search

flashlight / wav2letter

Facebook AI Research's Automatic Speech Recognition Toolkit
https://github.com/facebookresearch/wav2letter/wiki
Other
6.37k stars 1.01k forks source link

Unsupported gpu architecture 'compute_30 [gpu :RTX 3090] #924

Closed lovegu0713 closed 3 years ago

lovegu0713 commented 3 years ago

GPU: nvidia RTX3090: when I install wav2ltter use " pip install -e ." ,then get some error:

ERROE Description

[ 11%] Building NVCC (Device) object src/libraries/criterion/CMakeFiles/w2l-criterion-library-cuda.dir/cuda/w2l-criterion-library-cuda_generated_CriterionUtils.cu.o
[ 11%] Building NVCC (Device) object src/libraries/criterion/CMakeFiles/w2l-criterion-library-cuda.dir/cuda/w2l-criterion-library-cuda_generated_FullConnectionCriterion.cu.o
[ 11%] Building NVCC (Device) object src/libraries/criterion/CMakeFiles/w2l-criterion-library-cuda.dir/cuda/w2l-criterion-library-cuda_generated_ViterbiPath.cu.o
[ 11%] Building NVCC (Device) object src/libraries/criterion/CMakeFiles/w2l-criterion-library-cuda.dir/cuda/w2l-criterion-library-cuda_generated_ForceAlignmentCriterion.cu.o
nvcc fatal   : Unsupported gpu architecture 'compute_30'
nvcc fatal   : Unsupported gpu architecture 'compute_30'
nvcc fatal   : Unsupported gpu architecture 'compute_30'
CMake Error at w2l-criterion-library-cuda_generated_FullConnectionCriterion.cu.o.Release.cmake:220 (message):
  Error generating
  /home/Test/zhaojing/wav2letter/bindings/python/build/temp.linux-x86_64-3.8/src/libraries/criterion/CMakeFiles/w2l-criterion-library-cuda.dir/cuda/./w2l-criterion-library-cuda_generated_FullConnectionCriterion.cu.o

CMake Error at w2l-criterion-library-cuda_generated_ViterbiPath.cu.o.Release.cmake:220 (message):
  Error generating
  /home/Test/zhaojing/wav2letter/bindings/python/build/temp.linux-x86_64-3.8/src/libraries/criterion/CMakeFiles/w2l-criterion-library-cuda.dir/cuda/./w2l-criterion-library-cuda_generated_ViterbiPath.cu.o

Wav2ltter Unsupported GPU RTX3090?

lovegu0713 commented 3 years ago

I solved the problem： RTX3090 is not supported “compute_30”

nvcc --list-gpu-arch

log

compute_35 compute_37 compute_50 compute_52 compute_53 compute_60 compute_61 compute_62 compute_70 compute_72 compute_75 compute_80 compute_86

so I set "set(CUDA_COMMON_GPU_ARCHITECTURES "3.0" "3.5" "5.0")" ->"set(CUDA_COMMON_GPU_ARCHITECTURES "3.5" "5.0")" and "set(CUDA_ALL_GPU_ARCHITECTURES "3.0" "3.2" "3.5" "5.0")" -> "set(CUDA_ALL_GPU_ARCHITECTURES "3.5" "5.0")" in 'cmake/select_compute_arch.cmake'

finally,add 'add_compile_definitions(THRUST_IGNORE_CUB_VERSION_CHECK)' line to the CMakeLists.txt

this

----------------------------- Configuration ----------------------------- option(W2L_BUILD_LIBRARIES_ONLY "Build wav2letter-libraries only" OFF) option(W2L_BUILD_INFERENCE "Build inference even when W2L_BUILD_LIBRARIES_ONLY is set" OFF) option(W2L_LIBRARIES_USE_CUDA "Use CUDA in libraries-only build" ON) option(W2L_LIBRARIES_USE_KENLM "Use KenLM in libraries-only build" ON) option(W2L_LIBRARIES_USE_MKL "Use MKL in libraries-only build" ON) option(W2L_BUILD_FOR_PYTHON "Build Python bindings" OFF) option(W2L_BUILD_TESTS "Build tests for wav2letter++" ON) option(W2L_BUILD_EXAMPLES "Build examples for wav2letter++" ON) option(W2L_BUILD_EXPERIMENTAL "Build internal experimental components for wav2letter++" OFF) option(W2L_BUILD_SCRIPTS "Build internal scripts for wav2letter++" OFF) option(W2L_BUILD_RECIPES "Build recipes" ON) option(W2L_BUILD_TOOLS "Build audio tools" OFF) set(KENLM_MAX_ORDER 6 CACHE STRING "Maximum ngram order for KenLM") add_compile_definitions(THRUST_IGNORE_CUB_VERSION_CHECK)