8 8
qblueRed42

nvidia jetson nano

Recommended Posts

10 hours ago, TonyMac32 said:

I noticed that too (got mine in the mail finally).  I need to take a closer look, I don't know that they added any of the voltage hopping hardware on the board, which would make C an especially bad idea, not that micro USB wasn't just insanely inappropriate...

 

I think they did the MicroUSB just to get the board up and running...

 

low power mode == nvpmodel = 1

 

Full force - expect at least a 20W draw, which means 4A on the barrel to cover things when the GPU is fully engaged - whether Blender like above at the request of @NicoD or AI models...

 

My 5Vdc - 5A adapter is similar to a laptop power supply... I commented earlier on the GPIO expansion - and there one can push 6A into the board across things, but that's going to be more than just bodge wire...

Share this post


Link to post
Share on other sites
On 5/4/2019 at 5:10 PM, sfx2000 said:

Proper power - 5A/5VDC over the barrel... @tkaiser - opened an issue to report.

 

https://github.com/ThomasKaiser/sbc-bench

 

http://ix.io/1I4j

 

Again, general compute is not where nVidia is focused on, but there, as a big ARM, it does ok...

 

@tkaiser -- he updated the SBC bench page with my Jetson Nano report, along with a few other updates...

 

https://github.com/ThomasKaiser/sbc-bench/blob/master/Results.md

 

I've share a google worksheet with results from his page that covers the results he has recorded...

 

https://docs.google.com/spreadsheets/d/1RERLs9QBfBZ3l0crd6zDtb8zLJmwoeRPLQtw1yBBbzY/edit?usp=sharing

Share this post


Link to post
Share on other sites
43 minutes ago, sfx2000 said:

and there one can push 6A into the board across things

Assuming you have good Dupont connectors, yes.

 

45 minutes ago, sfx2000 said:

My 5Vdc - 5A adapter is similar to a laptop power supply..

I use a 5V 14 A chassis supply adjusted to 5.25 for my mess, I think I've got the right barrel jack on it.  Been reworking some floating ground pins on a TRRS jack...  :lol:

Share this post


Link to post
Share on other sites
10 hours ago, NicoD said:

If so, close the window, en click render and tell me the resulting time please.

 

Blender on Jetson Nano as requested - with Cuda, there's no other boards other than nVidia

 

Time 01:11:37.77 | Mem:140.48M, Peak 142.73M

 

I would expect Shield TV Box to do better, along with more recent tegra's that support Pascal and later GPU's.

 

Share this post


Link to post
Share on other sites
On 3/24/2019 at 7:43 AM, NicoD said:

Now the heavy work has been done by the small boys, the big ones come out to take things over :)
I'll need to earn a bit more money if I want to be able to buy it all....

 

Jetson Nano plus the Google TPU accelerator - it's a good combo based on models for that task...

 

They play well together - but this is likely outside of the general compute tasks that folks are mainly interested in over here on Armbian...

 

Coral-USB-Accelerator.jpg.2cd5659f129c6e10a45191338171961f.jpg

 

Share this post


Link to post
Share on other sites
6 hours ago, sfx2000 said:

 

Time 01:11:37.77 | Mem:140.48M, Peak 142.73M

 

I would expect Shield TV Box to do better, along with more recent tegra's that support Pascal and later GPU's.

Thank you. That's a rather ok time. But my N2 does it a lot faster with CPU alone. So I'm hapyy I made the right choice.
That's about the same time as the NanoPi M4 and the NanoPC t3+. So it isn't slow. Indeed the Tegra's should be able to do it in almost half the time.
Then it would beat the N2 its best 46m02s(Mate).
Thank you for doing this test. I learned from it. With Blender 2.80 it should be able to use gpu+cpu. So could still become a good one for that.

Share this post


Link to post
Share on other sites
On 5/5/2019 at 7:18 PM, sfx2000 said:

 

Blender on Jetson Nano as requested - with Cuda, there's no other boards other than nVidia

 

Time 01:11:37.77 | Mem:140.48M, Peak 142.73M

 

I would expect Shield TV Box to do better, along with more recent tegra's that support Pascal and later GPU's.

 

 

Looked back thru the logs - and there, Jetson was hitting thermal limits and backing down the clocks on the GPU during the Blender benchmark.... I've got excellent power, enough so that GPU will run fast enough to get throttled there - so maybe now time to find a fan to keep things a bit cooler...

 

For most use cases regarding this board, the blender test on Cuda is likely out of scope for SBC's for the most part, as this is a desktop application.

 

FWIW - along with the four hidden Cortex-A53 cores (which the nVidia ARMv8 platforms use for low power*), there's also an ARM Cortex-A9 core, which runs the audio engine, and a low-level Cortex-A7 that runs power management and the boot process... like I mentioned earlier, there's a lot happening under the hood inside Tegra, and most of it is not bad for the most part.

 

When I say Shield is going to give better numbers - mostly because the GPU is wider - 256 vs 128 on Nano, and similar with Nintendo Switch, which is a custom TX1 for that application.

