Compilation error to architecture with video card RTX 1660

[aldisio]

My setup

• Ubuntu 18.04

• $ nvcc --version
  nvcc: NVIDIA (R) Cuda compiler driver
  Copyright (c) 2005-2018 NVIDIA Corporation
  Built on Sat_Aug_25_21:08:01_CDT_2018
  Cuda compilation tools, release 10.0, V10.0.130

• $ cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2
  #define CUDNN_MAJOR 7
  #define CUDNN_MINOR 6
  #define CUDNN_PATCHLEVEL 5

Error occurs after this commands:
$ make clean
$ make GPU=1 CUDNN=1 LIBSO=1 ARCH=" -gencode arch=compute_75,code=[sm_75]"

Output error:
[ ... ]

nvcc -gencode arch=compute_75,code=[sm_75] -Iinclude/ -I3rdparty/stb/include -DGPU -I/usr/local/cuda/include/ -DCUDNN --compiler-options "-Wall -Wfatal-errors -Wno-unused-result -Wno-unknown-pragmas -fPIC -Ofast -DGPU -DCUDNN -I/usr/local/cudnn/include -fPIC" -c ./src/crop_layer_kernels.cu -o obj/crop_layer_kernels.o
nvcc -gencode arch=compute_75,code=[sm_75] -Iinclude/ -I3rdparty/stb/include -DGPU -I/usr/local/cuda/include/ -DCUDNN --compiler-options "-Wall -Wfatal-errors -Wno-unused-result -Wno-unknown-pragmas -fPIC -Ofast -DGPU -DCUDNN -I/usr/local/cudnn/include -fPIC" -c ./src/dropout_layer_kernels.cu -o obj/dropout_layer_kernels.o
./src/dropout_layer_kernels.cu(140): warning: variable "cur_scale" was declared but never referenced

./src/dropout_layer_kernels.cu(245): warning: variable "cur_scale" was declared but never referenced

./src/dropout_layer_kernels.cu(262): warning: variable "block_prob" was declared but never referenced

nvcc -gencode arch=compute_75,code=[sm_75] -Iinclude/ -I3rdparty/stb/include -DGPU -I/usr/local/cuda/include/ -DCUDNN --compiler-options "-Wall -Wfatal-errors -Wno-unused-result -Wno-unknown-pragmas -fPIC -Ofast -DGPU -DCUDNN -I/usr/local/cudnn/include -fPIC" -c ./src/maxpool_layer_kernels.cu -o obj/maxpool_layer_kernels.o
nvcc -gencode arch=compute_75,code=[sm_75] -Iinclude/ -I3rdparty/stb/include -DGPU -I/usr/local/cuda/include/ -DCUDNN --compiler-options "-Wall -Wfatal-errors -Wno-unused-result -Wno-unknown-pragmas -fPIC -Ofast -DGPU -DCUDNN -I/usr/local/cudnn/include -fPIC" -c ./src/network_kernels.cu -o obj/network_kernels.o
./src/network_kernels.cu(364): warning: variable "l" was declared but never referenced

./src/network_kernels.cu(706): error: identifier "cudaStreamCaptureModeGlobal" is undefined

./src/network_kernels.cu(706): error: too many arguments in function call

2 errors detected in the compilation of "/tmp/tmpxft_00005390_00000000-6_network_kernels.cpp1.ii".
Makefile:185: recipe for target 'obj/network_kernels.o' failed
make: *** [obj/network_kernels.o] Error 1

[aldisio]

The compilation process run without error when return to checkout version 64efa72.
Versions after this commit, can be incompatibility with CUDA version 10.0.

$ git checkout 64efa721ede91cd8ccc18257f98eeba43b73a6af
$ make clean
$ make GPU=1 CUDNN=1 LIBSO=1 ARCH=" -gencode arch=compute_75,code=[sm_75]"
´´´ 

[folkien]

The compilation process run without error when return to checkout version 64efa72.
Versions after this commit, can be incompatibility with CUDA version 10.0.

$ git checkout 64efa721ede91cd8ccc18257f98eeba43b73a6af
$ make clean
$ make GPU=1 CUDNN=1 LIBSO=1 ARCH=" -gencode arch=compute_75,code=[sm_75]"
´´´ 

So which cuda version should be used?

[stephanecharette]

See: #7486 (comment)