MinkowskiEngine Does Not Support RTX 5090 (sm_120 Architecture)

by ADMIN 64 views

MinkowskiEngine does not support RTX 5090 (sm_120 architecture)

Issue Description

CUDA Architecture Compatibility Issues with MinkowskiEngine

MinkowskiEngine is a popular PyTorch library for efficient 3D geometric computations. However, when attempting to run code using MinkowskiEngine on a system with an RTX 5090 GPU, I encountered errors related to CUDA architecture compatibility. The RTX 5090 uses the sm_120 architecture, while the current version of MinkowskiEngine (0.5.4) appears to only support up to sm_90. This incompatibility prevents MinkowskiEngine from running on the RTX 5090, resulting in a CUDA error.

Environment Information

  • GPU: NVIDIA RTX 5090
  • CUDA Version: CUDA 12.8 installed on the system (required for RTX 5090)
  • CUDA Environment: Using conda environment configured with CUDA 12.1 packages
  • PyTorch Version: 2.5.1
  • MinkowskiEngine Version: 0.5.4
  • Operating System: Ubuntu
  • Python Version: 3.10

Error Message

When attempting to run MinkowskiEngine operations, I encounter the following error:

CUDA error: no kernel image is available for execution on the device
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

The detailed CUDA error indicates this is due to the sm_120 architecture not being supported.

Analysis

Understanding the CUDA Architecture Incompatibility

The RTX 5090 uses NVIDIA's latest Blackwell architecture with compute capability sm_120. After examining MinkowskiEngine's installation process, I found that the CUDA architectures specified during compilation do not include sm_120, preventing it from running on the RTX 5090. This incompatibility is a significant limitation, as the RTX 5090 is a powerful GPU that can provide significant performance benefits for 3D geometric computations.

Potential Compatibility Issues

There may also be compatibility issues between the system-level CUDA 12.8 (required for RTX 5090) and the conda environment configured with CUDA 12.1 packages. While CUDA 12.8 is backward compatible, this combination might contribute to the problem. It is essential to ensure that the CUDA environment is properly configured to avoid such compatibility issues.

Request

  1. Update MinkowskiEngine to Support NVIDIA Blackwell Architecture (sm_120)

To resolve this issue, I request that MinkowskiEngine be updated to support the NVIDIA Blackwell architecture (sm_120). This will enable MinkowskiEngine to run on the RTX 5090 and other GPUs with the same architecture.

  1. Temporary Solution: Adding sm_120 Support When Compiling from Source

If an immediate update is not possible, I would appreciate guidance on how to add sm_120 support when compiling MinkowskiEngine from source. This would provide a temporary solution until the library is updated to support the sm_120 architecture.

Attempted Solutions

  • Running on CPU: I tried running on CPU, but many critical operations (such as furthest_point_sampling in PointNet++) show "CPU not supported" errors.
  • Using CUDA_LAUNCH_BLOCKING=1: I tried running with CUDA_LAUNCH_BLOCKING=1, but the same error persists.
  • Using Different CUDA Environments: I attempted to use different CUDA environments but encountered similar issues.

Thank you for your attention and assistance in resolving this issue. I look forward to hearing from you soon.

Conclusion

MinkowskiEngine is a powerful library for 3D geometric computations, but its current version does not support the NVIDIA Blackwell architecture (sm_120). This incompatibility prevents MinkowskiEngine from running on the RTX 5090 and other GPUs with the same architecture. I request that MinkowskiEngine be updated to support the sm_120 architecture or provide guidance on how to add sm_120 support when compiling from source.
MinkowskiEngine does not support RTX 5090 (sm_120 architecture) - Q&A

Q: What is the issue with MinkowskiEngine and the RTX 5090 GPU?

A: The issue is that the current version of MinkowskiEngine (0.5.4) only supports up to CUDA architecture sm_90, while the RTX 5090 uses the sm_120 architecture. This incompatibility prevents MinkowskiEngine from running on the RTX 5090.

Q: What are the system requirements for running MinkowskiEngine on the RTX 5090?

A: To run MinkowskiEngine on the RTX 5090, you need:

  • A system with an RTX 5090 GPU
  • CUDA 12.8 installed on the system (required for RTX 5090)
  • A conda environment configured with CUDA 12.1 packages
  • PyTorch Version 2.5.1
  • MinkowskiEngine Version 0.5.4
  • Operating System: Ubuntu
  • Python Version: 3.10

Q: What error message do I get when trying to run MinkowskiEngine on the RTX 5090?

A: When attempting to run MinkowskiEngine operations, you may encounter the following error:

CUDA error: no kernel image is available for execution on the device
CUDA kernel errors might be asynchronously reported at some other API call, so the stacktrace below might be incorrect.
For debugging consider passing CUDA_LAUNCH_BLOCKING=1.

Q: What are the potential causes of the CUDA error?

A: The CUDA error may be caused by:

  • The sm_120 architecture not being supported by MinkowskiEngine
  • Compatibility issues between the system-level CUDA 12.8 and the conda environment configured with CUDA 12.1 packages

Q: How can I resolve the issue?

A: To resolve the issue, you can:

  • Update MinkowskiEngine to support the NVIDIA Blackwell architecture (sm_120)
  • Add sm_120 support when compiling MinkowskiEngine from source
  • Use a different GPU that supports the sm_90 architecture

Q: What are the benefits of updating MinkowskiEngine to support the sm_120 architecture?

A: Updating MinkowskiEngine to support the sm_120 architecture will enable it to run on the RTX 5090 and other GPUs with the same architecture, providing significant performance benefits for 3D geometric computations.

Q: Can I add sm_120 support when compiling MinkowskiEngine from source?

A: Yes, you can add sm_120 support when compiling MinkowskiEngine from source. However, this may require additional configuration and may not be a straightforward process.

Q: What are the potential risks of using a different CUDA environment?

A: Using a different CUDA environment may introduce compatibility issues and may not resolve the issue.

Q: Can I run MinkowskiEngine on CPU?

A: Yes, you can run MinkowskiEngine on CPU. However, many critical operations (such as furthest_point_sampling in PointNet++) may show "CPU not supported" errors.

Q: What are the next steps to resolve the issue?

A: The next steps to resolve the issue are to:

  • Update MinkowskiEngine to support the NVIDIA Blackwell architecture (sm_)
  • Add sm_120 support when compiling MinkowskiEngine from source
  • Use a different GPU that supports the sm_90 architecture

Q: How can I get help with resolving the issue?

A: You can get help with resolving the issue by:

  • Contacting the MinkowskiEngine developers
  • Joining the MinkowskiEngine community
  • Seeking help from a professional developer or researcher

Conclusion

MinkowskiEngine does not support the RTX 5090 (sm_120 architecture) due to incompatibility issues. To resolve the issue, you can update MinkowskiEngine to support the sm_120 architecture, add sm_120 support when compiling from source, or use a different GPU that supports the sm_90 architecture. If you need help with resolving the issue, you can contact the MinkowskiEngine developers, join the MinkowskiEngine community, or seek help from a professional developer or researcher.