Installation error with RTX 2080, CUDA 9.0, CUDNN 7.4 nvcc fatal : Unsupported gpu architecture 'compute_75' #1190

nohboogy · 2019-01-21T02:03:14Z

While trying to install torch, same error comes out.

canning dependencies of target image
[ 87%] Building C object CMakeFiles/image.dir/image.c.o
[100%] Linking C shared module libimage.so
[100%] Built target image
cd build && make install
[ 25%] Built target jpeg
[ 50%] Built target ppm
[ 75%] Built target lua_png
[100%] Built target image
Install the project...
-- Install configuration: "Release"
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lib/libppm.so
-- Set runtime path of "/home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lib/libppm.so" to "$ORIGIN/../lib:/home/wooseok/torch/install/lib:/opt/OpenBLAS/lib"
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lib/libjpeg.so
-- Set runtime path of "/home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lib/libjpeg.so" to "$ORIGIN/../lib:/home/wooseok/torch/install/lib:/opt/OpenBLAS/lib"
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lib/liblua_png.so
-- Set runtime path of "/home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lib/liblua_png.so" to "$ORIGIN/../lib:/home/wooseok/torch/install/lib:/opt/OpenBLAS/lib"
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lib/libimage.so
-- Set runtime path of "/home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lib/libimage.so" to "$ORIGIN/../lib:/home/wooseok/torch/install/lib:/opt/OpenBLAS/lib"
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/init.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/win.ui
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/test.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/grace_hopper_512.jpg
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/corrupt-ihdr.png
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/gray3x1.png
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/grace_hopper_512.png
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/rectangle.png
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/P4.pbm
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/rgb2x1.png
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/foobar.png
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/P6.ppm
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/P5.pgm
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/P2.pgm
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/gray16-1x2.png
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/bmp-without-ext
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/fabio.png
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/fabio.jpg
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/assets/rgb16-2x1.png
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/image/1.1.alpha-0/lua/image/README.md
Updating manifest for /home/wooseok/torch/install/lib/luarocks/rocks
image 1.1.alpha-0 is now built and installed in /home/wooseok/torch/install/ (license: BSD)

cmake -E make_directory build && cd build && cmake .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_PREFIX_PATH="/home/wooseok/torch/install/bin/.." -DCMAKE_INSTALL_PREFIX="/home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0" && make

-- The C compiler identification is GNU 5.4.0
-- The CXX compiler identification is GNU 5.4.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Found Torch7 in /home/wooseok/torch/install
-- Configuring done
-- Generating done
-- Build files have been written to: /home/wooseok/torch/pkg/optim/build
cd build && make install
Install the project...
-- Install configuration: "Release"
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/adadelta.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/cg.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/lswolfe.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/nag.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/rmsprop.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/ConfusionMatrix.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/adagrad.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/sgd.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/polyinterp.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/lbfgs.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/adamax.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/adam.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/init.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/de.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/fista.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/cmaes.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/rprop.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/checkgrad.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/Logger.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/asgd.lua
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/intro.md
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/algos.md
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/logger.md
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/logger_plot.png
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/image
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/image/parameterflattening.svg
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/image/parameterflattening.png
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/doc/image/parameterflattening.svg.png
-- Installing: /home/wooseok/torch/install/lib/luarocks/rocks/optim/1.0.5-0/lua/optim/README.md
Updating manifest for /home/wooseok/torch/install/lib/luarocks/rocks
optim 1.0.5-0 is now built and installed in /home/wooseok/torch/install/ (license: BSD)

Found CUDA on your machine. Installing CUDA packages
Warning: unmatched variable LUALIB

jopts=$(getconf _NPROCESSORS_CONF)

echo "Building on $jopts cores"
cmake -E make_directory build && cd build && cmake .. -DLUALIB= -DLUA_INCDIR=/home/wooseok/torch/install/include -DCMAKE_CXX_FLAGS=${CMAKE_CXX_FLAGS} -DCMAKE_BUILD_TYPE=Release -DCMAKE_PREFIX_PATH="/home/wooseok/torch/install/bin/.." -DCMAKE_INSTALL_PREFIX="/home/wooseok/torch/install/lib/luarocks/rocks/cutorch/scm-1" && make -j$jopts install

