NVIDIA Unveils New Jetson Orin Nano Super Developer Kit

[GFreeman]

NVIDIA is taking the wraps off a new compact generative AI supercomputer, offering increased performance at a lower price with a software upgrade. The new NVIDIA Jetson Orin Nano Super Developer Kit, which fits in the palm of a hand, provides everyone from commercial AI developers to hobbyists and students, gains in generative AI capabilities and performance. And the price is now $249, down from $499.

Available today, it delivers as much as a 1.7x leap in generative AI inference performance, a 70% increase in performance to 67 INT8 TOPS, and a 50% increase in memory bandwidth to 102 GB/s compared with its predecessor. Whether creating LLM chatbots based on retrieval-augmented generation, building a visual AI agent, or deploying AI-based robots, the Jetson Orin Nano Super is an ideal solution to fetch.



The Gift That Keeps on Giving
The software updates available to the new Jetson Orin Nano Super will also boost generative AI performance for those who already own the Jetson Orin Nano Developer Kit.

Jetson Orin Nano Super is suited for those interested in developing skills in generative AI, robotics or computer vision. As the AI world is moving from task-specific models into foundation models, it also provides an accessible platform to transform ideas into reality.

Powerful Performance With Super for Generative AI
The enhanced performance of the Jetson Orin Nano Super delivers gains for all popular generative AI models and transformer-based computer vision.

The developer kit consists of a Jetson Orin Nano 8 GB system-on-module (SoM) and a reference carrier board, providing an ideal platform for prototyping edge AI applications.

The SoM features an NVIDIA Ampere architecture GPU with tensor cores and a 6-core Arm CPU, facilitating multiple concurrent AI application pipelines and high-performance inference. It can support up to four cameras, offering higher resolution and frame rates than previous versions.

Extensive Generative AI Software Ecosystem and Community
Generative AI is evolving quickly. The NVIDIA Jetson AI lab offers immediate support for those cutting-edge models from the open-source community and provides easy-to-use tutorials. Developers can also get extensive support from the broader Jetson community and inspiration from projects created by developers.

Jetson runs NVIDIA AI software including NVIDIA Isaac for robotics, NVIDIA Metropolis for vision AI and NVIDIA Holoscan for sensor processing. Development time can be reduced with NVIDIA Omniverse Replicator for synthetic data generation and NVIDIA TAO Toolkit for fine-tuning pretrained AI models from the NGC catalog.

Jetson ecosystem partners offer additional AI and system software, developer tools and custom software development. They can also help with cameras and other sensors, as well as carrier boards and design services for product solutions.

Boosting Jetson Orin Performance for All With Super Mode
The software updates to boost 1.7X generative AI performance will also be available to the Jetson Orin NX and Orin Nano series of systems on modules.

Existing Jetson Orin Nano Developer Kit owners can upgrade the JetPack SDK to unlock boosted performance today.

Introducing NVIDIA Jetson Orin™ Nano Super: The World’s Most Affordable Generative AI Computer

View at TechPowerUp Main Site | Source

 

[Daven]

Something like this is going into the Nintendo Switch 2.

And why on God's Green Earth, does that piece of shite CEO think anyone cares about his signature being on the box?

 

[TheLostSwede]

Something like this is going into the Nintendo Switch 2.

I hope not, that thing is using a Cortex-A78AE, which is an embedded Arm core from 2020. It's not even that power efficient, as it's not designed for mobile devices.

Arm Announces Cortex-A78AE, Mali-G78AE and Mali-C71AE Autonomous System IPs

www.anandtech.com

Besides, my understanding was that Nintendo had gone with MTK this time around, but possibly an Nvidia GPU.

 

[Daven]

I hope not, that thing is using a Cortex-A78AE, which is an embedded Arm core from 2020. It's not even that power efficient, as it's not designed for mobile devices.

Arm Announces Cortex-A78AE, Mali-G78AE and Mali-C71AE Autonomous System IPs

www.anandtech.com

Besides, my understanding was that Nintendo had gone with MTK this time around, but possibly an Nvidia GPU.

Orin is the last and latest release of the Tegra SoC series. SKUs were released between Oct 2022 and Jan 2023. This would probably be around the time Nintendo would lock-in the hardware for the Switch 2. The Orin series is the last rumor I read regarding the exact SoC.

Nintendo Switch 2 to Feature NVIDIA Ampere GPU with DLSS | TechPowerUp

But this rumor is over a year old. More recent articles talk about Nvidia DLSS and Ampere.

Alleged Nintendo Switch 2 design leaks out: 8-inch screen and 12GB LPDDR5X-7500 memory - VideoCardz.com

Since there are no later SoC's from Nvidia and Orin uses Ampere, what else could it be?

 

[Kasy21]

I am not into AI but how does something like this make sense when our modern gpus, for what I can understand about ai tops, outperform then by orders of magnitude?

 

[TheLostSwede]

Orin is the last and latest release of the Tegra SoC series. SKUs were released between Oct 2022 and Jan 2023.

And? It's still based on an old and embedded Arm core. Nvidia has always been behind the phone SoC companies when it comes to the Arm cores they use in their products.
A 7-25 W chip would be devastating for Nintendo in terms of battery life, we already know how mediocre the Steamdeck is with a 4-15 W SoC.

Since there are no later SoC's from Nvidia and Orin uses Ampere, what else could it be?

Did Microsoft and Sony take an off the shelf part from AMD? Does the Steamdeck use an off the shelf part?
It will hopefully, for Nintendo's sake, be a custom SoC made specifically for them, that offers decent battery life.

