DeepStream 7.x build with CUDA 12.9/13.0 crashes RTX PRO 500 Blackwell GPU - requires double restart to recover

Accelerated Computing Intelligent Video Analytics DeepStream SDK
1

- Hardware Platform (Jetson / GPU)
GPU: NVIDIA RTX PRO 500 Blackwell Generation Laptop GPU (compute capability 12.0)

- DeepStream Version
DeepStream 7.1 (building from nvcr.io/nvidia/deepstream container)

- JetPack Version (valid for Jetson only)
N/A (dGPU platform)

- TensorRT Version

TensorRT v100300

- NVIDIA GPU Driver Version (valid for GPU only)
Driver Version: 580.105.08

- Issue Type (questions, new requirements, bugs)
Bug

[02/04/2026-16:19:40] [E] Error[1]: IBuilder::buildSerializedNetwork: Error Code 1: Internal Error (Unsupported SM: 0xc00)
[02/04/2026-16:19:40] [E] Engine could not be created from network
[02/04/2026-16:19:40] [E] Building engine failed
[02/04/2026-16:19:40] [E] Failed to create engine from model or file.
[02/04/2026-16:19:40] [E] Engine set up failed
&&&& FAILED TensorRT.trtexec [TensorRT v100300] # trtexec --onnx=/models/e91ed15ffef9400787a1fa69e328cd3a/ghost_reid.onnx --saveEngine=/models/e91ed15ffef9400787a1fa69e328cd3a/ghost_reid.engine --minShapes=input:1x3x384x128 --optShapes=input:4x3x384x128 --maxShapes=input:16x3x384x128 --fp16

- How to reproduce the issue?

I’m attempting to build a DeepStream 7.1 application on a laptop with RTX PRO 500 Blackwell GPU using CUDA Toolkit 12.9/13.0. The goal is to cross-compile for Jetson Orin (sm_87) deployment, but the build process crashes the GPU.

Steps to reproduce:

  1. Host system: Laptop with NVIDIA RTX PRO 500 Blackwell Generation GPU
  2. Using CUDA Toolkit 12.9/13.0
  3. Building DeepStream 7.1 container/application (targeting ARM64/sm_87 for Jetson Orin)

Questions:

  1. Is RTX PRO 500 Blackwell officially supported for DeepStream 7.1builds (as host build machine)?
  2. Are there known issues with CUDA 12.9 + Blackwell architecture when building DeepStream?
  3. Should I use a different CUDA Toolkit version for building on Blackwell while targeting Orin sm_87?

Additional context:

  • Target deployment: Jetson Orin (sm_87), not the Blackwell GPU itself
  • Previous forum discussions mentioned DeepStream 7.1 doesn’t officially support Blackwell for runtime, but I’m using it only as a build host

Any guidance on stable build configuration or known compatibility issues would be appreciated.

4

DS-7.1 depends on TRT-10.3, but TRT-10.3 cannot run on Blackwell, therefore DS-7.1 also cannot run.

If you only need to cross-compile, you don’t need to install a GPU on the host machine. As long as nvcc works, that’s sufficient.

Please refer to this link.

5

There is no update from you for a period, assuming this is not an issue anymore. Hence we are closing this topic. If need further support, please open a new one. Thanks.