Shopping Cart

Your cart is empty.

Your cart is empty.

Intel Neural Compute Stick 2

4.5 (13) Write Review
Free shipping on orders over $29.99

$54.90

$ 26 .99 $26.99

In Stock

About this item

  • Deep learning prototyping is now available on a laptop, a single board computer or any platform with a USB port
  • Accessible and affordable—take advantage of more performance per watt and highly efficient fanless design
  • Combines the hardware-optimized performance of the newest Intel Movidius Myriad X VPU and the Intel Distribution of OpenVINO Toolkit to accelerate deep neural network-based applications


Computer vision and artificial intelligence are transforming IoT devices at the edge. Speed prototyping for your deep neural network application with the new Intel Neural Compute Stick 2 (NCS 2). Based on the Intel Movidius Myriad X VPU and supported by the Intel Distribution of OpenVINO Toolkit, the Intel NCS 2 delivers greater performance boost over the previous generation.


DougWorld
2025-08-20 12:15:51
If you are thinking about implementing a Neural Network for , e.g., image recognition, this small USB-3 device is on your critical path. I installed it on a Windows 10 and a Ubuntu 18.04 system with no problems. Intel provides detailed instructions for setting up a whole bunch of demos, which will spark your curiosity about NN.If you get more serious, you can use multiple instances of this device to spread the NN over more nodes.It also runs on Raspberry Pi 4, but I haven't had a chance to try it yet. It is a very worthy product and you will have Big Fun with it.
Bruno
2025-07-28 21:33:14
I got a more than x10 performance boost (with full YOLOv3) in Python on a Raspberry Pi 3B+.It's analyizing a motion detection push of security images 24/7 now (the CNN forward pass for a full YOLOv3 is ~548 ms, do add some CPU overhead to process bounding boxes etc). Ordered a second one for a similar task on a Pi 4.
Jonathan H.
2025-07-24 13:14:14
I love this strange hardware. I didn't hit any snags running the openvino demos on a pi, but would like to do more testing on something a little faster than the pi 3 b+ (until a carrier board comes along for the video, to reduce load on pi CPU).For now, as a tinkerer, I'm mostly just glad to have this in my tool palette and am using it to get my feet wet with AI at the edge. I can't seem to get my imagination off of it!
Customer
2025-07-15 12:50:28
Works well, a little challenging to setup and go.
Nick L
2025-05-31 16:47:39
Using with a raspberry pi 3 b+. Get about 14-15 fps on a 320x240 stream which is way more than i needed (versus 1ish fps without it). You have to use intels opencv version with their optimizations.It does require you to run intels model optimizer if you want to run custom nets but it was fairly easy to install their package.
IpslWon
2025-05-04 15:05:22
I've had this for a little over a week and I was able to convert a custom TensorFlow model into their framework, but can ONLY use it in their demo programs. Overall for being Intel, I'm more than frustrated. Some of the requirements they ask are just unacceptable if you're planning on using your machine for anything other than these products. Special versions of OpenCV (without using environments!!!) I've been able to get their pre configured "examples" that are binaries so hold no value in using this product for your own needs. Support and the forums are what you would expect from a large company like Intel: 0 help all "is it OK to close this thread". They have multiple releases but none I've found are plug and play like they suggest. Python libraries don't load. I'd like to speak on the documentation but outside of installing and running the demo I can't do anything, thereby making useless. I thought I'd purchase one of these since the Coral USBs are all on backorder.If you're in the same boat as me, do yourself a favor and just wait for your Coral sticks. You could always add more data to train your model!UPDATE: After digging though all the githubs and support request I could find I was able to get a working inference with my own custom model. In the process of finding a couple answers I found even more issues with their product and lack of notification on some major aspects of their software. From the order of the array input to changing the structure of blobs, this seems like one of those products that could only be used if you never plan on upgrading anything on the box. I was able to get some impressive improvements in my FPS 0.32 with a TFLite model vs 1.89 using this stick. I STILL wouldn't tell people to use it based on the complete lack of support.
@dace2IT
2025-04-28 18:36:12
Dace IT LLC d/b/a Sense Traffic Pulse™ is a Member of the Intel® IoT Solutions Alliance. A global ecosystem of more than 800 industry leaders, the Alliance offers its Members unique access to Intel® technology, expertise, and go-to-market support—accelerating deployment of best-in-class solutions. Intel Neural Compute Stick 2 allows developers to innovate at the edge!
David Shamma
2025-04-24 17:16:36
Works pretty fast when you get it going...just takes some work to configure correctly. It *is* pretty new as is the OpenVino toolkit...so hopefully in time more docs and tutorials will clear things up. That said, I got some detectors going and they fly on a RaspberryPi.
Colbert Philippe
2025-04-20 13:03:41
It's Intel second generation Neural Compute Stick. I'm a professional computer programmer in that area and need to be up-to-date. I bought two such devices and may buy a few more in the future. I use this Neural Computer Stick with Intel's new platforms of Intel OneAPI and Intel OpenVINO. This little Neural Compute sticks packs a lot of computing power. Don't let its small size fool you! It's packs a lot of power!
Georgy
2025-04-11 15:50:34
Entrega en tiempo y de buena calidad
Peter T.
2025-04-06 15:45:34
Habe diesen Stick gekauft um ein Rasp3 dabei zu entlasten Objekte durch eine Kamera zu erkennen.Muss sagen, es funktioniert . Man muss etwas Zeit und Geduld mitbringen um alles einzurichten, aber wenn es einmal funktioniert ist es genial. Konnte die Verarbeitungszeit auf 1/10 reduzieren und durch die Auslagerung der Berechnungen ist das Rasp nicht überlastet für andere aufgaben. Empfehle das SSD Netzwerk zusammen mit OpenCV + OpenVISION zu betreiben. Anleitung wie das einrichten funktioniert, gibt es bei Intel direkt. Schlechte Nachricht an alle YOLO v3 Fans..das geht gerade nicht wegen einem Bug in OpenCV..aber SSD ist auch nicht schlechter.
Tomsa Remus
2025-03-06 14:08:50
I bought this item for a rapsberry pi powered project and as expected it increased the yolo detection frames per second rate from 0.6 - 0.7 to almost 8 ! Great device!
天狗
2025-01-22 15:50:06
ThinkPad X220 (メモリ 16GB / SSD 500 / Ubuntu 16) の KVM (メモリ 8GB / CPU 2 コア/ HDD 40G / Ubuntu 20) から USB3 をパススルーして利用。NCS2 のプロービングが出来て、重みファイルのコンバートや Python からの呼び出し方さえ分かればやる事は簡単。Yolo を試してみたところ、1 フレームあたり CPU はで 24 秒のところが NCS2 では 2 秒といった感じ。ただ、今回の場合、世に出回っている既存の検出器(重みファイル)は、解決したい問題に適用でなかった。自前の学習ファイルか自前の画像処理アルゴリズムを検討中。よって、機械学習のアクセレータや Python から呼び出せる汎用的なコプロとして扱う方法を調査中。