Skip to content

Do not report for FP16 compatibility on old Nvidia cards #21461

New issue
New issue
Closed
#25880
Closed
Do not report for FP16 compatibility on old Nvidia cards#21461
#25880
Labels
category: dnncategory: gpu/cuda (contrib)OpenCV 4.0+: moved to opencv_contrib
Milestone
4.11.0
@JulienMaille

Description

@JulienMaille
JulienMaille
opened on Jan 17, 2022

opencv/modules/dnn/src/dnn.cpp

Lines 251 to 255 in 4801526

if (haveCUDA())
{
backends.push_back(std::make_pair(DNN_BACKEND_CUDA, DNN_TARGET_CUDA));
backends.push_back(std::make_pair(DNN_BACKEND_CUDA, DNN_TARGET_CUDA_FP16));
}

Since commit d0fe6ad cv::dnn::getAvailableBackends() will list CUDA_FP16 for all Cuda devices even the old ones that do not have __half compute capabilities
The simple check has been removed, and I would suggest to add it back, what do you think @YashasSamaga ?

Metadata

Metadata

Assignees

No one assigned

    Labels

    category: dnncategory: gpu/cuda (contrib)OpenCV 4.0+: moved to opencv_contrib

    Type

    No type

    Projects

    No projects

    Milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions