I run the code completely according to your environment , but it takes more than 12 hours to run COCO on 2080ti. This should be a very abnormal situation.
Hi, you can see my installation details here.
Since most code is built on MetaRCNN, I use CUDA 8.0 with PyTorch 0.4.1 to compile the make file.
As for the issue with 2080ti, I've never tried on it but I suppose the problem comes from the cuda version being too slow for 2080ti? (c.f. here)
I run the code completely according to your environment , but it takes more than 12 hours to run COCO on 2080ti. This should be a very abnormal situation.
I'm looking forward to your reply