Considering buying the Jetson board, and would like to seek advice from all to help make a decision, since this is not a cheap board to me.
What I have now, is a Win 7 notebook, with GeForce 610M, and have the free VS2010 with CUDA 5.0.
What I wanted to achieve with the Jetson, is mainly for learning DNN (caffe etc), and occasional hobby projects since the CV is also attractive. In my day job, i saw an opportunity in using natural language processing, which i solved the major part with clumpsy regular expressions. I envisage that that is not something like I can apply the GPU/DNN/Jetson directly, but rather an interesting research topic in itself. i.e. the goal is to whether DNN will do better than the current implementation.
My hobby included arduino and various diy electronics, so embedded devices will not scare me. Is the Jetson a sensible buy to me?
A bit of description is possible, though I don’t have the exact advice you want. This is a full linux system with the power of a desktop. You can access Jetson through windows, but Jetson itself is a different architecture and you won’t be able to load windows onto Jetson. This gives you several advantages, but if you are used to working under windows for code that runs on windows, it is a learning curve. The libraries available and running under Jetson are quite good for someone who intends to build CUDA (and a few other) apps. Jetson typically arrives with L4T R19.2 operating system (linux version for the L4T flavor of Ubuntu), you’ll need a native linux host if you want to flash and upgrade…mandatory for most recent CUDA 6.5.
Thank you very much for the information. It looks like I’ll be good in handling the board as at work i use ubuntu servers. Still doing some research and so far figured that the critical point is whether this 192 core will beat my current 48 core mobile nvidia. If, say for example, the Jetson is 10x faster, then it is very attractive when experimenting or developing…
If you only use 48 cores currently, Jetson will likely be far faster. The only remaining question is if memory in the 48 core system is dedicated and fast, or if it uses system memory shared with CPU. Most mobile platforms share memory with CPU and do not have dedicated memory like a full size PCIe video card would.