Building on 8 cores
-- The C compiler identification is GNU 5.4.0
-- The CXX compiler identification is GNU 5.4.0
-- Check for working C compiler: /usr/bin/cc
-- Check for working C compiler: /usr/bin/cc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
-- Detecting C compile features
-- Detecting C compile features - done
-- Check for working CXX compiler: /usr/bin/c++
-- Check for working CXX compiler: /usr/bin/c++ -- works
-- Detecting CXX compiler ABI info
-- Detecting CXX compiler ABI info - done
-- Detecting CXX compile features
-- Detecting CXX compile features - done
-- Found Torch7 in /home/wooseok/torch/install
-- Found CUDA: /usr/local/cuda-9.0 (found suitable version "9.0", minimum required is "6.5")
-- Removing -DNDEBUG from compile flags
-- TH_LIBRARIES: TH
-- MAGMA not found. Compiling without MAGMA support
-- Autodetected CUDA architecture(s): 7.5
-- got cuda version 9.0
-- Found CUDA with FP16 support, compiling with torch.CudaHalfTensor
-- CUDA_NVCC_FLAGS: -D__CUDA_NO_HALF_OPERATORS__;-gencode;arch=compute_75,code=sm_75;-DCUDA_HAS_FP16=1
-- THC_SO_VERSION: 0
-- Performing Test HAS_LUAL_SETFUNCS
-- Performing Test HAS_LUAL_SETFUNCS - Failed
-- Configuring done
-- Generating done
-- Build files have been written to: /home/wooseok/torch/extra/cutorch/build
[ 2%] Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir/THC_generated_THCHalf.cu.o
[ 2%] Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir/THC_generated_THCReduceApplyUtils.cu.o
[ 3%] Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir/THC_generated_THCBlas.cu.o
[ 4%] Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir/THC_generated_THCSleep.cu.o
[ 5%] Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir/THC_generated_THCStorage.cu.o
[ 6%] Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir/THC_generated_THCStorageCopy.cu.o
[ 7%] Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensor.cu.o
[ 8%] Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensorCopy.cu.o
nvcc fatal : Unsupported gpu architecture 'compute_75'
nvcc fatal : Unsupported gpu architecture 'compute_75'
CMake Error at THC_generated_THCSleep.cu.o.cmake:207 (message):
Error generating
/home/wooseok/torch/extra/cutorch/build/lib/THC/CMakeFiles/THC.dir//./THC_generated_THCSleep.cu.o

CMake Error at THC_generated_THCReduceApplyUtils.cu.o.cmake:207 (message):
Error generating
/home/wooseok/torch/extra/cutorch/build/lib/THC/CMakeFiles/THC.dir//./THC_generated_THCReduceApplyUtils.cu.o

lib/THC/CMakeFiles/THC.dir/build.make:77: recipe for target 'lib/THC/CMakeFiles/THC.dir/THC_generated_THCSleep.cu.o' failed
make[2]: *** [lib/THC/CMakeFiles/THC.dir/THC_generated_THCSleep.cu.o] Error 1
make[2]: *** Waiting for unfinished jobs....
lib/THC/CMakeFiles/THC.dir/build.make:63: recipe for target 'lib/THC/CMakeFiles/THC.dir/THC_generated_THCReduceApplyUtils.cu.o' failed
make[2]: *** [lib/THC/CMakeFiles/THC.dir/THC_generated_THCReduceApplyUtils.cu.o] Error 1
nvcc fatal : Unsupported gpu architecture 'compute_75'
nvcc fatal : Unsupported gpu architecture 'compute_75'
CMake Error at THC_generated_THCBlas.cu.o.cmake:207 (message):
Error generating
/home/wooseok/torch/extra/cutorch/build/lib/THC/CMakeFiles/THC.dir//./THC_generated_THCBlas.cu.o

CMake Error at THC_generated_THCTensor.cu.o.cmake:207 (message):
Error generating
/home/wooseok/torch/extra/cutorch/build/lib/THC/CMakeFiles/THC.dir//./THC_generated_THCTensor.cu.o

lib/THC/CMakeFiles/THC.dir/build.make:70: recipe for target 'lib/THC/CMakeFiles/THC.dir/THC_generated_THCBlas.cu.o' failed
make[2]: *** [lib/THC/CMakeFiles/THC.dir/THC_generated_THCBlas.cu.o] Error 1
lib/THC/CMakeFiles/THC.dir/build.make:98: recipe for target 'lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensor.cu.o' failed
make[2]: *** [lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensor.cu.o] Error 1
nvcc fatal : Unsupported gpu architecture 'compute_75'
CMake Error at THC_generated_THCTensorCopy.cu.o.cmake:207 (message):
Error generating
/home/wooseok/torch/extra/cutorch/build/lib/THC/CMakeFiles/THC.dir//./THC_generated_THCTensorCopy.cu.o

