L4T on Acer Chromebook 13 CB5

Will Linux4 Tegra run on Acers’s new Chromebook 13 CB5 with Tegra K1?

Also interested in this. I’m hoping to give it a try on a HP Chromebook under chroot using crouton as soon as Oct 22 arrives :)

That Acer web link makes it look like the people advertising the Chromebook don’t understand their own product. It shows CPU as Tegra K1, but Tegra is not a CPU and Keplar K1 is used with a CPU, not as a CPU.

If it was meant that Chromebook uses the Tegra124 SoC, then it is a quad core ARM15 (ARMv7) and adapting L4T should be possible…but beware that Jetson is tied to the circuit board and peripherals on it known as the PM375. Should the Chromebook really be a Tegra124, and if the firmware and/or software is available for the surrounding components, I’d think L4T would work with some effort. I guess the first thing I’d wonder is if an existing boot loader could be used, and if not, what would be required for adapting L4T’s boot loader.

But what if the Tegra component is not the Tegra124 SoC? Is it possibly something else?

It is the same Tegra K1 SOC as in Jetson TK1. But having a support for the SOC is not yet enough, as linuxdev said. The kernel has to support all the peripherals and there might not be many of those in common with the two boards.

There has been some patches in the upstream kernel to support the Acer’s Chromebook with the open source versions (at least for a basic boot-up) so using pure open source might also be an option in the future. That is of course highly experimental and doesn’t support as much features as the L4T.

Similarly, if it were possible to run L4T on the Shield Tablet, that would be a very handy. I’d get the Chromebook too if we found a way to flash L4T to it. As kulve mentioned, there are people out there experimenting with open source versions, but I need the hw-accelerated features like CUDA and nv_omx…

Another question:

I saw this which may be of interest:
http://www.howtogeek.com/162120/how-to-install-ubuntu-linux-on-your-chromebook-with-crouton/

It looks like Chrubuntu can actually offer hw acceleration on Acer CB5:

Crouton can run ChromeOS and Ubuntu side by side:

Best Regards,
Peter Bauer

I am using Chrubuntu…

It uses driver R19.3 on Kernel 30.10.18 from ChromeOS, the root-filesystem is Ubuntu 14.04 lts.

Have a nice day

I change the Nvidia divers pointed in the script to the updated 21.3 version
[url]Dropbox - chrubuntu-install-21.3.sh - Simplify your life
now I can install CUDA ver6.5
The original script is this one:
[url]http://svn.clifford.at/handicraft/2014/chrbook13/chrubuntu.sh[/url]

Nice to know I’m not the only one. I also modified the script, I was having a few errors getting the install to work correctly. I ran it, got it all install, had it running, and even installed the CUDA Toolkit 6.5. The system correctly identifies the CUDA and nvcc versions, I can compile the samples with no issues, but whenever I try to run one of the compiled programs I get an error that it cannot execute the binary. No other information.

Any ideas on what to check or what to try?

If an executable won’t execute, there’s a chance it was compiled for the wrong architecture or CPU. Run “file yourprogram” and see what architecture. Does it say something like one of these?

ELF 32-bit LSB  executable, ARM, EABI5 version 1 (SYSV)
ELF 64-bit LSB executable, x86-64, version 1 (SYSV)

I’ll check as soon as I get home to the CB5. Thanks for that.

If it’s compiling in anything other than the ARM, which seems odd since I installed the ARM versions of everything, how would I force it to compile in ARM architecture?

Will update you on what I find.

Cross compile is an extremely complicated topic. Kernel cross compile is simple because it’s already set up to understand the environment, and the kernel does not link to outside programs. When I cross compile for Jetson (which I try to avoid) I use a loopback mounted image of the entire Jetson environment. Even then, it can be a challenge because anything that links to anything else can accidentally try to link against something in the host.

I’d first see if you need cross compile fixes. The short answer is that having cross compile tools installed on a host is only a tiny part of proper cross compile…everything wants to put its fingers into everything else on the system…and it doesn’t care if it gets the x86 version or ARM version.

