep going in circles with it and the online guides. I have flashed the SD card as per the instructions but no examples where installed by default nor I understand how to load them using the SDK manager tool as in the instructions for the nano it says it’s not needed
Hi, Compared to Jetson TX2, how much is the performance improvement? E.g., running a same deep learning model, such as YOLO, SSD, etc, how faster can we get w.r.t to Jetson TX2. Thanks.
If you would like a more comprehensive discussion of the architecture and performance (measured against the Jetson TX2) of the AGX Xavier, please see the link in the JetsonHacks article: wp.me/p7ZgI9-13x under "Peak Performance".
I do not have anything to share on this subject. Samples are available on the device. Please refer to the documentation or ask on the Jetson Developer forum if you need more information.
+JetsonHacks do you recommend without being influenced by Nvidias sayings that the Xavier is worth buying? How well do you think it would work if it was used in something like your MIT Racecar but outside, with a Zed Cam (once updated) do you think it will build better maps without having problems?
That's a hard question to answer, there are many tradeoffs. First, the money part. If you are comparing with a Jetson TX2 with a carrier board, then an Xavier with the developers discount is pretty close in price. That makes the Xavier a good choice. If you are talking just TX2 with the developer kit then the Xavier is 2-4x as much money (EDU discount possible). Either way, it's a bunch of coin. So there's that part. The second part is computational performance. The ZED camera in particular takes a huge number of compute cycles on the GPU, which leaves a relatively small amount of computing capacity left to do everything else (or anything clever). The Xavier wins there, it was built for that type of application. Still, it depends on how clever you are as to what you can get a TX2 to do, I don't think many people have really explored the limits. The image processing approach requires more than just one stereo camera, of course. In the real world, something like a Tesla has ~7 cameras as I recall. The TX2 bandwidth would be saturated in that type of situation (assuming that you are using all of the different data lanes available). Again, the Xavier can ingest that many quite easily and process them all in real time too. It's probably ~10x the TX2 performance in this early software release cycle. I view it as if you're buying the Xavier, you're looking to go big. You should work on the really interesting problems, not just some 2D video processing, but also sensor fusion. Add a 3D Lidar (like they do on some of the MIT RACECARs with a Velodyne Puck), or radar and multiple cameras. If you're just exploring, something like a TX2 does all the basic stuff and is a great place to start. Plus when you crash it going 35 mph it will be much less painful to watch. On the other hand, if you're looking to do *really* interesting things, go with the Xavier. It also depends on if you're doing this professionally or not. If you are doing it for a work project, go Xavier because by the time you're project is done, that will be the new norm and the next bright and shiny will come along. If you are exploring on your own, you probably don't know what you don't know and should explore to figure it out. A F1 driver doesn't start off in a F1 car to learn to drive after all. Instead they work their way up the driving circuit. You can do that with a lower power computer; you can always trade up. Hope this helps. Thanks for watching!
Hey thanks for the quick response, I appreciate it! So i've been working on a rescue robot with my team, Hokuyo is letting us borrow the smaller version of the UXM 30LX EW so the UTM but we have both for lidar processing, the Zed I understand, so I want to shoot my brains out when I use it because its always so slow and when I try to run it and RTABMAP and all the other junk I need the Jetson just fails completely. We are trying to build something like a Ros Husky Rover type of outdoor rover, my team and I have access to our high schools new maker lab and were trying to design a new base for this year, and I in terms of hardware i'm torn between using just 2 linked Jetsons together or buying 1 Xavier with the dev discount. Professionally with a different group of people we were told to run our ML models on a custom security camera with a Tx2 built in but the best the whole thing gets with the camera inputs and the models is running at 15 frames per sec on overclock and the system sometimes crashes. Suppose Stereolabs updates the zed to cuda 10 and everything, if I run the zed and rtab and go run around outside, what are the chances that my map, and say later some other stuff like a mono cam does feature tracking to 3d with the zed, will that run smoothly? Also the Xavier is the ultimate solution to having more than 1 high quality camera connected right? I'm on the border of buying the Xavier and I think for the amount of issues it solves it could prove useful. Thanks again for the response!
I wonder why they don't pair them up with TU106 to TU 102 chipsets? These singleboard Computers offer way too few cuda cores for Boinc or Folding. This unit has a graphics performance between a GT1030 and GTX1050, which is 3x slower than an RTX 2080 Ti.
Not sure what kind of power envelope you are trying to outline here. The Jetson AGX Xavier runs in 30W, the devices you describe need 100s of watts. the Jetsons are mostly meant for deploying trained models (inferencing).Thanks for watching!
Not sure I understand the question. Typically you would add a PCIe peripheral such as a disk drive. You can ask in the forums if anyone has experience with GPUs in that slot. Thanks for watching!
It might be challenging, depending on your development level skills. First, it's an ARM board so it's not compatible with PC games. Second, it's rather expensive ($1300), so you'd really have to be motivated to change over to this. On the other hand, if you're developing a product around this (like Nintendo did on the previous generation Jetsons with their Switch), there's a lot there to work with. Thanks for watching!