nvcc fatal : Unsupported gpu architecture 'compute_75'
CMake Error at THC_generated_THCStorage.cu.o.cmake:207 (message):
Error generating
/home/wooseok/torch/extra/cutorch/build/lib/THC/CMakeFiles/THC.dir//./THC_generated_THCStorage.cu.o

nvcc fatal : Unsupported gpu architecture 'compute_75'
nvcc fatal : Unsupported gpu architecture 'compute_75'
CMake Error at THC_generated_THCHalf.cu.o.cmake:207 (message):
Error generating
/home/wooseok/torch/extra/cutorch/build/lib/THC/CMakeFiles/THC.dir//./THC_generated_THCHalf.cu.o

CMake Error at THC_generated_THCStorageCopy.cu.o.cmake:207 (message):
Error generating
/home/wooseok/torch/extra/cutorch/build/lib/THC/CMakeFiles/THC.dir//./THC_generated_THCStorageCopy.cu.o

lib/THC/CMakeFiles/THC.dir/build.make:105: recipe for target 'lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensorCopy.cu.o' failed
make[2]: *** [lib/THC/CMakeFiles/THC.dir/THC_generated_THCTensorCopy.cu.o] Error 1
lib/THC/CMakeFiles/THC.dir/build.make:84: recipe for target 'lib/THC/CMakeFiles/THC.dir/THC_generated_THCStorage.cu.o' failed
make[2]: *** [lib/THC/CMakeFiles/THC.dir/THC_generated_THCStorage.cu.o] Error 1
lib/THC/CMakeFiles/THC.dir/build.make:560: recipe for target 'lib/THC/CMakeFiles/THC.dir/THC_generated_THCHalf.cu.o' failed
make[2]: *** [lib/THC/CMakeFiles/THC.dir/THC_generated_THCHalf.cu.o] Error 1
lib/THC/CMakeFiles/THC.dir/build.make:91: recipe for target 'lib/THC/CMakeFiles/THC.dir/THC_generated_THCStorageCopy.cu.o' failed
make[2]: *** [lib/THC/CMakeFiles/THC.dir/THC_generated_THCStorageCopy.cu.o] Error 1
CMakeFiles/Makefile2:172: recipe for target 'lib/THC/CMakeFiles/THC.dir/all' failed
make[1]: *** [lib/THC/CMakeFiles/THC.dir/all] Error 2
Makefile:127: recipe for target 'all' failed
make: *** [all] Error 2

Error: Build error: Failed building.

I think it is because I'm using RTX graphic driver, but I cannot find any way to install torch now.
Anyone can solve this problem?

zhangqijun · 2019-01-24T02:44:57Z

same question，need some help

pablovicentem · 2019-02-15T16:15:19Z

Did you achieve to solve the problem? I have the same problem with the RTX2080 and ubuntu 18.04 and I really need to solve it. Thanks

nohboogy · 2019-02-15T18:06:53Z

@zhangqijun @pablovicentem
I solved this problem with reinstalling CUDA 10.0.
CUDA 9.0 dosen't support rtx 2080 even though it looks like we could install many applications with it.

WanliXue · 2019-03-28T23:15:25Z

I am with CUDA 10.0, still suffers this issue:

nvcc fatal : Unsupported gpu architecture 'compute_75' when install Torch on Ubuntu 16.04.6 LTS

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    3      1055      G   /usr/lib/xorg/Xorg                            59MiB |
+-----------------------------------------------------------------------------+

ha-lins · 2019-05-20T12:44:02Z

I am with CUDA 10.0, still suffers this issue:

nvcc fatal : Unsupported gpu architecture 'compute_75' when install Torch on Ubuntu 16.04.6 LTS

Fri Mar 29 10:12:20 2019
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 410.78 Driver Version: 410.78 CUDA Version: 10.0 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce RTX 208... Off | 00000000:19:00.0 Off | N/A |
| 27% 29C P8 11W / 250W | 11MiB / 10989MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 1 GeForce RTX 208... Off | 00000000:1A:00.0 Off | N/A |
| 27% 31C P8 10W / 250W | 11MiB / 10989MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 2 GeForce RTX 208... Off | 00000000:67:00.0 Off | N/A |
| 27% 35C P8 21W / 250W | 11MiB / 10989MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 3 GeForce RTX 208... Off | 00000000:68:00.0 On | N/A |
| 27% 34C P8 21W / 250W | 71MiB / 10986MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    3      1055      G   /usr/lib/xorg/Xorg                            59MiB |
+-----------------------------------------------------------------------------+

Hi, have you solved the torch installition problem ? I have the same problem and cannot solve it.
Thanks!

