Jetson Orin Nano - 75x Faster Than A Raspberry Pi

The new Jetson Orin Nano is an embedded/edge module from NVIDIA. There is also a Dev Kit that includes a carrier board. The Orin Nano has a 1024 core Ampere GPU, 32 Tensor Cores, 6 core Cortex-A78 CPU, 8GB of RAM, and more. Compared to the original Jetson Nano is can perform certain Machine Learning tasks about 80 times faster!
---
Gary Explains GitHub: github.com/garyexplains/examples
Twitter: / garyexplains
Instagram: / garyexplains
#garyexplains

[@filleswe91]

Thanks for the explanation Gary. ♥ It's always exciting to see big leaps in hardware and software. :)

[@ryuhaneda]

Very interesting to see continued development in this space. I’ve never paid much attention to NVIDIA’s offerings before (was focused more on Raspberry Pi and Arduino), may have to note them more. Still too expensive for me to consider it (I’m still only chipping in as a hobby dev for those platforms), but I wonder if the shape of future PCs is significantly smaller.

[@amcadam26]

It's nearly £400! You have to ask at this price how it compares to a low end PC with an Nvidia 1660 GPU for around the same price

[@amcadam26]

I guess it's all about where you want to use it.

[@soragranda]

The price is more cause of how much of this are manufactured rather than the nvidia tax we usually get. Demand is good but not to the point of making produce way more.

[@ivand5699]

Well the thing is also de power draw and size. You can use this in a Switch 2 or cars but a PC with a 1660…

[@GaryExplains]

Yes it depends where you want to use it, you can't use a mini PC in a self driving car or on a factory floor robot. Also there are the GPIO pins etc. Plus what mini PC runs at 15W and has 1024 GPU cores plus tensor cores for ML?

[@densepixel]

@@GaryExplains Ah yes, I'm a chip integrator. I sell companies on selling their guided factory floors with redundancy and safety measures, for an AI system with a level of uncertainty of its next action and its current process. I tell them that these chips have more GPU cores and that is what makes it better. I tell them that having each unit being independent and not controlled by a system much larger and more informed is better, it's freedom. Clients do get mad when I integrate development hardware, but they don't know the balanced power. Trust me no person is as tall, with a better haircut, for this hot body.

[@Wol333]

If only it were affordable. At its current price I'd rather slot a RTX card in an older desktop.

[@esphilee]

The product is not for gaming. He showed the power of the product using the game. This product serve different purpose. The form factor, connectivity, power efficiency are the considering factors in its intended application. Say AI or autonomous control system for battle bots, UAV, submersible drone, and so on.

[@Wol333]

@@esphilee I didn't mention gaming, nor was gaming implied as an application of interest. For robotics, I'd rather rely on Snapdragon or an SBC with an AI accelerater. This needed to be cheaper to be appealing given the competition.

[@Mkungaa]

@@Wol333 Jetson is an SBC with AI accelerator, duh. And what competition? Coral dev board with 4 TOPS? Rock Pi with 3 TOPS? Hikey970 with 6 TOPS for $300? 4 TOPS Intel or Coral USB stick plugged into Raspberry Pi? That is why Nvidia can charge $500 for dev kit and $200 for each module, which is reasonable, when you consider there is literally no competition.

[@bman3c915]

@@Wol333 Please list those Snapdragon SoCs or AI accelerator SBCs with comparable performance to size/form factor for a better price. If you're referring to the RB5 kit, those are less than half as performant and aren't exactly cheap either. The RB6 is the only option in the same realm as an Orin but is still pricey ($2,199.00). This isn't targeted to hobbyists. Its targeted to industries where high-performance compute in an embedded/edge form-factor brings value and is priced accordingly. If you want cheap, buy a Jetson Nano or Tinker Edge T.

[@antonioferraro7972]

I think that also comparing with the raspberry pi does not make sense, the purpose of these boards are totally different...

[@Mkungaa]

$300 for modules to get the more performance than Xavier, which costs $1000. That is actually impressive. $500 for devkit is also fine, considering that you might need at most several devkits to design product, but you will mostly order modules to build final product. I would preferred if they have developed carrier board instead of dev kit, but I guess we can wait for 3rd party solutions.

[@kyleDoesCoding]

Im working on a project and was wondering what your thoughts were about this question I have. I'm building a quadruped robot and I want to run real time face recognition and a few other ai models on the board. Will the orin nano be fast enough for real time face recognition and other capabilities such as, running speech recognition and controlling motors using ai to walk around?

