Nvidia's new Jetson TX1 board could enable a new generation of autonomous drones

midian182

Posts: 9,726   +121
Staff member

People sometimes forget that Nvidia does more than just graphics cards, and the latest area that the company is expanding into involves improving the computing capabilities and AI functions of drones and robots. While this may sound a bit Skynet-like, the company has assured people that the technology will be used for good; Nvidia says it could enable drones to scan crowds to “identify suspicious activity,” or to help them “navigate through a forest for search and rescue.”

The company’s new Jetson TX1 developer kit is a credit card-sized supercomputer that will “enable a new generation of incredibly capable autonomous devices," says Deepu Tallia, vice president and general manager of the Tegra business at Nvidia. "They will navigate on their own, recognize objects and faces, and become increasingly intelligent through machine learning. It will enable developers to create industry-changing products."

Jesse Clayton, product manager at Nvidia, said that as well as having better recognition capabilities, robots and drones will also be able avoid collisions using deep learning algorithms and image processing engines on the board. The TX1, which replaces last year’s Nvidia TK1, offers 1 teraflop of performance alongside its 256 graphic cores to process images. The board could also connect to powerful cloud services for post-processing of images, Clayton said.

Nvidia’s developer kit for the TX1 includes, amongst other things, a debugger, compiler and libraries. Launch partners include the Massachusetts Institute of Technology, Kespry, Herta, Percepto and Stereolabs. The board’s full specs are as follows:

  • GPU: 1 teraflops, 256-core Maxwell architecture-based GPU offering best-in-class performance
  • CPU: 64-bit ARM A57 CPUs
  • Video: 4K video encode and decode
  • Camera: Support for 1400 megapixels/second
  • Memory: 4GB LPDDR4; 25.6 gigabits/second
  • Storage: 16GB eMMC
  • Wi-Fi/Bluetooth: 802.11ac 2x2 Bluetooth ready
  • Networking: 1GB Ethernet
  • OS Support: Linux for Tegra
  • Size: 50mm x 87mm, slightly smaller than a credit card

The developer kit will be available for preorder in the US from November 12 for $599. It will expand outside the region a few weeks later. The final TX1 module version, which ships without some major components and connectors, will be available early next year at a suggested price of $299 (in quantities of 1000 or more) from distributors around the world.

Permalink to story.

 
I can't be the only one that's secretly terrified of drones. On the one hand they can help survey farmer's crops and perform search and rescue. On the other hand groups of dozens of militarized drones can sweep across cities in tight surveillance patterns. All just to keep us safe right? If law enforcement in your town wanted to buy a dozen of these things and fly them around your city all day and night and monitor cameras from a central location would you think that was great or terrifying?
 
Are you terrified of helicopters too by any chance?

"I can't be the only one that's secretly terrified of helicopters. On the one hand they can help survey farmer's crops and perform search and rescue. On the other hand groups of dozens of militarized helicopters can sweep across cities in tight surveillance patterns. All just to keep us safe right? If law enforcement in your town wanted to buy a dozen of these things and fly them around your city all day and night and monitor cameras from a central location would you think that was great or terrifying?"

:)
 
If law enforcement in your town wanted to buy a dozen of these things and fly them around your city all day and night and monitor cameras from a central location would you think that was great or terrifying?
In my country, people would see it as target practice
 
I don't know if I like of idea of Nvidia being in drones. I'm sure they'll intentionally cripple their hardware every generation so that next generation sales increase. Anyone notice that the 780 is now behind the 280x? How does a flagship card slip so much?
 
I don't know if I like of idea of Nvidia being in drones. I'm sure they'll intentionally cripple their hardware every generation so that next generation sales increase. Anyone notice that the 780 is now behind the 280x? How does a flagship card slip so much?
280x came out 5 months after 780 and please post your sources
 
I don't know if I like of idea of Nvidia being in drones. I'm sure they'll intentionally cripple their hardware every generation so that next generation sales increase. Anyone notice that the 780 is now behind the 280x? How does a flagship card slip so much?
1. What does commercial/industrial hardware have to do gaming graphics cards? Are you assuming that M.I.T, Stanford University, Amazon, Microsoft, various commercial and industrial concerns, and a whole raft of HPC vendors don't have enough sense to make an informed decision on the viability of the hardware they choose?
2. WRT graphics cards only, most sites have the roles reversed from the scenario you've proposed.
2a: The GTX 780 was never Nvidia's flagship card. The correct order of Kepler flagships is: GTX Titan -> GTX 780 Ti -> GTX Titan Black
3. You are dismissing the possibility that AMD have been able to improve drivers
4. The whole "Nvidia crippling Kepler performance" was grossly overstated - I say this as the current owner of two GTX 780's. There is plenty of factual analysis based on sound empirical benchmarking that put the performance loss (since regained) at 3-5%.
 
