DRIVE OS 6.0.10 / DriveWorks on RTX 5070 Ti (Blackwell) – general compatibility

Autonomous Vehicles DRIVE AGX Orin DRIVE AGX Orin General
1

Hi all,

we are setting up multiple new x86 development workstation and are trying to run DRIVE OS / DriveWorks samples on the host GPUs (not just cross-compiling for Orin). Our goal is to use the workstation GPUs for day-to-day development and for running the provided samples, while targeting multiple DRIVE AGX Orin devices.

Host setup

  • OS: Ubuntu 24.04
  • GPU: GeForce RTX 5070 Ti (sm_120)
  • NVIDIA driver: (from nvidia-smi) 580.95.05 (CUDA Version 13.0)
  • Using DRIVE OS Linux SDK 6.0.10 (Docker-based workflow)

Problem description
When running DriveWorks / DRIVE OS samples that use CUDA (in our case something that triggers the feature detector path), we get:

Driveworks exception thrown:
src/dw/imageprocessing/featuredetetor/implFeatureDetectorEx.cu:1649
cudaErrorNoKernelImageForDevice: no kernel image is available for execution on the device

Questions

  • Is DRIVE OS / DriveWorks 6.0.10 on x86 expected to work on Blackwell (sm_120 / RTX 50-series) for running samples?
  • If not supported: is there any workaround (driver/CUDA settings, PTX JIT, etc.) to run the samples on RTX 5070 Ti?
  • Since we target multiple DRIVE AGX Orin devices: is there any supported way to keep Orin compatibility but use a host setup that also supports sm_120, or is the only solution to use an older supported GPU for host-side sample execution?
2

You probably want the DRIVE AGX Orin forum.

3

Moving to DRIVE AGX Orin forum.

5

Dear @Jis ,
we have not tested RTX 50 series and may notice issues when running DW samples. For example, DNN sample uses ampere TRT engine which will fail to run. You need to generate the model again using tensorRT_optimzation tool and see if it works.

Are you using CUDA preinstalled on docker i.e CUDA 11.4? Did you compile the sample and ran or directly using the preinstalled binaries?