is it possible compile on 2080Ti?

[AlgorithmicIntelligence]

I've been tried many the version of pytorch, eventually, it's success at 0.4.0. But it seems pytorch 0.4.0 not compatible with cuda 10.0, and 2080Ti need cuda >= 10.0.

So I want to know is anybody successful run it on 2080Ti?

[igygi]

You can try building the environment using conda. Try this for your torch installation:

conda install pytorch=0.4.0 cuda90 -c pytorch. From my experience, this works with 2080Ti.

[ZY-Zeng]

I have a same problem about 2080Ti and try building the environment that you said. But it not works. So, can you give me more detail information about your environment. (The version of all packages) Thank thank thank you !!!

[igygi]

What error are you encountering when building the environment?

Here are the steps I took to build my environment:

1. Create a conda environment for PANet with python version 3.6: $ conda create -n <env-name> python=3.6.8
2. Activate the conda environment: $ conda activate <env-name>
3. Install the required packages:
cython			$ conda install cython
matplotlib		$ conda install matplotlib
scipy			$ conda install scipy
opencv			$ conda install -c conda-forge opencv
pyyaml			$ conda install pyyaml
packaging		$ conda install packaging
pycocotools		$ conda install -c hcc pycocotools
tensorboardX		$ conda install -c conda-forge tensorboardx
pandas			$ conda install pandas
pytorch 0.4.0		$ conda install pytorch=0.4.0 cuda90 -c pytorch
torchvision 0.2.1		$ conda install torchvision=0.2.1

Note: Make sure to install pytorch 0.4.0 with cuda 9.0 support. The command specified above should accomplish this.
	
4. Go to the lib directory inside the PANet folder: 
$ cd PANet
$ cd lib
5. Edit make.sh file and uncomment the line -gencode arch=compute_70,code=sm_70 "
5. Run sh make.sh

[ZY-Zeng]

@igygi Well! Very very very thank you for your timely reply. I will try it in the afternoon.

[ZY-Zeng]

cool!!!!! It works! Thank you very much again!!!

[lixiuyu-cumt]

@ZY-Zeng @igygi I have the same question, I want to know whether you install the cuda9.0? ,or you just use the commad "conda install pytorch=0.4.0 cuda90 -c pytorch" and not install cuda9.0?

[ZY-Zeng]

@ZY-Zeng @igygi I have the same question, I want to know whether you install the cuda9.0? ,or you just use the commad "conda install pytorch=0.4.0 cuda90 -c pytorch" and not install cuda9.0?

Of course. You need install cuda 9.0 before creation of environment.

[lixiuyu-cumt]

@ZY-Zeng @igygi I have the same question, I want to know whether you install the cuda9.0? ,or you just use the commad "conda install pytorch=0.4.0 cuda90 -c pytorch" and not install cuda9.0?

Of course. You need install cuda 9.0 before creation of environment.

I installed cuda 9.0 after creating the environment, and I followed the steps step by step installation.But I encountered the question

"THCudaCheck FAIL file=/opt/conda/conda-bld/pytorch_1524586445097/work/aten/src/THC/THCGeneral.cpp line=844 error=11 : invalid argument cudaCheckError() failed : no kernel image is available for execution on the device " Has this ever happened to you? Have you installed the correspoding cudnn?Maybe RTX2080Ti is not compatible with CUDA9.0?