1. What does commercial/industrial hardware have to do gaming graphics cards? Are you assuming that M.I.T, Stanford University, Amazon, Microsoft, various commercial and industrial concerns, and a whole raft of HPC vendors don't have enough sense to make an informed decision on the viability of the hardware they choose?
2. WRT graphics cards only, most sites have the roles reversed from the scenario you've proposed.
2a: The GTX 780 was never Nvidia's flagship card. The correct order of Kepler flagships is: GTX Titan -> GTX 780 Ti -> GTX Titan Black
3. You are dismissing the possibility that AMD have been able to improve drivers
4. The whole "Nvidia crippling Kepler performance" was grossly overstated - I say this as the current owner of two GTX 780's. There is plenty of factual analysis based on sound empirical benchmarking that put the performance loss (since regained) at 3-5%.

Well for one, no most places don't have common sense to choose correct hardware.

The GTX Titan was never aimed at the consumer market. It had excellent double point precision and a 1k price tag for a reason. You could call it prosumer at best.

You say the 780 regained performance? I don't see it

https://www.techspot.com/review/1000-project-cars-benchmarks/page3.html

getting beaten by a 970
 
Well for one, no most places don't have common sense to choose correct hardware.
That would include basically every scientific, medical, research, and teaching organization market and the vendors that service them.
The GTX Titan was never aimed at the consumer market. It had excellent double point precision and a 1k price tag for a reason. You could call it prosumer at best.
GTX Titan was still the flagship of the 700 series. The rest is semantics and market segment. The fact that the Titan bridges markets doesn't alter the fact that it is still both a gaming and GPGPU SKU
You say the 780 regained performance? I don't see it...[ ].getting beaten by a 970
Hardly surprising. ProjectCARS is a new game. Kepler is an old arch so unsurprising that Nvidia (or anyone else for that matter) spends more resources on newer architectures. You might also note that Maxwell is much better suited to graphically intensive games, as any synthetic test will show...
67929.png

...while the same can be said of compute...
67918.png

Given that newer games in general leverage a higher level of compute, and going forward, Gameworks (inc Flex /PhysX) and tessellation (where the GTX 970 also shades the GTX 780) will also feature more prominently I really don't expect the gap to close as a trend. It never has in the past. Fermi and AMD's VLIW4 and 5 both suffered growing disparity as the vendors focused on current series and the games required more resources.
 
I don't mind where Nvidia is going with this, I am sure there are a lot of applications out there, that could benefit from such technology. However, how about developing a board that can make the skies safer.

I am talking about ADS-B, a sense and avoid technology, so that way people can fly their drones with the assurance that they will not collide with a commercial jet.
 
I don't mind where Nvidia is going with this, I am sure there are a lot of applications out there, that could benefit from such technology. However, how about developing a board that can make the skies safer.

I am talking about ADS-B, a sense and avoid technology, so that way people can fly their drones with the assurance that they will not collide with a commercial jet.
Since the dev kit is aimed at deep learning, it probably isn't beyond the realms of possibility to "teach" the drone to fully integrate and autonomously use avoidance and transponder hardware - but like any new tech, the early adopters will probably pay a hefty price.
 
Are you terrified of helicopters too by any chance?

"I can't be the only one that's secretly terrified of helicopters. On the one hand they can help survey farmer's crops and perform search and rescue. On the other hand groups of dozens of militarized helicopters can sweep across cities in tight surveillance patterns. All just to keep us safe right? If law enforcement in your town wanted to buy a dozen of these things and fly them around your city all day and night and monitor cameras from a central location would you think that was great or terrifying?"

:)
No because there aren't dozens upon dozens of them in the sky patrolling autonomously. There's a big difference between helicopters and drones.
 
I don't mind where Nvidia is going with this, I am sure there are a lot of applications out there, that could benefit from such technology. However, how about developing a board that can make the skies safer.

I am talking about ADS-B, a sense and avoid technology, so that way people can fly their drones with the assurance that they will not collide with a commercial jet.
Since the dev kit is aimed at deep learning, it probably isn't beyond the realms of possibility to "teach" the drone to fully integrate and autonomously use avoidance and transponder hardware - but like any new tech, the early adopters will probably pay a hefty price.

I hope the open source community can come out with an effective but cheap alternative. That way it would make much more affordable for those on tight budget.
 
Back