YOLOv4-tiny darknet on Xavier NX

Hi,

I have been running Yolov4-tiny with darknet on my jetson nx setting

CUDNN=0
CUDNN_HALF=0

running inference with no problem at all.

Now, When setting:

CUDNN=1
CUDNN_HALF=1

I keep getting the following error:

CUDA status Error: file: ./src/convolutional_kernels.cu : () : line: 138 : build time: Jan 12 2022 - 11:01:56 
CUDA Error: invalid resource handle

System info:

Darknet Makefile:

GPU=1
CUDNN=1
CUDNN_HALF=1
OPENCV=1
AVX=0
OPENMP=0
LIBSO=1
ZED_CAMERA=1
ZED_CAMERA_v2_8=0
.
.
.
ARCH= -gencode arch=compute_72,code=[sm_72,compute_72]
.
.
.
NVCC=/usr/local/cuda/bin/nvcc

Has anyone encounter this before?