So that seems to not be the problem. Here’s the file readout, and what happens when i try to execute:

user@chrubuntu:/usr/local/cuda-6.5/samples/1_Utilities/bandwidthTest$ file bandwidthTest
bandwidthTest: ELF 32-bit LSB  executable, ARM, EABI5 version 1 (SYSV), dynamically linked (uses shared libs), for GNU/Linux 2.6.32, BuildID[sha1]=eb8e3aa231018fba75d9d496c70269f4f5a6c75f, not stripped
user@chrubuntu:/usr/local/cuda-6.5/samples/1_Utilities/bandwidthTest$ bash bandwidthTest
bandwidthTest: bandwidthTest: cannot execute binary file

Sudo doesn’t work, either.

I went ahead and compiled another of the samples, here’s the readout:

/usr/local/cuda-6.5/bin/nvcc -ccbin g++ -I../../common/inc  -m32   -Xcompiler -mfloat-abi=hard  -gencode arch=compute_20,code=compute_20 -o deviceQueryDrv.o -c deviceQueryDrv.cpp
/usr/local/cuda-6.5/bin/nvcc -ccbin g++   -m32   -Xcompiler -mfloat-abi=hard  -Xlinker --dynamic-linker=/lib/ld-linux-armhf.so.3  -gencode arch=compute_20,code=compute_20 -o deviceQueryDrv deviceQueryDrv.o  -lcuda
mkdir -p ../../bin/armv7l/linux/release/gnueabihf
cp deviceQueryDrv ../../bin/armv7l/linux/release/gnueabihf

nvcc -V gives me:

nvcc: NVIDIA (R) Cuda compiler driver
Copyright (c) 2005-2014 NVIDIA Corporation
Built on Wed_Nov_12_15:57:57_CST_2014
Cuda compilation tools, release 6.5, V6.5.30

Checking Driver version gives me:

user@chrubuntu:~$ grep -i "x driver" /var/log/Xorg.0.log
[     5.735] (II) NVIDIA dlloader X Driver  21.3  Release Build  (integ_stage_rel)  (buildbrain@mobile-u64-720)  Mon Feb  2 17:51:55 PST 2015
[     5.907] (II) NVIDIA(0):     ACPI event daemon is available, the NVIDIA X driver will

Can you even run that with bash?

I’d figure you would just run it with ./bandwidthTest if the file is an executable and has the execute permission flag set.

Huh, well there you go. I did try that, I swear, but I checked the permissions and I was not the owner of the file, which seems odd. I sudo chown the file, then ./filename, and it runs. Interesting…

Thanks for the help. Now I can start messing around. This is an excellent way to run cuda dev. Integrated keyboard and monitor on what amounts to basically a jetson.

How long until someone loads Ubuntu on the new Shield X1 console, I wonder?

If the interface for flashing the device is there, I think someone will figure it out pretty soon. They got Android running on the TK1 after all!

I installed ubuntu using Dropbox - chrubuntu-install-21.3.sh - Simplify your life
then cuda 6.5

I can compile the samples, it complained about gcc-4.9 (which is a dependency for cuda6.5?) but went on after disabling the version check.

Running deviceQuery works and says Device 0: “GK20A”, etc, … Anything else fails with an error

[CUDA Bandwidth Test] - Starting...
Running on...

 Device 0: GK20A
 Quick Mode

CUDA error at bandwidthTest.cu:753 code=46(cudaErrorDevicesUnavailable) "cudaEventCreate(&start)"

Doing the same as sudo adds: modprobe: FATAL: Module nvidia not found.
X Driver is 21.3
Kernel is 3.10.18
Nodes in /dev/nv* are

crw-rw---- 1 root video  10,  51 Jun 12 23:26 /dev/nvhdcp1
crw-rw---- 1 root video 247,   1 Jun 12 23:26 /dev/nvhost-as-gpu
crw-rw---- 1 root video 251,   0 Jun 12 23:26 /dev/nvhost-ctrl
crw-rw---- 1 root video 247,   2 Jun 12 23:26 /dev/nvhost-ctrl-gpu
crw-rw---- 1 root video 247,   0 Jun 12 23:26 /dev/nvhost-gpu
crw-rw---- 1 root video 249,   0 Jun 12 23:26 /dev/nvhost-msenc
crw-rw---- 1 root video 248,   0 Jun 12 23:26 /dev/nvhost-tsec
crw-rw---- 1 root video 246,   0 Jun 12 23:26 /dev/nvhost-vic
crw-rw-rw- 1 root root  195, 255 Jun 12 23:26 /dev/nvidiactl
crw-rw---- 1 root video  10,  62 Jun 13 06:05 /dev/nvmap

lspci does nothing, there is no /proc/bus/pci

I’m a bit stuck here …