[@johnsmith9205]

50 FPS at 720p for Doom 3 seems a bit low, I wonder how that compares with modern Snapdragon CPU/GPU combos.

[@Shubhamrawat364]

When will we start to see some competition for Nvidia in this space, from years iam hearing that competition is coming but there doesn't seem to be anyone else, what amd, intel or startups are doing ??

[@HitAndMissLab]

what is the maximum speed of SPI and I2C interfaces on the Orin Nano?

[@DigitalDiabloUK]

Surprised at the lack of hardware encoding. I guess the industry needs less power for video encode than the ML recognition components. I use both the Nano and TX1 hardware to quietly and efficiently convert videos into HEVC and is definitely one of the best options I've found for faster than real-time encode at low power consumption.

[@davidpintous6425]

Kris Kersey has a cool Iron Man helmet video about the s/w vs h/w encoding. Was skeptical, but says he moves his code and setup over from XNX to the new dev kit in 2 hours (with faster s/w encoding in play). kzbin.info/www/bejne/h4O7hY2CgZKWgsk

[@katerineloterolondono4063]

Thanks for the video. Does it have Wifi?

[@shadowarez1337]

I'm surprised these never replaced the Nvidia Shield Tegra X1 it would have been amazing.

[@a.thales7641]

I'm sure we'll get one in a few years.

[@mza.2]

And considering how ARM is now with v9, we really need new Tegra chip

[@v.gedace1519]

Did some one tryed to mine (bit) coins on this dev kit? Hey! We a ret alking about a SBC from NVidea with TerraFlops capabilities at low power consumption (Compared to a NVidea GPU) ... So please can some one give this a shot?

[@Stopinvadingmyhardware]

How long before NVidia is able to release a SoC device as an independent device that can be incorporated into a custom PCB.

[@dwwolf4636]

Expect this in drones soon.

[@kamikazilucas]

I really hope the switch 2 docked can go up to 40 watts or something so it can take better advantage of the orin chip

[@filleswe91]

Have you tested AV1 encoding on it and compare the speeds?

[@GaryExplains]

If you watched the video (which I guess you haven't) I point out that it doesn't have hardware encoding.

[@filleswe91]

@@GaryExplains Thanks, I missed that section

[@D9ID9I]

Orin Nano NX actually does have hardware video encoding.

[@alexandrunistoroiu452]

Would be nice to test those for blender&unreal rendering

[@jacquesdurand4416]

When ARM workstation !?

[@sprocket5526]

I woudl argue is infinitely faster, since Pi's are that mystical unicorn no one seems to able to find.

[@xfndzx1]

Would these embedded devices be suitable for mining?

[@GaryExplains]

No.

[@DinoFancellu]

Would be nice to see how StableDiffusion runs on it, compared to a modern PC and an RTX card

[@lodoss118]

Can it run crysis

[@jeffg4686]

Only question I have is - what are we all gonna do when all the jobs go away?

[@gureguru4694]

behold. the switch 2.0 hardware. (if Nintendo knows what they are doing)

[@majicdude88]

Absolutely won't happen. Cost is way too high for Nintendo. Maybe it would be in the third or fourth switch when this tech is nearly a decade old first lol

[@bulletpunch9317]

If its gonna be come after 3 years, maybe.

[@qzorn4440]

Very nice information. Could you please do a comparison using the new Orange Pi 5 8GB Rockchip RK3588S 8-Core 64 ARM Mali-G610 GPU, AI accelerator NPU with up to 6 TOPS? 😎 Thank you. Amazon... $101.99

[@densepixel]

For the price of a GPU, which is later might be sold as a Nvidia shield for way less.

[@ca9inec0mic58]

it will be in the next Nintendo switch for like $500, just watch

[@densepixel]

@@ca9inec0mic58 Hope it would compete with Valve on price. If it is more expensive it better perform better considering it's a newer design vs a shrunk console chip (v8 + ampre vs Zen 2 + RDNA 2).

[@CharlesVanNoland]

I would hope it's faster than an Rpi, being as expensive as it is.

[@Originalimoc]

My AMD16C is like ~30x RPi4.

[@denvera1g1]

HOLY PHUK an actual NANO? Does this mean we'll be getting a new Nvidia Shield Pro replacing the 2015 CPU design? Edit, i'd rather them wait for TSMC 4/5nm ADA based, because really the Samsung 8nm silicon used for Orin isnt that much more efficient than the TSMC 12 and 16nm used for the Nvidia 10 and 20 series graphics cards

[@nashaeshire6534]

So the Orin Nano cost 200-300$ and 500 with dev kit. It seems the dev kit contain a PSU and the carrier board. Because the Orin carrier board is the same as the old jetson one, why not buying the Orin Nano + the old Jetson Nano dev kit? 149$ + 300$ => 450$ plus a free Jetson Nano.

