failed from TEST#206 with gtx3090

[black6816]

hi, when i test with gtx3090, it failed from TEST#206，58 test failed out of 263.

[lightbulb128]

I think this should be a problem with the CUDA environment. Do you have any specific details on the output of the failed tests? You could run one specific test with verbose output by, for example, ctest -R EncryptorCudaTest.BFVEncryptDecrypt -V. And you could also try to run the time test.

[black6816]

i use nvcr.io/nvidia/nvhpc:23.1-devel-cuda_multi-ubuntu20.04 docker image, which nvcc version is release V11.8, V11.8.89. I rerun ctest -R EncryptorCudaTest.BFVEncryptDecrypt -V, nothing except 1/1 Test #206: EncryptorCudaTest.BFVEncryptDecrypt ...***Failed 1.05 sec

[black6816]

when i run ctest with A100, all test cases are passed. `261/263 Test #261: EvaluatorCudaTest.SerializeTest ........................................ Passed 0.33 sec Start 262: CKKSEncoderCudaTest.CKKSEncoderEncodeVectorDecodeTest 262/263 Test #262: CKKSEncoderCudaTest.CKKSEncoderEncodeVectorDecodeTest .................. Passed 0.53 sec Start 263: CKKSEncoderCudaTest.CKKSEncoderEncodeSingleDecodeTest 263/263 Test #263: CKKSEncoderCudaTest.CKKSEncoderEncodeSingleDecodeTest .................. Passed 0.65 sec

100% tests passed, 0 tests failed out of 263 `

[lightbulb128]

I am not very familiar with docker. Can you build and run any other CUDA program with gtx3090 and that docker container?

[black6816]

yes, i can build and run cuda-samples/Samples/0_Introduction/vectorAdd, Test PASSED

[lightbulb128]

I test this with my A100 and RTX4090 and both pass the test. Here are my steps

sudo docker run -d -it --net=host --gpus all --name troy nvcr.io/nvidia/nvhpc:23.1-devel-cuda_multi-ubuntu20.04
sudo docker attach troy

cd root
git clone https://github.com/lightbulb128/troy.git
cd troy
git submodule update --init
mkdir build
cd build
cmake .. -DTROY_TEST=ON
make
ctest -R Cuda

Probably should add some basic unit tests for GPU availability