Ukraine Finds Nvidia AI Chip In New Russian Missile (kyivpost.com) 56
Longtime Slashdot reader AmiMoJo shares a report from Kyiv Post: Ukraine's military intelligence (HUR) has identified an Nvidia Jetson Orin computer module inside Russia's new S-71 Monochrome air-launched cruise missile, potentially indicating the use of artificial intelligence technology in the weapon, the agency said Wednesday, Aug. 12. The finding was published in the 'Components in Weapons' section of the HUR's War&Sanctions portal, which documents foreign electronic components found in Russian weapons used against Ukraine.
The latest research identified 35 electronic components in Russian weapons, according to the HUR. HUR specialists identified the Nvidia Jetson Orin module in the S-71 Monochrome, a new Russian air-launched cruise missile. The agency said the use of the computer module may indicate the integration of artificial intelligence technologies into the missile. The HUR did not provide further details on the specific functions performed by the Nvidia module in the weapon.
The latest research identified 35 electronic components in Russian weapons, according to the HUR. HUR specialists identified the Nvidia Jetson Orin module in the S-71 Monochrome, a new Russian air-launched cruise missile. The agency said the use of the computer module may indicate the integration of artificial intelligence technologies into the missile. The HUR did not provide further details on the specific functions performed by the Nvidia module in the weapon.
Re: (Score:2)
Re: (Score:2)
Seems like it would be too slow for missiles, but whatever...
Which is exactly why you would use a relatively fast and powerful chip in a missile. A slow chip can do basic image recognition at normal speed. A fast chip can do a more advanced recognition at normal speed or more basic recognition at high speed. "AI" is just a label for a bunch of more advanced algorithms many of which could use this chip and be very useful for a missile designer. Image recognition is one of the earliest tasks that AI took on and which is done both with and without neural networks.
Re: (Score:2)
My guess is that it was probably just whatever was available. If the storage on the device survived they could figure out what actually is on it.
Re: (Score:2)
Re:Why?! (Score:5, Interesting)
They also recovered chinese PTZ camera, so probably someone just importing these from chinese old-stocks. Although, current-gen Jetsons may still be importable to china since they are definitely not in the 'AI Datacenter' class
Re: (Score:2)
Can it run Crysis?
Re: (Score:2)
This one can actually cause Crysis.
Re: (Score:2)
Why?!
Target recognition. For example a particular building in a city.
Seems most likely to be a misinformation campaign. There's plenty of ways to navigate a middle and none of them require AI.
In an electronic warfare environment using visual identification is an important capability. Note “visual” is not necessarily limited to the human visual spectrum.
Re: Why?! (Score:2, Insightful)
If you're hitting something the size of a building you don't need an Nvidia chip.
My issue with it isn't that they're using a processor, it's that this class of processor is likely overkill. So, either it's so readily available they have a warehouse full of them, or somebody wants to stir up shit. I'm guessing it's effectively the latter. It's more politically with it to raise the concern that it is to bother sourcing this component for production.
Basic image recognition sufficient for mapping navigation can
Re: (Score:2)
Re: (Score:2)
What a nonsense comment you've made.
That's the basic way that much of Russian propaganda works throw some random comments at the wall [rand.org] and hope that it stops action. The thing to do when seeing these nonsense comments is to remember that they are trying to stop you acting. Do something like donating [u24.gov.ua] or calling your politician to either complain about inadequate Nvidia export controls or alternatively to demand that they help get Ukraine air defenses to counter these missiles.
The main thing is that you can never be sure whether it's deliberate
Re: (Score:2)
Re: (Score:2)
If they've got new Nvidia chips, why aren't they targeting Zelensky's bodyguards' cell phones?
Re: (Score:2)
If you're hitting something the size of a building you don't need an Nvidia chip. My issue with it isn't that they're using a processor, it's that this class of processor is likely overkill.
I don’t think so, terrain contour matching and other similar navigational techniques are serious computer vision problems.
So, either it's so readily available they have a warehouse full of them, or somebody wants to stir up shit.
Off the shelf DJI drones have had NVIDIA GPUs for over a decade.
