I pretend to buy this board and it will be my first time with it. I have some questions about the performance of the board. I am working with Ros, with depth cameras, 2 LiDAR and I am using the ubunto 16.04 64 bits on a old gaming laptop to implement a vision system. I pretend to use this board instead the laptop.
Do you think i will have dependencies problems with my ros libraries or do you think i will have compatibility issues?.
The behavior of the board will be the same of my laptop if well configured?
It’s impossible to compare the Jetson TX2 with “an old laptop” because we have no idea what that laptop is.
Which CPU? How fast? How much RAM? What graphics card? Built-in to CPU or discrete? Which vendor, generation, core count, memory bus bandwidth, and so forth?
Also, it’s impossible to know whether your ROS modules will work or not. Which depth cameras? Which LiDARs? Do they use custom driver modules? Did you have to install kernel modules to make them work? Could you build everything from source?
What is the core of the vision system? Are you using the GPU on the old laptop? Will you be updating the software to use the GPU on the Jetson? Which algorithms are you using? How well versed are you in this kind of software porting?
Sorry for not be much specific in the first post. My old laptop is a gammer version that i bought in 2010. But it has an i7 720QM Processor, 8gb ram ddr3 and anNVIDIA® GeForce® GTX 260M with 1GB DDR3 VRAM.
My ROS modules are all open source and are all available on the ros repos. The liDAR and camera (kinect or other depth camera) have all the drivers needed on the ros repo so i only have to build it from source. The vision system uses cuda and hog algorithm and is available to ros kinetic (only for ubuntu 16).
Dont forget Nvidia Tx1/2 are small devices, and nothing compared to a laptop of 5-6 years old. Except the tremendous power of the cuda cores and hardware acceleration for specific types of work. Its not as good a general purpose unit as an older laptop.
An older laptop 5-6 years old would have tremendous issues encoding 2x 4k streams to h264. But Tx2 dont play 3d games.
I know that small embedded devices are not general propose. But in this case i pretend to use it to detect persons and objects with a depth camera and Lidar. The vision system uses Opencv and other librarys from ROS.
My afraid is to have some dependencie issues with some libraries, because i already build and tested the system on my labtop and everything is working well.
The GTX260M is a DX10 part – I think that means shader level 4, if I recall correctly.
It also has 112 “stream processors,” which should be compared to the 256 “cuda cores” on the TX2.
I think the modern technology and higher core count means the TX2 is actually faster and better for GPU workloads.
The i7-7200QM is a low-power Intel part with 4 cores / 8 threads and 6 MB of cache at 1.6 GHz. It’s very likely to be faster at CPU work than the quad-A57 and dual-Denver2 on the TX2.
So, if your workload is mainly GPU based, you will see an improvement on the TX2.
If you’re maxing out the CPUs on the laptop, then the TX2 will not be able to keep up.
If runtime of the laptop per battery charge is an issue, the TX2 will absolutely excel.
You helped me a lot with that.
The vision system that i want to apply in the board runs at 20-30hz on a gaming laptops, and itself requires only approx 10% of 1 CPU core.
As i said before i already have the system build in my laptop and everything uses the middleware ROS. So if the board is ROS compatibility and supports ubuntu 16.04 i hope not have any problem with the drivers and dependencies issues with some libraries(like opencv…);
USB3 has a raw throughput of 5000Mbit/s. Actual throughput is less than that. A HUB can add a bit of overhead, but basically you still want to know how much throughput total is needed, and how much throughput of individual devices. So…“it depends”.
Since your project is imagery and image recognition based, they yes, the TX2 will probably work much better in this problem domain only. For other tasks that are cpu intensive, then no.
And you might check out the JetsonHacks website (http://www.jetsonhacks.com/) for some very cool solutions to implementing ROS and OpenCV on the TX2, among many other fun things.
I think you would fare better with a realsense R200 for your project as well. About $100.
Thank you for your advice. I will test first with my laptop and then i will see if i have the results i want.
Thank you for your tip. I already saw some videos from jetsonhacks to understand better the jetson board and to see how it works.
About the realsense, i already worked with it and i did not like the range of the depth camera. It is to short. I am using a kinect v1 only for tests but i will use a sick 3vistor-t with a range of 0.5m - 7.2m.