WanliXue · 2019-05-21T04:59:56Z

I am with CUDA 10.0, still suffers this issue:
nvcc fatal : Unsupported gpu architecture 'compute_75' when install Torch on Ubuntu 16.04.6 LTS
Fri Mar 29 10:12:20 2019
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 410.78 Driver Version: 410.78 CUDA Version: 10.0 |
|-------------------------------+----------------------+----------------------+
| GPU Name Persistence-M| Bus-Id Disp.A | Volatile Uncorr. ECC |
| Fan Temp Perf Pwr:Usage/Cap| Memory-Usage | GPU-Util Compute M. |
|===============================+======================+======================|
| 0 GeForce RTX 208... Off | 00000000:19:00.0 Off | N/A |
| 27% 29C P8 11W / 250W | 11MiB / 10989MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 1 GeForce RTX 208... Off | 00000000:1A:00.0 Off | N/A |
| 27% 31C P8 10W / 250W | 11MiB / 10989MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 2 GeForce RTX 208... Off | 00000000:67:00.0 Off | N/A |
| 27% 35C P8 21W / 250W | 11MiB / 10989MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
| 3 GeForce RTX 208... Off | 00000000:68:00.0 On | N/A |
| 27% 34C P8 21W / 250W | 71MiB / 10986MiB | 0% Default |
+-------------------------------+----------------------+----------------------+
+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    3      1055      G   /usr/lib/xorg/Xorg                            59MiB |
+-----------------------------------------------------------------------------+
Hi, have you solved the torch installition problem ? I have the same problem and cannot solve it.
Thanks!

I am sorry, I didnt get a solution for this.

talkenig · 2019-06-05T04:30:50Z

@ha-lins @WanliXue @pablovicentem @nohboogy @zhangqijun
Had the same problem and got to the bottom of it. My configuration is a Tesla T4 with 410.92 driver and CUDA 9.2, Ubuntu 18.10.
The thing is, the torch installer tries for some reason to use the highest compute capability supported by the device (or the driver - not sure which), but ignores the compute capability supported by the CUDA toolkit.
So, in my case the device supports compute capability 7.5, but CUDA 9.2 supports only 7.0 or 7.2 (not sure which one). You can guess I got the same nvcc fatal : Unsupported gpu architecture 'compute_75'
The solution is to force the nvcc compile options to use a lower compute capability. This can be achieved by setting the following environment variable:

export TORCH_CUDA_ARCH_LIST="7.0"

Just before running ./install.sh from the torch directory.
Note that the list may contain more than one compute capability, e.g. it can be "6.0 6.2 7.0 7.2". This will be reflected in the CUDA_NVCC_FLAGS -gencode arch=compute_70,code=sm_70 and those will be outputted to the terminal during the build.

RishengZhou · 2019-08-01T14:13:20Z

@ha-lins @WanliXue @pablovicentem @nohboogy @zhangqijun
Had the same problem and got to the bottom of it. My configuration is a Tesla T4 with 410.92 driver and CUDA 9.2, Ubuntu 18.10.
The thing is, the torch installer tries for some reason to use the highest compute capability supported by the device (or the driver - not sure which), but ignores the compute capability supported by the CUDA toolkit.
So, in my case the device supports compute capability 7.5, but CUDA 9.2 supports only 7.0 or 7.2 (not sure which one). You can guess I got the same nvcc fatal : Unsupported gpu architecture 'compute_75'
The solution is to force the nvcc compile options to use a lower compute capability. This can be achieved by setting the following environment variable:

export TORCH_CUDA_ARCH_LIST="7.0"

Just before running ./install.sh from the torch directory.
Note that the list may contain more than one compute capability, e.g. it can be "6.0 6.2 7.0 7.2". This will be reflected in the CUDA_NVCC_FLAGS -gencode arch=compute_70,code=sm_70 and those will be outputted to the terminal during the build.

