Created by: thinkall

    ret = get_gpu_count() > 0
    if _HAS_FLUID:
        from paddle import fluid
        if ret is True and not fluid.is_compiled_with_cuda():
            logger.warning("Found non-empty CUDA_VISIBLE_DEVICES. \
                But PARL found that Paddle was not complied with CUDA, which may cause issues."
                           )
    return ret

In machine with CUDA GPU enabled, but cpu version paddle installed. Here when cpu version paddle is detected, it will still return ret > 0. Therefore, the following error will raise:

[06-16 14:16:01 MainThread @logger.py:224] Argv: PARL/examples/QuickStart/train.py [06-16 14:16:01 MainThread @machine_info.py:86] nvidia-smi -L found gpu count: 4 [06-16 14:16:01 MainThread @machine_info.py:105] WRN Found non-empty CUDA_VISIBLE_DEVICES. But PARL found that Paddle was not complied with CUDA, which may cause issues. [06-16 14:16:01 MainThread @machine_info.py:86] nvidia-smi -L found gpu count: 4 [06-16 14:16:01 MainThread @machine_info.py:105] WRN Found non-empty CUDA_VISIBLE_DEVICES. But PARL found that Paddle was not complied with CUDA, which may cause issues. E0616 14:16:01.705137 1281 pybind.cc:1277] Cannot use GPU because you have installed CPU version PaddlePaddle. If you want to use GPU, please try to install GPU version PaddlePaddle by: pip install paddlepaddle-gpu If you only have CPU, please change CUDAPlace(0) to be CPUPlace().

As a workaround, when cpu version paddlepaddle is detected, the function "is_gpu_available" should return False.