I am not into AI but how does something like this make sense when our modern gpus, for what I can understand about ai tops, outperform then by orders of magnitude?

There are multiple kinds of "AI", this is used for edge devices, for example something in your home or in a "robot" or some other smaller applications that does on-device processing.
What you're thinking of is LLM, whereas this is more for ML.

 

[bonehead123]

And why on God's Green Earth, does that piece of shite CEO think anyone cares about his signature being on the box?

Cause that makes it run faster, hahaha

Now if it included a signed jacket, I might buy one, you know, just because !

 

[Ferrum Master]

I see even deeper meaning to the last sketch Jensen talking all this to probably his dogs.

 

[igormp]

And the price is now $249, down from $499.

Now I want to return my orin nano

Available today, it delivers as much as a 1.7x leap in generative AI inference performance, a 70% increase in performance to 67 INT8 TOPS, and a 50% increase in memory bandwidth to 102 GB/s compared with its predecessor. Whether creating LLM chatbots based on retrieval-augmented generation, building a visual AI agent, or deploying AI-based robots, the Jetson Orin Nano Super is an ideal solution to fetch.

So it's just a regular orin nano with higher clocked GPU, CPU and faster memory? Nice still.

I hope not, that thing is using a Cortex-A78AE, which is an embedded Arm core from 2020. It's not even that power efficient, as it's not designed for mobile devices.

Arm Announces Cortex-A78AE, Mali-G78AE and Mali-C71AE Autonomous System IPs

www.anandtech.com

The current switch was released in 2016 (or was it 2017?) with a CPU core that came out in 2012 (A57).
The A78 is pretty much a mobile-focused core, the AE variant just adds some extra security stuff but the core design itself is still pretty much the same.

I am not into AI but how does something like this make sense when our modern gpus, for what I can understand about ai tops, outperform then by orders of magnitude?

This is not meant to replace a GPU, but rather be used in embedded/low power scenarios. I can train a model with my 3090s and use that same model for inference in my Orin without any code changes (since both have the same architecture), and deploy it with a pack of batteries anywhere.

A 7-25 W chip would be devastating for Nintendo in terms of battery life, we already know how mediocre the Steamdeck is with a 4-15 W SoC.

The current switch uses a 10~15W chip, and the original Orin is able to fit within this budget at ease while delivering way more performance.
FWIW, I have both the original Jetson Nano (with a Maxwell GPU, the same SoC of the current switch), and a Orin Nano, the performance difference is mind blowing at a similar power envelope.

 

[TheLostSwede]

The current switch uses a 10~15W chip, and the original Orin is able to fit within this budget at ease while delivering way more performance.
FWIW, I have both the original Jetson Nano (with a Maxwell GPU, the same SoC of the current switch), and a Orin Nano, the performance difference is mind blowing at a similar power envelope.

And this is why the battery life is the second biggest complaint.

 

[tvshacker]

Can it run Crysis?

 

[TheDeeGee]

... that piece of shite CEO...

Wait, what did he do to you and/or your family?

 

[FreedomEclipse]

I am not into AI but how does something like this make sense when our modern gpus, for what I can understand about ai tops, outperform then by orders of magnitude?

It could also be a way to possibly bypass export restrictions for things going to China.

 

[Vayra86]

Wait, what did he do to you and/or your family?

Everything! I can't even buy a leather jacket anymore these days. Let alone a GPU.

 

[igormp]

It could also be a way to possibly bypass export restrictions for things going to China.

Don't think that's the case at all since this device is really underpowered for stuff like training models and large-scale inferencing, and this product is not really new but just a clock bump + price cut of an existing model.

 

[Daven]

Wait, what did he do to you and/or your family?

Worship of any single individual no matter how small the gesture hurts everyone.

 

[konga]

I hope not, that thing is using a Cortex-A78AE, which is an embedded Arm core from 2020. It's not even that power efficient, as it's not designed for mobile devices.

Arm Announces Cortex-A78AE, Mali-G78AE and Mali-C71AE Autonomous System IPs

www.anandtech.com

Besides, my understanding was that Nintendo had gone with MTK this time around, but possibly an Nvidia GPU.

There have been multiple leaks that have basically confirmed that the Switch 2 will be using an Orin chip, starting with the Nvidia leak from a few years ago that confirmed they were working on an updated API for the switch 2 and a leak of an SoC model number that coincided with one of the Orin models. Everything we've heard since then has corroborated the Orin leaks. It's gonna be the 2020 Arm cores and 2020 Ampere architecture in the 2025 Switch 2. Nintendo never goes cutting edge, instead preferring tried-and-tested architectures that are cheap and readily available.

 

[trsttte]

And why on God's Green Earth, does that piece of shite CEO think anyone cares about his signature being on the box?

And on a developer device no less. I could see this on a gpu, a special edition for the fanboys, but on a devkit!?

Orin is the last and latest release of the Tegra SoC series. SKUs were released between Oct 2022 and Jan 2023. This would probably be around the time Nintendo would lock-in the hardware for the Switch 2. The Orin series is the last rumor I read regarding the exact SoC.

Nintendo Switch 2 to Feature NVIDIA Ampere GPU with DLSS | TechPowerUp

But this rumor is over a year old. More recent articles talk about Nvidia DLSS and Ampere.

Alleged Nintendo Switch 2 design leaks out: 8-inch screen and 12GB LPDDR5X-7500 memory - VideoCardz.com

Since there are no later SoC's from Nvidia and Orin uses Ampere, what else could it be?

It makes sense since they want to keep game compatibility.

It could also be a way to possibly bypass export restrictions for things going to China.

Not at these performance levels, it's still way bellow what a cut down gpu can do.