Does anyone have any experience with the Nivdia Jetson Nana AI board

My company is purchasing one of these and a camera for evaluation for use in our project. I was wondering if anyone has some thoughts or experiences with this board. It seems well built and a lot of YouTubers I follow like it. If you have used one of this I would be interested to read your experiences using this system.

Reference: www.nvidia.com...

Well , looks like I am the first to comment
Thanks for showing me something I did not know about.
It uses ARM64 (RISC) as a CPU
The GPU is nice as well
The overall specs are sort of "bland"
But the price is bland as well starting at $100 or so.
Lots of applications and modules.

This review is provided by a gamer , although I understand what they are for .

Now that I know , I am getting one for a conversation starter.

a reply to: Gothmog

Quad-core ARM® Cortex®-A57 MPCore processor

4 GB 64-bit LPDDR4
1600MHz - 25.6 GB/s

16 GB eMMC 5.1 Flash

This is not bad specs for a $99 board

500 MP/sec
1x 4K @ 60 (HEVC)
2x 4K @ 30 (HEVC)
4x 1080p @ 60 (HEVC)
8x 1080p @ 30 (HEVC)

We are evaluating this board for object information collection like facial recognition. ALPR, and other evidence gathering for law enforcement applications.

I think it is a tiny price considering the regular price in our industry is a minimum of 4 times that price point and is a more general use board not designed for ai.

The 128 CUDA core GPU provides for very quick video information processing at a very cheap price point. The better models in the series are far better but I rate this as a good entry-level board for machine learning and neural networking.

a reply to: machineintelligence

That’s really cool, did they just unbox an nVidia shield and add a gig of ram and sell it as an AI board?

The one we have is excellent at processing, the new Kodi updates have issues with 4K but the box upscales beautifully and HDR is very solid. The 1080p upscaling to 4K looks almost like 4K, better than our Q8FN or Pioneer Elite can do.

a reply to: TheAMEDDDoc

It seems to be more oriented at AI functions like machine learning and visual data characterization. The one you are talking about seems more set-top TV box oriented. This one is focused more on character recognition and object recognition than anything else.

a reply to: machineintelligence

Seems like its going to be a end point board probably for automotive and security camera work where being easily replaced is as important as its specs (aka you can send out a chimp and they can swap out the faulty module)

NV have the upper end already sorted with their pci-e cards so it makes financial sense to have something in every market segment to be able to offer the full unified experience which will appeal to senior manglement.

Can remember that at one point NV put license limitations on what could be done on what as people were slapping in 1080ti's into machines and creating great rigs for such work and not paying out silly money.

Wonder with a bit of work you could get a beowulf style cluster going and hit a point where you could equal something much more expensive if just in a larger box?

a reply to: TheAMEDDDoc

That’s really cool, did they just unbox an nVidia shield and add a gig of ram and sell it as an AI board?

The nano uses a cut down version of the tegra processor.

Shield's Tegra has 4xA57 @ 1.9 GHz and 4xA53 @ 1.3 GHz with 256 GPU cores
Whereas Nano's Tegra has 4xA57 @ 1.43 GHz with 128 GPU cores.

So performance of Nano will likely be 50-75% slower than the shield in many benchmarks.

a reply to: glend

I looked at the Shield Tegra. It seems more oriented for a set-top box.

We are also evaluating the Nvidea Jetson Xavier NX for a higher-end system board: developer.nvidia.com...

a reply to: machineintelligence

Yes Shield Tegra is specifically targetted for media requirements.

NVidea has aimed the Xavier series with cuda core technology specifically at AI. But other mainstream manufacturers have stepped up. The 2019 HiSilicon Kirin 990 for example is 200% faster than 2014 K1 at GPU FP32 floating point operations. So the lead NVidea once enjoyed, is diminishing.

We will hopefully see A1 processors at a far more reasonable price in coming years.

a reply to: glend
I am working with state and federal government on this project so they do not want Huawei products in systems connected to their data networks at present. It is interesting to see their offerings though from a technical standpoint. 7nm that is awesome.

a reply to: machineintelligence

I understand, nano and xavier are best options available to you today. CPU's like Fujitsu A64GFX arm chip that doesn't need seperate gpu might be a game changer. NVidea, Intel and AMD better watch out.

a reply to: glend

When I can evaluate one of these in an SBC or MOC format and field one I would be interested.