Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Nvidia's $99 Jetson Nano Is an AI Computer for DIY Enthusiasts

engadget.com

Post reply on HN
Print view View original

21–30 of 218 posts

Re: Nvidia's $99 Jetson Nano Is an AI Computer for DIY Enthusiasts

#21

Post by thisacctforreal »

epynonymous wrote: post #19

the pricepoint is good, relative to the jetson tx1/2 (299-749 usd) and xavier (1099 usd). some of my notes, there is a massive heatsink on this thing, probably for both the a57 cpu and maxwell gpu, this will make your case a bit larger than say the rpi. not sure how the a57 compares to rpi’s a53, i assume both are armv8, quad-core. the inputs seem identical to rpi 3 model b+, hdmi, ethernet, 4 usb (seems 2.0), mini u…

It would be good to know what bus the Ethernet talks on; with the Pi it's shared over the USB bus and it hampers the performance considerably.

For this reason I went with an Odroid for a small network storage system.

Edit: it has PCIe so it shouldn't be a problem.

Re: Nvidia's $99 Jetson Nano Is an AI Computer for DIY Enthusiasts

#23

Post by morganvachon »

dvhh wrote: post #18
billconan wrote: post #14

I hope to put it into a pi-top https://accounts.pi-top.com/products/pi-top/ not sure if this fits

I kind of miss the nvidia transformer line of android tablet, I am still running a TF101 as a light client.

I have a TF700T collecting dust on my workbench. It was a great machine for its time though! Probably the best looking display available on an Android tablet at its release, and amazing battery life.

Re: Nvidia's $99 Jetson Nano Is an AI Computer for DIY Enthusiasts

#25

Post by im_down_w_otp »

thisacctforreal wrote: post #21
epynonymous wrote: post #19

the pricepoint is good, relative to the jetson tx1/2 (299-749 usd) and xavier (1099 usd). some of my notes, there is a massive heatsink on this thing, probably for both the a57 cpu and maxwell gpu, this will make your case a bit larger than say the rpi. not sure how the a57 compares to rpi’s a53, i assume both are armv8, quad-core. the inputs seem identical to rpi 3 model b+, hdmi, ethernet, 4 usb (seems 2.0), mini u…

It would be good to know what bus the Ethernet talks on; with the Pi it's shared over the USB bus and it hampers the performance considerably. For this reason I went with an Odroid for a small network storage system. Edit: it has PCIe so it shouldn't be a problem.

The on-board Ethernet on the TX1 is USB3 integrated, IIRC. The system having PCIe is not necessarily a sign of how it gets used or what is using it, unfortunately.

Re: Nvidia's $99 Jetson Nano Is an AI Computer for DIY Enthusiasts

#26

Post by aseipp »

newen wrote: post #24
tuxxy wrote: post #20

No dedicated memory for the GPU?

Yeah, I couldn't find any information on it. TX2 has 8GB of GPU RAM. I don't even understand how it will work on anything reasonable without any GPU RAM..

Tegra systems use a unified memory controller for both GPU and CPU RAM. This means the GPU can access essentially all system memory as video memory and there is no meaningful distinction between the two. This has always been the case -- similarly, the Xavier Jetson, and the TX2 do not have split GPU/CPU RAM either. They all have unified controllers, so the "8GB of GPU RAM" for the TX2 you mention is also "8GB of CPU RAM".

This thing essentially looks like a chopped down version of the TX1, considering the specs are all pretty much identical. These have 4GB of memory.

Most of the time you'll be running a stripped down headless Linux 4 Tegra on this device, so something like 80% of your memory can otherwise be dedicated entirely to your application (both GPU and non-GPU compute portions) anyway.

Re: Nvidia's $99 Jetson Nano Is an AI Computer for DIY Enthusiasts

#27

Post by dheera »

yolobey wrote: post #4

Hope it'll ship with better support that the first Jetson. The one they marketed with all that AI/Machine Vision stuff and then shipped without a camera driver. This is just following Google's Edge TPU, which probably competes with a Raspberry Pi + Movidius stick. The market there is getting interesting.

This. And all of the third party camera solutions for TX2 cost $500+ when equivalent USB cameras with the same sensor cost $50. They really need to get their act together and sell some NVIDIA-sanctioned camera solutions at scale, and at price points similar to Raspberry Pi cameras.

A lot of third party carrier boards also have a complete sh_tshow of connectors. Auvidea's boards, for example, ship with a Raspberry Pi camera connector, but Raspberry Pi NoIR cameras don't have TX2 drivers, and there are hardly any other cameras that ship with that connector.

Seriously, NVIDIA: Please sell a TX2 devkit that has six non-weird 2-lane CSI connectors and some IMX290 or AR0521 or any other commonly-used robotics sensors for $100 each that plug in and "just work". It would make a lot of people happy to have something to at least start with, and pave the way for third party options to follow the same form factor, connectors, pinouts, board sizes, and so forth.

Re: Nvidia's $99 Jetson Nano Is an AI Computer for DIY Enthusiasts

#28

Post by voycey »

We are using Rapids.Ai and unfortunately this requires Pascal as a minimum Architecture - unfortunate as we have been looking at a cheap method to create proof of concepts for code we would eventually push to the cloud (functionality rather than speed for example) - this would have been ideal!

Re: Nvidia's $99 Jetson Nano Is an AI Computer for DIY Enthusiasts

#29

Post by aninteger »

mntmn wrote: post #13

Anyone know about the binary blob / firmware situation on this one? Will it run mainline Linux?

It's Nvidia, so you have to decide whether you want to give ANY money at all to a company that ships binary blobs or not (whether or not this individual product uses them). I'm sure there are product teams at Nvidia that ship "open" products but unfortunately there doesn't seem to be a way to support the good parts of a corporation while neglecting the parts that ship closed binary blobs.
Post reply on HN