[@GaryExplains]

Because I don't think they are pin compatible. In others words the Jetson Nano board is not the same as the Orin Nano one.

[@eid0eid0]

$500 for the devkit is a bad joke

[@GaryExplains]

Why?

[@tong.clement]

Too expensive. If it was $80 it would be game changing

[@GaryExplains]

$80 😂

[@tabaghdissar]

How does the performance compare to the M2 Mac mini, they're are almost the same price. WE NEED powerful ARM chips that are powerful as well as powerefficient for projects like home assistant, NAS, VPN, docker, ...

[@D9ID9I]

This could be a nice core board for a drone to hit kremlin habitants with some nice warhead.

[@TheUnofficialMaker]

and 75x more $$

[@ChicagoBob123]

$499.00 Raspberry pi 5 8GB $80 NVidia is pretty expensive for an out of pocket development kit. I am sure is faster as well.

[@kungfujesus06]

Only 500 dollars you say? Good God I wish they tried harder to bring that price down

[@GaryExplains]

You think the price is high for what you get? If you want level entry try the Jetson Nano.

[@kungfujesus06]

@@GaryExplains even that is a bit steep compared to other arm sbcs like the odroid. The Nvidia GPU is enticing but, what is the driver configuration for that? Are they using proprietary drivers or open ones? As far as I know nothing in mesa works with cuda so I imagine it's a proprietary stack. While this isn't uncommon with arm sbcs, those framerates in doom3 aren't even competitive with haswell based video chipsets. It's kind of impressive for an arm sbc but the performance isn't compelling enough for it to be much of a value proposition. The cuda support alone does offer some interesting potential use cases for this and for sure it'd be way more palatable if it were in the sub 200$ range.

[@GaryExplains]

I think you have misclassified this board in your mind. If you want a cheaper SBC with an NVIDIA GPU look into the Jetson Nano.

[@eliotcole]

$500 and only gigabit Ethernet? no.

[@psxtuneservice]

Great for $150, not for $500.

[@GaryExplains]

NVIDIA has lots of options in its range. When you place it among all the other boards and modules then I don't think $150 is a reasonable price for this.

[@Simoss13]

If Nintendo can cut some ports and AI features to lower costs, this is a great upgrade for next gen hardware

[@NizarElZarif]

The rumors that the next gen Nintendo console (switch 2 ?) would have system-level AI upscaling like DLSS. removing AI cores mean a lot of changes to the architecture, so i assume they would always be there, wether they are disabled or enabled depends on the agreements between the two (if it exist in the first place).

[@xtr.7662]

​@@NizarElZarif its gonna be better at least in terms of memory bandwith the custom t239 they r making for the next switch will be based on the higher end orin but of course much downgraded and underclocked for the next switch

[@soraaoixxthebluesky]

This board will always going to be better than the competition because it supports CUDA. At $499 vs current scalped price of Raspberry Pi, I will choose this one.

[@aaroncatolico7550]

*It's also 75x more expensive than a RPI as well.*

[@GaryExplains]

I guess you should check your math.

[@electricworld1968]

This is great performance but a modern 5 watt smartphone chip does have more performance than this embedded module. like modern snapdragon processors have amazing gpu, cpu, ai processor along with other components on very small chip size. it could replace these embedded modules I think.

[@Halfjera]

Why even compare it to the pi? They're not meant for the same things and not even in the same price range.

[@GaryExplains]

Did you watch the video?

[@GustavoMsTrashCan]

As much "generic" as this answer may be, but at that price tag....? I'd rather buy "generic" x86_64 mini pcs instead of that. Unless there is something unique it provides to developers that does not have anywhere else... ...which I suppose its more like the latter option. Then again, I'm just a hobbyist.

[@GaryExplains]

You can't put a PC in a self driving car or in a robot on a factory floor, etc.

[@GaryExplains]

Plus what PC can you get that runs in 15w and has a 1024 core GPU and has tensor cores?

[@bulletpunch9317]

@@GaryExplains do you have these two replies on your clipboard? Or is it a script where the trigger word is 'mini pc'

[@GaryExplains]

I replied to a couple of comments at the same time and used cut and paste. Why do you ask?

[@bulletpunch9317]

@@GaryExplains just was seeing it a lot.

[@theabyss5647]

And 13x as expensive.

[@GaryExplains]

Indeed, so overall a better value proposition.

[@theabyss5647]

@@GaryExplains That's hilarious and absurd.

