search

NVIDIA-AI-IOT / Lidar_AI_Solution

A project demonstrating Lidar related AI solutions, including three GPU accelerated Lidar/camera DL networks (PointPillars, CenterPoint, BEVFusion) and the related libs (cuPCL, 3D SparseConvolution, YUV2RGB, cuOSD,).
Other
1.34k stars 236 forks source link

The GPU Compute Capability #213

Closed ChunyuFeng closed 1 year ago

ChunyuFeng commented 1 year ago

There is a limitation in terms of GPU compute capability, specifically ">=sm_80", and several other issues mentioned this restriction. I am curious if the libspconv.so library is solely responsible for this limitation. Additionally, if I only use the RGB branch and avoid utilizing sparse convolution, can I use devices with lower compute capability (like Xavier, sm_72)?

hopef commented 1 year ago

Yes, you can. the SM limitation is valid in libspconv.so only, but the RGB branch has no.

ChunyuFeng commented 1 year ago

Yes, you can. the SM limitation is valid in libspconv.so only, but the RGB branch has no.

Thanks! I'll try this!