Thanks! It works.

talkenig · 2019-08-01T16:22:11Z

@RishengZhou
By the way, another approach that worked for me without any tweaks needed, on a machine with Ubuntu 18.04 and CUDA 10.1 is simply using the following git repo instead of the official one:
https://github.com/nagadomi/distro

vicya · 2019-10-08T08:39:17Z

@ha-lins @WanliXue @pablovicentem @nohboogy @zhangqijun
Had the same problem and got to the bottom of it. My configuration is a Tesla T4 with 410.92 driver and CUDA 9.2, Ubuntu 18.10.
The thing is, the torch installer tries for some reason to use the highest compute capability supported by the device (or the driver - not sure which), but ignores the compute capability supported by the CUDA toolkit.
So, in my case the device supports compute capability 7.5, but CUDA 9.2 supports only 7.0 or 7.2 (not sure which one). You can guess I got the same nvcc fatal : Unsupported gpu architecture 'compute_75'
The solution is to force the nvcc compile options to use a lower compute capability. This can be achieved by setting the following environment variable:

export TORCH_CUDA_ARCH_LIST="7.0"

Just before running ./install.sh from the torch directory.
Note that the list may contain more than one compute capability, e.g. it can be "6.0 6.2 7.0 7.2". This will be reflected in the CUDA_NVCC_FLAGS -gencode arch=compute_70,code=sm_70 and those will be outputted to the terminal during the build.

Thanks ! My configuration is a RTX 2080 CUDA 7.5, Ubuntu 16.04,

export TORCH_CUDA_ARCH_LIST="5.0"

Succeed

viven12138 · 2019-10-22T03:19:06Z

@ha-lins @WanliXue @pablovicentem @nohboogy @zhangqijun
Had the same problem and got to the bottom of it. My configuration is a Tesla T4 with 410.92 driver and CUDA 9.2, Ubuntu 18.10.
The thing is, the torch installer tries for some reason to use the highest compute capability supported by the device (or the driver - not sure which), but ignores the compute capability supported by the CUDA toolkit.
So, in my case the device supports compute capability 7.5, but CUDA 9.2 supports only 7.0 or 7.2 (not sure which one). You can guess I got the same nvcc fatal : Unsupported gpu architecture 'compute_75'
The solution is to force the nvcc compile options to use a lower compute capability. This can be achieved by setting the following environment variable:

export TORCH_CUDA_ARCH_LIST="7.0"

Just before running ./install.sh from the torch directory.
Note that the list may contain more than one compute capability, e.g. it can be "6.0 6.2 7.0 7.2". This will be reflected in the CUDA_NVCC_FLAGS -gencode arch=compute_70,code=sm_70 and those will be outputted to the terminal during the build.

Thanks! I met the issue when i install detectron2. It works! You are amazing!

morganmcg1 · 2020-01-30T20:10:52Z

I had 2 different versions of CUDA, but solved it by removing everything and re-installing 10.1

I'm on Ubuntu 18.04 with 2080

Highly recommend following the instructions in this blog, worked perfectly for me: https://medium.com/@exesse/cuda-10-1-installation-on-ubuntu-18-04-lts-d04f89287130

duybn-0885 · 2020-09-22T07:50:11Z

@ha-lins @WanliXue @pablovicentem @nohboogy @zhangqijun
Had the same problem and got to the bottom of it. My configuration is a Tesla T4 with 410.92 driver and CUDA 9.2, Ubuntu 18.10.
The thing is, the torch installer tries for some reason to use the highest compute capability supported by the device (or the driver - not sure which), but ignores the compute capability supported by the CUDA toolkit.
So, in my case the device supports compute capability 7.5, but CUDA 9.2 supports only 7.0 or 7.2 (not sure which one). You can guess I got the same nvcc fatal : Unsupported gpu architecture 'compute_75'
The solution is to force the nvcc compile options to use a lower compute capability. This can be achieved by setting the following environment variable:

export TORCH_CUDA_ARCH_LIST="7.0"

Just before running ./install.sh from the torch directory.
Note that the list may contain more than one compute capability, e.g. it can be "6.0 6.2 7.0 7.2". This will be reflected in the CUDA_NVCC_FLAGS -gencode arch=compute_70,code=sm_70 and those will be outputted to the terminal during the build.

