In recent years, advances in AI have produced algorithms for everything from image recognition to instantaneous translation. But when it comes to applying these advances in the real world, we’re only just getting started. A new product from Nvidia announced today at GTC — a $99 AI computer called the Jetson Nano — should help speed that process.
The Nano is the latest in Nvidia’s line of Jetson embedded computing boards, used to provide the brains for robots and other AI-powered devices. Plug one of these into your latest creation, and it’ll be able to handle tasks like object recognition and autonomous navigation without relying on cloud processing power.
Fascinating little device that could be a great boon for the maker community.
I don’t know how well this product actually delivers in practice, however if it works as advertised the cuda/opencl support could be a huge perk over other ARM SBC that run linux yet lack GPU acceleration. Most ARM SBCs technically do have GPU acceleration, yet the proprietary hardware & drivers are limited to specific android builds, which sucks for those of us who run standard linux.
I’ll be honest, I’m skeptical about the claims of “open” in this article. I have a strong suspicion that neither the hardware nor drivers are actually as open the article would have us think. NVidia’s SBC could be just as proprietary (and therefor problematic for homebrew/mainline linux) as other ARM SBCs on the market. “Open source” probably only applies to the samples, which remain dependent upon nvidia’s proprietary drivers.
I hope I’m wrong though, I’m so tired of getting stuck with proprietary code dependencies. This could be my goto hardware if everything is genuinely open source!
Alfman,
The GPU is NViDIA’s, so it’s defintely never going to be open. I have yet to find an ARM board that is even well documented enough to write my own bootloader.
These links can help :
http://www.forlinx.net/products_detail/productId=28.html
https://github.com/gitempty/uboot-s3c6410
https://github.com/haimiange/uboot-for–s3c6410
Kochise,
Thanks.
No problem, the Samsung’s S3C6410 GPU is mostly open as well :
https://github.com/tom3q/openfimg
https://github.com/tom3q/xf86-video-openfimg
Kochise,
I would nitpick that the manufacturer’s drivers are still closed, but that guy put a lot of work into reverse engineering it to provide his own version of open source drivers.
https://github.com/tom3q/openfimg/wiki
(emphasis mine)
Congrats to him for doing what he did, but the status page reveals a lot is incomplete and the developer hasn’t touched it in many years, meaning the open source drivers realistically may never get finished. Technically maybe somebody like me could pick it up where he left off, but both those ARM SBCs you cited are very old with significantly lower specs than the ones we’re buying today.
Not to come across as overly entitled here, those boards could be helpful to others, but do you know of a more modern ARM SBC (comparable to the one in the article) with well supported open drivers? If you know of one that supports accelerated openCl, I’d be very interested in buying it for an upcoming project.
Alfman, the most open recent ARM board if either the Raspberry Pi (Broadcom has mostly opened the VideoCore driver) or the UDOO iMX.6 (Freescale has mostly opened the Vivante driver)
https://github.com/anholt/mesa/wiki/VC4
https://github.com/raspberrypi/userland
https://github.com/shacharr/videocoreiv-qpu-driver
https://github.com/etnaviv
Kochise,
I’ll need to investigate it when I have more time, but in the past I know there had been complaints from the raspberri pi community regarding broadcom’s proprietary blobs. If this has truly been permanently fixed, that’s great news to me.
This very recent link suggests it may work with the mainline kernel too (though the author didn’t indicate she did any testing or whether the kernel actually works correctly. Need more confirmation, haha).
https://www.collabora.com/news-and-blog/blog/2019/03/12/quick-hack-raspberry-pi-meets-linux-kernel-mainline/
>> The GPU is NViDIA’s, so it’s defintely never going to be open.
“NVIDIA ported the free nouveau driver to the Maxwell GPU (GM20B) of the Tegra X1.”
viton,
Did you even read what you quoted?
They ported the open driver, which is a reverse engineering effort. They did not publish the specs, let alone open their own driver. So no, it’s still not open.
Nothing has changed.
Yes, but the driver is open source. That’s enough, as well as a fact, that nVidia is not terrorising the developers (like ARM or Imagination would do)
viton,
But that isn’t what is being discussed.