@@JetsonHacks my goal is to make a portable powerful pocket pc like the gdp win that is able to run aaa games via virtual machine and play all emulator at full 60fps.... I know it's a dream but basically my goal is a more powerful switch
Thanks for the lesson, though I don't know who you mean by 'we'. I figure if Sir Patrick Stewart can get away with it, so can I. Also, see the official NVIDIA video, notice how he pronounces it. Could be a reason for that. Thanks for watching! ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-BNaV2CZ7Ed0.html ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-XoWW5HiGHsg.html
In fairness to the TX1 and TX2 formats if you were to use them in practice you'd replace the carrier board with something like this: connecttech.com/product/astro-carrier-for-nvidia-jetson-tx2-tx1/ and get much smaller form factor. If you need the compute horsepower though this thing looks epic.
The developer kits are just that, deployed these products take much different form factors. Skydio and Magic Leap are great examples. However, for practical prototyping on smaller form factors (and for those with smaller budgets), it's nice not to have to spend a bunch of money for a carrier board and be able to have a rough prototype working straight away. The Astro with breakout board costs as much as the Jetson. But as you would point out, the Xavier is 4X the cost of a TX2, so a carrier board isn't nearly the burden in comparison to the TX2. The TX2 and Xavier are not the same class of computing device, the Xavier is a little monster. I'm going to use this one on Halloween to scare small children away. Thanks for watching!
interesting... this thing could be next in PC, whole system is running at 15W... but... why only demos? I'd like to see a real performance... is it still boot Ubuntu 18.04 of the SD card? Can NVMe M.2 drives be used as system drive? Did you try a simple encoding/decoding with any open source editors? They claim 4 streams of 4K@60p H.265 encoding.... true or false? nVidia support is horrible, they tell you to go to Forums... anyone compared this 8-cores chip +GPU with new Apple M1?
It sounds like you have some pretty good ideas for Jetson AGX Xavier videos. You should do some videos and answer those questions on your channel. I can answer some of your other questions: 1. Why only demos?: Received the Xavier a day before the product launch and made a video released on launch day. September, 2018.The video is entitled 'Unboxing and Demonstration', which states explicitly that this is an Unboxing and Demonstration video. The video matches the description. 2. The AGX Xavier does not have a SD card. 3. Yes, NVMe drives can be used to both boot the system and used as the roofs. 4. I don't know what the term 'open source editor' means wrt encoding/decoding You should get an Apple M1 if you think the Xavier and M1 are the same class of machine. I've heard really good things about the M1. Thanks for watching!
@@JetsonHacks thanks Jim for quick reply and answers for some of my questions! Open source editors are usually free software with unlocked features we can install and run on any platforms, here, I meant something like Openshot or Kdenlive video editors built for Ubuntu OS... or... even regular FFMPEG commands to try to run, to make videos. I don't think M1 and this chip are the "same class", I know it's ARM architecture, which raises lots of questions about software we can use, including the version of Ubuntu OS. I've heard Apple engineers did a very good job for optimizing their software and next M1X chip will blow all existing PCs... it's good... but its all proprietary software of Apple closed eco system, running same Kernel, which I don't use... and yes, since the launch of these chips nVidia is not making any effort to build a complete system... and reviewers who posted demos still don't have answers to my simple questions since the launch(sorry I didn't see all your videos)... So the idea here is to use these chips for content creation and as a regular every day computer at ~15W of power. If you have those videos already posted, pls , send me the links.
@@naturepi The Jetson AGX Xavier is a developer kit. The main purpose is to provide a platform for developers to build applications to integrate the Xavier hardware into their own designs. The JetsonHacks channel is for Jetson developers, the intent being on how to program and develop on the platform. For example, there are videos here on how to write code or pipelines to encode/decode video. Here is a Jetson Nano (~ quite a bit slower, older software) running Kdenlive: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-kbn_6ixtIpA.html If you are developing a desktop computer which integrates the AGX Xavier, you will need to benchmark it yourself to see if it fits your needs. The expectation that a 3 year old machine that runs on ~20W has the same performance capability as new machines that run at 10-20x that seems optimistic to me.
The CPU activity monitor is the System Monitor available through the launchpad. For the GPU monitor: ru-vid.com/video/%D0%B2%D0%B8%D0%B4%D0%B5%D0%BE-7HyKqm6DqnI.html
I think thats a good atitude, people know what should be clicked in Ubuntu, and so there wouldn't be another distro to take a look on, so you could update and install your known apps and known package names :)
I do not believe that the pricing has been announced (Sept, 2018). Expect pricing when the module itself becomes available with the thermal transfer plate solution later this year. You can probably get a better answer on the official NVIDIA Jetson AGX Xavier forum: devtalk.nvidia.com/default/board/326/jetson-agx-xavier/ Thanks for watching!