I follow this solution but seem it does not work for me. I have tried other lower versions, but still not working.

nohboogy changed the title ~~Installation error with RTX 2080, CUDA 9.0, CUDNN 7.4~~ Installation error with RTX 2080, CUDA 9.0, CUDNN 7.4 nvcc fatal : Unsupported gpu architecture 'compute_75' Jan 21, 2019

nohboogy closed this as completed Feb 15, 2019

VitalyFedyunin mentioned this issue Jun 24, 2019

Building from source failed. Multiple errors in the printout pytorch/pytorch#22129

Closed

Kaushik-Raghavan mentioned this issue Aug 21, 2019

nvcc fatal : Unsupported gpu architecture 'compute_75' leehomyc/Faster-High-Res-Neural-Inpainting#40

Open

viven12138 mentioned this issue Oct 22, 2019

Kernel not compiled with GPU support facebookresearch/detectron2#55

Closed

VitalyFedyunin mentioned this issue Mar 19, 2020

How install old version pytorch 1.2.0 from source? pytorch/pytorch#35035

Open

indulgingcode mentioned this issue Jun 3, 2020

nvcc fatal : Unsupported gpu architecture 'compute_75' open-mmlab/mmdetection#2880

Closed

gmt710 mentioned this issue Dec 2, 2020

nvcc fatal : Unsupported gpu architecture 'compute_75' facebookresearch/detectron2#149

Closed

laygond mentioned this issue Jan 1, 2021

warning: setUpNet DNN module was not built with CUDA backend; switching to CPU opencv/opencv#15858

Closed

Xpangz mentioned this issue Mar 7, 2021

error: command 'g++' failed with exit status 1, maybe due to python version CharlesShang/DCNv2#98

Open

yuval-alaluf mentioned this issue Sep 10, 2021

ImportError: No module named 'fused' yuval-alaluf/SAM#35

Closed

Taka152 mentioned this issue Sep 28, 2021

nvcc Unsupported gpu architecture error bytedance/lightseq#189

Open

ClementPinard mentioned this issue Mar 20, 2023

python setup.py install failed ClementPinard/Pytorch-Correlation-extension#92

Closed

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Installation error with RTX 2080, CUDA 9.0, CUDNN 7.4 nvcc fatal : Unsupported gpu architecture 'compute_75' #1190

Installation error with RTX 2080, CUDA 9.0, CUDNN 7.4 nvcc fatal : Unsupported gpu architecture 'compute_75' #1190

nohboogy commented Jan 21, 2019 •

edited

Loading

zhangqijun commented Jan 24, 2019

pablovicentem commented Feb 15, 2019

nohboogy commented Feb 15, 2019

WanliXue commented Mar 28, 2019

ha-lins commented May 20, 2019 •

edited

Loading

WanliXue commented May 21, 2019

talkenig commented Jun 5, 2019 •

edited

Loading

RishengZhou commented Aug 1, 2019

talkenig commented Aug 1, 2019 •

edited

Loading

vicya commented Oct 8, 2019 •

edited

Loading

viven12138 commented Oct 22, 2019

morganmcg1 commented Jan 30, 2020

duybn-0885 commented Sep 22, 2020 •

edited

Loading

Installation error with RTX 2080, CUDA 9.0, CUDNN 7.4 nvcc fatal : Unsupported gpu architecture 'compute_75' #1190

Installation error with RTX 2080, CUDA 9.0, CUDNN 7.4 nvcc fatal : Unsupported gpu architecture 'compute_75' #1190

Comments

nohboogy commented Jan 21, 2019 • edited Loading

zhangqijun commented Jan 24, 2019

pablovicentem commented Feb 15, 2019

nohboogy commented Feb 15, 2019

WanliXue commented Mar 28, 2019

ha-lins commented May 20, 2019 • edited Loading

WanliXue commented May 21, 2019

talkenig commented Jun 5, 2019 • edited Loading

RishengZhou commented Aug 1, 2019

talkenig commented Aug 1, 2019 • edited Loading

vicya commented Oct 8, 2019 • edited Loading

viven12138 commented Oct 22, 2019

morganmcg1 commented Jan 30, 2020

duybn-0885 commented Sep 22, 2020 • edited Loading

nohboogy commented Jan 21, 2019 •

edited

Loading

ha-lins commented May 20, 2019 •

edited

Loading

talkenig commented Jun 5, 2019 •

edited

Loading

talkenig commented Aug 1, 2019 •

edited

Loading

vicya commented Oct 8, 2019 •

edited

Loading

duybn-0885 commented Sep 22, 2020 •

edited

Loading