 

* There's precedent for this - nVidia has done this before on earlier SoC's and post Tegra X1 chips - makes things easier from a programming perspective when dealing with the kernel and CPU space for GPU compute tasks.

Share this post


Link to post
Share on other sites
On 5/5/2019 at 7:04 PM, TonyMac32 said:

Assuming you have good Dupont connectors, yes.

 

Or...

 

LOL...

 

91xdZRESz2L._SX466_.jpg.517bb7df535b26caed0950b3aa5d7e36.jpg

Share this post


Link to post
Share on other sites
11 minutes ago, sfx2000 said:

Looked back thru the logs - and there, Jetson was hitting thermal limits and backing down the clocks on the GPU during the Blender benchmark.... I've got excellent power, enough so that GPU will run fast enough to get throttled there - so maybe now time to find a fan to keep things a bit cooler...

 

For most use cases regarding this board, the blender test on Cuda is likely out of scope for SBC's for the most part, as this is a desktop application.

 

Some insight into the thermals and throttles on the SoC and GPU load with Blender - I'm a bit busy with Jobby-Job stuff, so lower priority, but still curious as this has popped up in other forums/threads...

 

Heatsink, as handsome as it is, was beyond a tolerable thumb test there when running the blender thingy...

 

Spoiler

[92499.066572] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[92515.763741] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[96071.070082] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[96440.844398] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[96441.578263] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[96442.149107] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[96458.718751] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[100385.641517] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[100402.343238] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[104323.512122] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[104340.583626] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[108279.329854] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[108279.859863] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[112212.702834] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[112213.192071] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[116128.385345] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[120122.268882] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[124066.804941] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[124067.294174] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[130360.077815] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[130365.377915] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[130391.919238] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[130429.222648] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[130568.698365] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[130569.391453] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[130576.974703] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[130587.697200] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[130591.774186] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[130596.218115] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[130641.679917] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[130648.698702] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[130673.746026] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[130708.492132] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[139877.520948] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[139878.050993] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[143836.929104] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[143837.377572] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[143854.005833] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[147811.022242] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[151781.672138] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[151798.916726] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[155719.362588] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[155719.770277] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[155736.465137] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[159658.550405] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[159675.583086] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[161371.588937] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[161391.129331] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[161393.131147] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[161395.992820] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[163592.093699] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[163608.805714] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[167575.163508] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[167592.242280] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[171526.279881] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[171543.042570] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[175493.601077] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000
[175510.415425] tegra_soctherm 700e2000.soctherm: soctherm: trip temperature -2147483647 forced to -127000

 

 

Share this post


Link to post
Share on other sites
1 hour ago, sfx2000 said:

was beyond a tolerable thumb test there when running the blender thingy...

Time to use the fan mounting holes

Share this post


Link to post
Share on other sites
On 5/6/2019 at 3:36 AM, sfx2000 said:

 

Anyways - under full power... work in progress... CPU load is actually pretty low, so my thoughts are that this is all running on the GPU itself, because CUDA... 

 

 

it's a good idea to use tegrastats utility(for logging) or top-looks jtop from Raffaello Bonghi - it's show CPU\GPU\total power consumption, frequency and load graph for GPU too(on second tab - press 2)

Share this post


Link to post
Share on other sites
On 5/21/2019 at 1:32 AM, ayaromenok said:

it's a good idea to use tegrastats utility(for logging) or top-looks jtop from Raffaello Bonghi - it's show CPU\GPU\total power consumption, frequency and load graph for GPU too(on second tab - press 2)

 

that's a good tip - I've been really busy with work-stuff, so the Nano is on the side table for the moment...

Share this post


Link to post
Share on other sites

There is a ready image of Armbian for Nano. Anyone interested ? :)

 

Working the sound output to HDMI and analog, wired network. Full-screen video in Chromium at 1080p works without brakes. :)

Share this post


Link to post
Share on other sites
On 4/27/2019 at 3:49 PM, NicoD said:

How do you come to this conclusion? Do you have info on that?

 

nVidia doesn't expose the A53's directly, it's all under the hood - cluster migration is their strategy - and this in many ways makes sense, but one is dependent on nVidia's firmware/hardware to make the right choice to move load over... Pretty much similar to Pixel-C or Shield...

 

It's an either/or, not an "and" where all cores, big or LITTLE are available.

 

In many ways, TX1 isn't that much different than Broadcom's VC4 - it's a GPU with ARM's attached, not the other way around - and that's ok, and makes sense with the rest of the nVida platforms, where the Nano allows for small/medium/large (Nano, TX2, Xavier) and have scalable consistency in what they're trying to do.

 

Which is obviously a different approach than the China chips...

Share this post


Link to post
Share on other sites
Just now, sfx2000 said:

Which is obviously a different approach than the China chips...

 

And to that end, the current sbc-bench is going to favor the big.LITTLE chips, and may not reflect performance in other use cases.

 

Comparing Jetson Nano to other SBC's that might be doing AI oriented stuff - again, consider the ecosystem, and where nVidia is positioning this part, and that is consistency across their product stack - and this was related to me from a friend that does Jetson development - he was pretty happy that Jetson Nano was around, as this is business for him...

