How to install Torch with new Turing GPU (RTX) - nvcc fatal: unsupported gpu architecture

[blog...@gmail.com]

Hello,

I have the following problem. I have a new Turing GPU (RTX). I want to install Torch.

I have CUDA 9.0, because Tensorflow supports 9.0.

And Tensorflow is working. Now, when i want to install Torch i get the error:

"nvcc fatal   : Unsupported gpu architecture 'compute_75'". I tried to manually add "7.0"/"7.5" in "select_compute_arch.cmake", but nothing changed. What can I do?

 

-- 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/joe/torch/extra/cutorch/build
[  2%] Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir/THC_generated_THCBlas.cu.o
[  2%] Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir/THC_generated_THCHalf.cu.o
[  5%] 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
[  4%] Building NVCC (Device) object lib/THC/CMakeFiles/THC.dir/THC_generated_THCReduceApplyUtils.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'
nvcc fatal   : Unsupported gpu architecture 'compute_75'
nvcc fatal   : Unsupported gpu architecture 'compute_75'

Thank you.

Kind Regards

[blog...@gmail.com]

However, I found out how to do so. It is probably not working with a CUDA 9.0 version (prove me wrong). But I installed CUDA 10.0, and used the advice in this thread: https://github.com/torch/cutorch/issues/834. Now, I can use Torch and TensorFlow together, with CUDA 10.0.