[@GaryExplains]

@@theabyss5647 I guess you and I do math differently. 75x faster for only 13x the price seems like a bargain for. I wish I could buy 75x more food, in bulk, and only pay 13x more.

[@lazerusmfh]

Raspberry pi 4 w/ hailo.

[@Build_the_Future]

Still to slow for running, my robotic AI models

[@whothefoxcares]

A #DISPLAYPORT has got to know it can't 'transition' into #HDMI

[@frankroemer9059]

A Steam deck ist similar to that price and have alot more Power , i didnt Test but Doom shall have one Steam deck way more fps as 25 😅

[@GaryExplains]

Chalk and cheese. About the only thing they have in common is that they can run Doom.

[@ca9inec0mic58]

oh man why the hell is everyone comparing this to PCs? this is meant for a completely different use case my man

[@GaryExplains]

@Ca9ine C0mic Exactly!

[@frankroemer9059]

@@ca9inec0mic58 they compare it to an Raspberry Pi and that has also a different use case, so it ist ok for me to do the same

[@ca9inec0mic58]

@@frankroemer9059 that's because a Raspberry Pi and the Jetson have alot more in common than an x86 PC, both are based on ARM, have a small-form factor and are Single Board computers besides, the Jetson doesn't really have a competitor so comparing it to a popular arm based SBC is the most logical choice

[@VAMSHI_DHAR]

WTF 500 dollars for this Better to get apple Mac mini m2 16 core neural engine also

[@GaryExplains]

Really? Does it have GPIO pins, i2c, SPI, etc How would you connect a Mac Mini to robot in a factory?

[@JCS707]

This clearly defeats the object if it is more expensive than an actual computer. Lots of mini pc’s and second hands laptops that can run debian ect…

[@GaryExplains]

You can't put a PC in a self driving car or in a robot on a factory floor, etc.

[@GaryExplains]

Plus what PC can you get that runs in 15W and has a 1024 core GPU and has tensor cores and can do ML?

[@JCS707]

Yes for industrial grade R&D or function, the entire point of the raspberry pi was for amateurs, hobbyists, students ect to learn play and experiment. With the current RPi shortage the prices increased and getting hard to find. These prices are pushing it out of reach for the intended market of the RPi

[@GaryExplains]

It isn't intended for the same market as the Pi.

[@vitalyl1327]

how many GPIOs your second hand PC have? How many CAN interfaces? I2C? SPI?

[@tempacc9589]

Why are you reading Nvidia MSRP prices as if they're reality? Don't even have to see the rest of your review.

[@GaryExplains]

Eh? Weird comment! 🤷‍♂️

[@tempacc9589]

@@GaryExplains Weird that anyone is still reading Nvidia's MSRP out loud as if they're real prices. Open Google and look any of these parts up mate.

[@GaryExplains]

Well, a few things to note "mate". First, any shortages which have let scalpers ask for ridiculous prices is a common problem in the entire industry and not something specific to NVIDA. Second, this video will last longer than any problems due to point 1. Third, these aren't PC graphic cards. Fourth, I took your advise and went to the Arrow to buy a Jetson Nano Dev kit. It sells for $149 (i.e. the RRP) and there are 449 ready to ship today (i.e. it is in stock). So basically, yeah, I did what you suggested and NVIDIA's MSRP is correct. Over to you.

[@tempacc9589]

@@GaryExplains You're just proving my point with that price. The Jetson Nano debuted with an MSRP of 100 dollars, which to be fair it held fairly well until Covid. Now they've propped the price up while it's already EOL just to make the other boards seem closer in value. It was after this that Nvidia started their sneaky "MSRP" tactics. The Xavier NX devkit debuted for "$399" but I've never seen it at that price. The Orin Nano module will almost certainly not be availible for its MSRP of "$300" either since the Xavier NX is already unobtainable for that price. The Orin Nano pricing just looks like a very obvious marketing stunt to pretend the prices are lower to grab attention.

[@GaryExplains]

Dude, have you heard that everything is going up in price everywhere. If you call that sneaky, you should come to my local supermarket, simple things like butter have doubled in price. RRP indeed! The Jetson Nano was launched in 2019. You want no price change over four years? Also it isn't EOL, the Nano modules will be a available through Jan 2027. Get a grasp. Sounds to me like you have a problem with NVIDIA, something has upset you, but I doubt it is the price of these modules. But anyway you do you, I will do me. I am sad that you didn't watch the video because I quoted the RRP, sad for you that is, seems like a very odd thing to get upset about. But hey, this is KZbin. 🤷‍♂️

[@aftab277]

🌞 Drop your most recent emoji here!