Cuda version on Host pc

Please provide the following info (tick the boxes after creating this topic):
Software Version
[*] DRIVE OS 6.0.8.1
DRIVE OS 6.0.6
DRIVE OS 6.0.5
DRIVE OS 6.0.4 (rev. 1)
DRIVE OS 6.0.4 SDK
other

Target Operating System
[*] Linux
QNX
other

Hardware Platform
[*] DRIVE AGX Orin Developer Kit (940-63710-0010-300)
DRIVE AGX Orin Developer Kit (940-63710-0010-200)
DRIVE AGX Orin Developer Kit (940-63710-0010-100)
DRIVE AGX Orin Developer Kit (940-63710-0010-D00)
DRIVE AGX Orin Developer Kit (940-63710-0010-C00)
DRIVE AGX Orin Developer Kit (not sure its number)
other

SDK Manager Version
1.9.3.10904
[*] other

Host Machine Version
native Ubuntu Linux 20.04 Host installed with SDK Manager
[*] native Ubuntu Linux 20.04 Host installed with DRIVE OS Docker Containers
native Ubuntu Linux 18.04 Host installed with DRIVE OS Docker Containers
other

Hello,

We wanted to know if the host pc needs to have the exact same cuda version of 11.4 which is the same on Nvidia Drive AGX Orin? or Could the host pc have the latest cuda version installed?

We also wanted run compiled code sent to the Drive AGX Orin and would it be okay if the host machine has a newer version of cuda compared to target device (Drive AGX Orin) having cuda version 11.4?

Do have have to compile the codes using Cuda version 11.4? or will using a newer version of Cuda be an issue for Drive AGX Orin?

Thanks

Dear @kevin.kal,
You need to use same TRT/CUDA version on host provided by DRIVE OS which is flashed on target. When you use sdkmanager, it installs the requires DRIVE OS compatible CUDA/TRT on host. But, I see you selected DRIVE OS 6.0.8.1. For this version, We don’t have host installer using sdkmanager and provided DRIVE OS 6.0.8.1 docker. You can cross compile your sample on docker and sent the binaries to run on target.

Hello @SivaRamaKrishnaNV

Thanks for replying back

Yes, I am aware sdkmanger doesn’t flash the target device for drive os 6.0.8.1.

However does that mean the host pc can have a higher cuda version? because on my current host pc, I have ubuntu 20.04 and installed cuda version 12.1. I was able to successfully flash the Drive OS using docker and I was able to go into NVIDIA DriveWorks SDK as guest docker container compile the samples and run the sample codes on docker and display it on my Host pc and I was able to cross compile for target device and send the files to target device.

Dear @kevin.kal,
The docker uses the CUDA/TRT headers or libs installed on docker while cross compiling the application. So it should not be a problem. It looks you could cross compile or run the sample already. Do you see any issue?

I wasn’t sure before thanks for clarifying. Basically we have another PC and we wanted to do some testing with driveworks. I installed cuda version 11.4.4 but was not able to get run the basic helloworld sample in the docker, I kept getting this error below:

“Driveworks exception thrown: Platform: cannot acquire CUDA context. Error cudaErrorCompatNotSupportedOnDevice: forward compatibility was attempted on non supported HW”

I found out the driver version I installed that came with cuda toolkit 11.4.4 was 470.82.01 when on the system requirements for host machine section on DRIVE OS 6.0.8.1 shows a driver version of >=470.182.03 is needed.

Hence why I wanted to make sure downloading and using 12.1 cuda version is okay to do so, that’s all.

1 Like

Dear @kevin.kal,
Can we close the topic or do you have any follow up queries?

No questions, Yes you may close topic. Thanks for the information.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.