“Nov 2025
Leading camera drone manufacturer DJI unveiled their new NVIDIA Tegra TK1-powered “Manifold” embedded computer this week.”
https://developer.nvidia.com/b... [nvidia.com]
Re: (Score:2)
Yes, but Orin is already an older part, and is widely available to hobbyists for cheap. Building a custom solution with a supply chain to match would certainly be more cost effective, but if they're simply looking to slap some basic vision algorithms and decision-making on board with standard Linux, and their volumes aren't in the millions, Orin is perfectly fine. I would also guess that based on their supply chain restrictions, sourcing something that isn't built around readily availably commodity parts wo
Re: (Score:2)
Re: Why?! (Score:3)
Because a raspi can do the job they are presumably doing with it. There's no need for this class of hardware unless you're really incompetent.
Oh wait, we're talking about Russia...
Re: (Score:2)
If they are using Jetson Orins similar to the ones we use in our company then it has far more RAM and GPU power than any Raspi. I can image work in missiles and drones that could use all that and be impossible on a Raspi.
Re: (Score:2)
Someone else cited FLOPS per gram as the measurement where these chips excel. Grams become important in something that flies and has to balance itself.
Re: (Score:2)
Jetson is Nvidia's version of a Raspberry Pi.
What "class of hardware" do you think it is?
Re: (Score:2)
Sure, let's discuss that. You've got a handful of methods historically:
- GPS/GLONASS/Starlink - can be jammed.
- Wire - inaccurate, not suitable for OTH guided missiles
- Pigeon - inaccurate and expensive
- Radar - can be jammed
- IR/laser - line of sight/requires IR signatures
- inertia - inaccurate, expensive.
There's nothing which really compares to "AI" in capability or accuracy. It's the only option which can't be jammed directly short of active EMP and provides the required/desired accuracy.
Re: (Score:2)
Tomahawks have had visual terrain following for forty years. They even had strobe lights so they could see the ground at night. They had a whole bunch of clever software and a bunch of mainframe preprocessing to make up for the lack of processing power, and have improved their capabilities over the years as computers got better.
Now you can pop in a $200 mass produced board that's orders of magnitude better, feed it some aerial photos over USB C and say sic 'em. Of course both the Russians and Ukrainians are
Re: (Score:1)
These are likely chips used for missile guidance and thrust/fin control.
Probably NOT an AI chip (Score:4, Informative)
Re: (Score:2)
You can get this kind of basic image recognition on $20 ARM boards from China. I've got a rock3a (older, barely capable compared to whats available now) board cluster that does this kind of image recognition for my cameras. rpi3 level stuff. That it was nvidia just speaks to what is available.
Re: (Score:2)
It's an Nvidia Jetson chip, which is an embedded AI chip per Nvidia marketing. And it's designed to run AI models on it - maybe nothing sophisticated nowadays, but you can certainly use it to train and run your own image recognition models.
I believe Dave's Garage used it to train a model on what vehicles are his and what are not his, so it can send a notification when an u
Re: Probably NOT an AI chip (Score:2)
People use Orin in the previous generation of self driving cars. It has compute and inference subsystems on the chip and in the software. Difference between Jetson and DRIVE are things like canbus and a safety microcontroller, and jetson board is smaller and significantly lighter weight.
Re: (Score:2)
Quid Pro Quo (Score:1, Troll)
The US intelligence agencies determined it was not in the United State's national security interest for Nvidia's most advanced chips to be exported without control, for fear they'd fall into the wrong hands, (like the Chinese). The UAE really wanted to buy those chips so they bribed Trump [theguardian.com] and voila [reuters.com]!
Re: (Score:2)
LOL they aren't putting Blackwell into rockets.
Re: (Score:2)
can't afford to, heck probably more cost effective to resell them then to fire them
Not very cutting edge.... (Score:1)
Jetson Orin is roughly 4 year old tech that's EOL now. I'm not sure how newsworthy this is. It's a graphics chip for mobile applications, not really an AI chip.
Nothing to see here....
Re:Not very cutting edge.... (Score:5, Informative)
Jetson Orin is roughly 4 year old tech that's EOL now. I'm not sure how newsworthy this is. It's a graphics chip for mobile applications, not really an AI chip.
