Intel this week initiated an end-of-life plan for its Movidius Neural Compute Stick, which is based on the Myriad 2 VPU. The product will be available for another year and Intel will continue to provide technical support for the device for another two years. For developers currently working on products powered by the NCS, Intel suggests developers move over to the more advanced Myriad X-based Neural Compute Stick 2.

The Intel Movidius Neural Compute Stick is based on the Myriad 2 vision compute unit (VPU), and offered up to 100 GFLOPS/1 TOPS compute capacity when it was launched in mid-2017. The USB compute stick was primarily aimed at developers of low-power products that require a compute vision and/or deep learning accelerator, with developers both prototyping on the NCS and in some cases outright using it in production devices. Since then, Intel released a considerably more powerful Myriad X VPU, which offers four times higher TOPS performance and providing a better feature-set. In addition, Intel offers a variety of vision accelerator products based on Myriad-series VPUs as well as Arria-series FPGAs available in various form-factors.

One of the key differences between the Intel NCS and the more advanced solutions like those powered by the Myriad X is that while the former uses Intel’s proprietary Movidius Neural Compute SDK, the latter rely on Intel’s distribution of OpenVINO toolkit, which is compatible with open-source OpenVINO industry-standard software packages.

With new hardware and SDKs available, and the market for AI ASICs as a whole quickly heating up with more powerful competitors, it looks like the market is winding down for the original Movidius NCS. As a result, Intel will cease taking for the hardware on October 30, 2019, and will ship all outstanding orders April 30, 2020. The company will continue to offer technical support for the part till April 30, 2021. As a result, it looks like the Movidius NCS will have a considerably shorter lifespan than the Myriad X (which was also launched in 2017).

Related Reading:

Sources: Intel, Intel

Comments Locked

24 Comments

View All Comments

  • CajunArson - Friday, May 3, 2019 - link

    The headline is majorly misleading since Intel is just having people move to their more advanced compute stick.

    I sure didn't see you say that AMD "discontinued" Bulldozer when RyZen launched.
  • HStewart - Friday, May 3, 2019 - link

    Yes this headline is extremely misleading, this product is only for developers only and really just mean they got a better version of it.

    It was probably a temporary product since Intel purchase the company and now has a much improved version that fits their needs. Even as developer for 30 years, I not sure what use this thing has.
  • dullard - Friday, May 3, 2019 - link

    Uses: Google Clips learns when to take photos depending on your location, lighting, facial expressions, etc. https://store.google.com/us/product/google_clips_s...

    FLIR does quality control with it (imagine it on manufacturing lines where you train it to spot defects): https://www.flir.com/news-center/press-releases/fl...

    Video surveillence with facial recognition: https://us.dahuasecurity.com/2018/04/12/dahua-tech...

    Any product where you want inexpensive, low power artificial intelligence. I could see this in medical devices such as cultures, microscope slide analysis, image analysis for any type of scan, etc.
  • HStewart - Friday, May 3, 2019 - link

    I understand the idea of the chip, but on a stick - maybe it helps a developer - but it sounds like eventually the product will be in a custom device - video processing and such.
  • dullard - Saturday, May 4, 2019 - link

    It says right in the article that the stock was for prototyping. Yes, many times these things will be custom into a product. But as a hardware/firmware developer myself, I find products like this invaluable to get the code working while the hardware is still in its infancy.
  • muddling - Friday, July 12, 2019 - link

    It also runs as an accelerator just plugged into low power COTS products.
  • name99 - Friday, May 3, 2019 - link

    Yes and no.
    IF you've developed a complex app that's deeply reliant on SW that's going away, it doesn't help that there's supposedly better HW now available.

    Of course churn is always part of the computer landscape. But in the past one of Intel's big selling points was that they promise long-term SW compatibility. It doesn't help their attempts to enter a new space if it seems like they're abandoning that promise in these new spaces.

    This may just be a temporary glitch. But they are trying to cover a lot of superficially similar territory with a range of apparently similar hardware (movidius, nervana, Xeons, GPUs, FPGAs). It would be reasonable for someone betting on one of these to wonder if their product will be the next one to be abandoned either as HW (phi...) or as SW. Seems like a shortsighted move in terms of the big strategy picture, especially following right after the 5G announcement --- really suggests that Intel is switching to a mindset where anything that doesn't achieve traction soon will be abandoned.
  • berry12 - Saturday, May 4, 2019 - link

    > IF you've developed a complex app that's deeply reliant on SW that's going away

    They are not. Firstly, we are talking about pretty old hardware. You are talking about a 2016 product. Four years later, that's the usual time when Intel discontinues things. Secondly, the software is compatible with the 2nd gen stick, so nothing is "going away". You are creating unnecessary FUD.

    > But in the past one of Intel's big selling points was that they promise long-term SW compatibility.

    Stick 2 is SW compatible with stick 1. That's 5+ years of SW compatibility so far. Stop with the FUD.

    > It would be reasonable for someone betting on one of these to wonder if their product will be the next one to be abandoned either as HW (phi...) or as SW.

    Movidius = 1 W. Xeons = 65 - 250 W. Nervana = 250 W - 400 W. Where are you seeing overlap? Secondly, Phi was not abandoned, they have Xeon 9200 series which is a successor. Thirdly, Phi was and still is compatible with Xeons.
  • name99 - Saturday, May 4, 2019 - link

    "Secondly, the software is compatible with the 2nd gen stick, so nothing is "going away""

    https://movidius.github.io/ncsdk/release_notes.htm...

    "This is a 2.x release of the Intel NCSDK which is not backwards compatible with the 1.x releases of the Intel NCSDK. Please look at the documentation for differences in tools and APIs."
  • berry12 - Monday, May 6, 2019 - link

    Take a look at the migration page. In the program we personally used, it was almost entirely a simple matter of function naming. The reason for the change is well understood and further diminishes your argument. Intel is moving to a unified nGraph and OneAPI framework to make everything future-proof EVEN in the case of product discontinuation. As part of that transition, the Movidius 2 took the necessary first steps to adopt nGraph.

    https://movidius.github.io/ncsdk/ncapi/c_api_migra...

Log in

Don't have an account? Sign up now