Optical Flow SDK fail with cudaErrorPeerAccessNotEnabled

I’ve run sample of Optical Flow SDK 1.1 successfully, and my app integrates the code of Sample. The following errors are reported when running on the same machine. I really appreciate some help ASAP.

CUDA error at /app/.../kernel.cu:113 code=705(cudaErrorPeerAccessNotEnabled) "cudaCreateSurfaceObject(&srcSurfObj, &srcSurfRes)"

Can any one give me some suggestions on what I should/can do, thanks a lot!

Here is my configuration:

  • CPU : Intel® Xeon® Silver 4114 CPU @ 2.20GHz
  • RAM : 32 GB
  • DISK : 1 TB SSD

Current Driver 435.21
Cuda: 10.1 with cudnn
Ubuntu 18.04


Is optical Flow SDK1.1 incompatible with Deepstream 4.0? I did experiments and found that it works well before creating ds4.0 pipeline,but the result is wrong. After creating ds4.0 pipeline, it runs out and reports errors as above.

What’s your GPU?
As you can find in Optical Flow download page, only Turing GPUs have Optical Flow capability.

Hi mchi,
My Gpu is T4.