Really? The wiki says it's most recent 2023 version has "up to 2048-core Nvidia Ampere architecture GPU with 64 Tensor cores" and supports 4-64GB ram. That sounds pretty AI to me.
Is it cutting edge? No. But it sure seems to be more than just a graphics chip for mobile applications.
Re: (Score:3)
Re: (Score:2)
How is it not AI? It's replacing a task that formerly required human intelligence with some artifice. Such tasks have been called AI for years, on and off. In fact the very first attempt at AI (the perceptron Mark 1) was an image recognition device.
Re: (Score:2)
Well, AI relies on a lot of matrix algebraic computation; which is what DSP and GPU chips are generally optimized to do.Google's Tensor AI accelerator chips are basically made to do lots of parallel matrix integer and floating point algebra.
Re: (Score:2)
>The problem is that no one distinguishes between AI and machine learning anymore
The founding definition of "AI" was behavioral. ML is a way of doing AI. Calling it a wholly separate category creates a distinction the original terminology doesn't support. ML may be thought of as a subfield of AI, rather than separate from it.
>These chips do image processing, which is not what any sensible person would call AI, but you gotta get those clicks somehow
No, it it *you* who have a totally narrow view of what
Re: (Score:2)
Well, I guess you could call a cruise missile a "mobile" application.
Bottom line it's a Russian weapons system enabled by an American electronics component.
Re: Not very cutting edge.... (Score:4, Informative)
Orin is absolutely not EOL. Jetpack 7.2.1 was literally just released for it, and I have one with r39 firmware built by Nvidia last week.
So Nvidia can sell to Russia or North Korea? (Score:2)
Or China?
Cool!
Not the first time (Score:3)
American boards and ICs have been finding their way into Russian weapons for quite some time now.
Re: (Score:2)
American boards and ICs have been finding their way into Russian weapons for quite some time now.
Whether Russia, China, Iran, or North Korea (pick your favorite country), chips leak all the time, in smaller quantities (unless you are going to implement extensive and intrusive export checks of every package, suitcase, backpack, or pants pocket, you simply can't stop it). As long as there is advantage (usually money) to be made, someone will find a way.
Re: (Score:2)
There aren't American sanctions against Ukraine.
What a waste (Score:3, Funny)
Ukraine's military intelligence (HUR) has identified an Nvidia Jetson Orin computer module inside Russia's new S-71 Monochrome air-launched cruise missile ...
A high-end graphics company chip in a monochrome system. :-)
"Potentially" is doing a lot of work here (Score:3)
For example, given Ukraine's steadily-improving defenses, it might make sense to equip a missile with enough data acquisition and processing capacity to select targets based on real-time conditions while en route -- e.g., to shift from a primary target to a secondary one if the defenses around the primary seem to be working too well. Given all the RF jamming in play, that'd have to be done by onboard processing with no reliance on external communications.
But this is all speculation. Until they get their hands on one that's intact enough to recover the code running it, we won't really know. But one thing that's clear is that the Russians didn't install it in a munition arbitrarily: they had a reason. (Insert quote from "The Hunt for Red October" about Russians and plans.)
Re:"Potentially" is doing a lot of work here (Score:4, Interesting)
Russia has a history there. In the 70s they had missiles that could prioritise targets based on size (for attacking ships), and share target info with other missiles.
Ukrainian forests are now criss crossed by kilometres and kilometres of fibre optic cable. RF control of drones is dead due to all the jamming.
Re: (Score:1)
RF control drones are nowhere near dead. Even DJI Mavics are still used daily for recon missions, nothing beats their combination of price to video quality.
The board source is more important (Score:3)
While it's fun to point out components, I think the bigger story would lie in where the signal processing board(s) are coming from. Russia does have fairly modern in-house electronics design and manufacturing but without vendor support integration of things like Jetsons into bespoke designs is pretty difficult. I would expect that they're actually sourcing whole boards and sensors (cameras) and hacking together interfaces to their known missile control systems.
AliExpress? (Score:3)
Aha! (Score:2)