I was able to resolve this issue by upgrading cuda & cudnn. In my case, I had been running my program on an outdated Nvidia Docker image with cuda 11.8 and cudnn 8:

nvidia/cuda:11.8.0-cudnn8-runtime-ubuntu22.04

The issued disappeared when I upgraded to a more recent Nvidia Docker image (as of Oct '24), which has cuda 12.6 and cudnn devel:

nvidia/cuda:12.6.2-cudnn-devel-ubuntu22.04

A full list Nvidia Docker images are available here. For those who aren't using docker, you can try upgrading your cuda and cudnn versions manually!