Share this post


Link to post
Share on other sites

Just got my Nano in the mail. @balbes150 Is there a PR or Armbian fork I can build from? Really interested of comparing the server performance between the RK3399 and the TX1 :P

Share this post


Link to post
Share on other sites
20 hours ago, Redferne said:

Just got my Nano in the mail. @balbes150 Is there a PR or Armbian fork I can build from? Really interested of comparing the server performance between the RK3399 and the TX1 

As far as I know, there is only one ready-made image of Armbian. Sources to build Armbian for Nano yet . The process of building the kernel and u-boot are very specific, I have a plan how to try to implement it, but I don't have time to do it yet. :ph34r:

Share this post


Link to post
Share on other sites

New versions of 20190621 images for Jetson Nano. In these images we use a standard procedure of initial setting as in all the other images Armbian. The primary root password is 1234.  :)

 

https://yadi.sk/d/W9SHvYXAIpFEew

 

Automatic partition expansion may work, but I didn't check it. Please note that these are test images and I did not have time to check them all. I would appreciate feedback on what works or not in these images. Importantly. If image capture is used SD card which was previously recorded in any system for Jetson Nano. It is necessary to completely erase (zero) SD card. Details can be seen in these messages.

 

https://devtalk.nvidia.com/default/topic/1052684/jetson-nano/armbian-for-jetson-nano/post/5351568/#5351568

 

By default, audio output via analog is enabled, to output audio via HDMI, you need to switch the default device to HDMI in the settings panel (in the upper right corner, click on the dynamics and select the first device in the list by default).

Share this post


Link to post
Share on other sites
21 hours ago, Igor said:

Is there any existing scripting to make u-boot/kernel/image or it has to be make?

As far as I could understand, Nano has a very complicated kernel build system and u-boot (placing the sources in a bunch of different sources). Therefore, I have not yet done build scripts. You need to pull the source code in some sane new structure (GIT). And only then you can create scripts. Or go the "stupid" way, build from ready-made binary kernel files and modules (they can be taken from the finished Ubuntu Nano image) immediately packages "deb" and use them in the image Assembly. With u-boot it will be more difficult, there is a very "stupid" breakdown by a huge number of partitions. It can also be taken as a ready-made binary file (DD image from a working SD disk). But I don't have time to do it yet. :ph34r:

Share this post


Link to post
Share on other sites
5 hours ago, balbes150 said:

placing the sources in a bunch of different sources


Yes, I went into a small research and got that far. And stopped there. Also low on time for this kind of complexity. Mainline u-boot/kernel exists, where at least building should be less complicated but its apparently not very usable ...

Not very happy to go the "stupid" way :huh:

Share this post


Link to post
Share on other sites

Thanks balbes150, nice one.

 

I get Igor's concerns going the binary way. I've tested the Yocto meta-tegra layer a few weeks ago and I got a bootable image.

 

@balbes150

 

This is the layer here: https://github.com/madisongh/meta-tegra

 

This is the yocto recipe that makes the kernel:

https://github.com/madisongh/meta-tegra/blob/master/recipes-kernel/linux/linux-tegra_4.9.bb

 

This is the defconfig file for nano (tegra210):

https://github.com/madisongh/meta-tegra/tree/master/recipes-kernel/linux/linux-tegra-4.9/tegra210

 

This is the recipe for u-boot:

https://github.com/madisongh/meta-tegra/blob/master/recipes-bsp/u-boot/u-boot-tegra_2016.07.bb

 

And that's the fw_env file:

https://github.com/madisongh/meta-tegra/tree/master/recipes-bsp/u-boot/files/tegra210

 

This is where partitioning happens:

https://github.com/madisongh/meta-tegra/blob/master/classes/image_types_tegra.bbclass

 

I know that Yocto has this mixed bash/python scripting language, but I think it's quite straight forward to understand.

 

Maybe you could use this as a guide to build the kernel and the bootloader.

Share this post


Link to post
Share on other sites
2 hours ago, haileykhysen said:

at just using the ARM version of Ubuntu server


That is just a userland (script and apps) compiled for ARM platform and just mainline kernel compiled for ARM. There is no sign of this hardware in mainstream boot loader and mainline kernel. Similar goes for some other devices. Some are supported well, but they will not boot Ubuntu server just like that. That's why Armbian exists = improved Ubuntu with ready to boot images for various hardware.

Mainlining for one ARM chip require huge (and costly) resources. Nvidia open source policy is on the close/conservative side which means time can be only wasted in higher numbers ...

Share this post


Link to post
Share on other sites

Join the conversation

You can post now and register later. If you have an account, sign in now to post with your account.
Note: Your post will require moderator approval before it will be visible.

Guest
Reply to this topic...

×   Pasted as rich text.   Restore formatting

  Only 75 emoji are allowed.

×   Your link has been automatically embedded.   Display as a link instead

×   Your previous content has been restored.   Clear editor

×   You cannot paste images directly. Upload or insert images from URL.

Loading...
8 8