Ubuntu 18.04 安裝NVIDIA Driver 418 & CUDA 10 & Miniconda
文章推薦指數: 80 %
如果不太相信我,可以關掉這一頁,直接照著官網教學來安裝也沒問題XD。
安裝過程:. 0. 將NVIDIA repository 將入你的apt wget https://developer.download.nvidia.com/ ...
UpgradeOpeninappHomeNotificationsListsStoriesWriteUbuntu18.04安裝NVIDIADriver418&CUDA10&Miniconda&TensorFlow1.13作業系統:Kubuntu18.04(Ubuntu的KDE版本)顯示卡:NVIDIAGeForceGTX1080Ti預計目標是可以成功執行TensorFlow1.13的GPU版本。
參考:https://www.tensorflow.org/install/gpu#software_requirementsThefollowingNVIDIA®softwaremustbeinstalledonyoursystem:NVIDIA®GPUdrivers—CUDA10.0requires410.xorhigher.CUDA®Toolkit—TensorFlowsupportsCUDA10.0(TensorFlow>=1.13.0)CUPTIshipswiththeCUDAToolkit.cuDNNSDK(>=7.4.1)(Optional)TensorRT5.0toimprovelatencyandthroughputforinferenceonsomemodels.重點:安裝NVIDIAdriver版本410以上安裝CUDA10.0安裝CUPTI安裝cuDNN7.4.1以上(可選)安裝TensorRT5.0版本號要特別注意,經過測試以後發現如果使用最新的CUDA10.1是沒辦法跑的...另外,TensorFlow官網提供了幾乎最方便通用的安裝方式,參考網址:https://www.tensorflow.org/install/gpu所以基本上照著官網指示安裝就可以用了。
我以下記錄的指令依照個人喜好稍作修改。
如果不太相信我,可以關掉這一頁,直接照著官網教學來安裝也沒問題XD。
安裝過程:0.將NVIDIArepository將入你的aptwgethttps://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/cuda-repo-ubuntu1804_10.0.130-1_amd64.debsudodpkg-icuda-repo-ubuntu1804_10.0.130-1_amd64.debsudoapt-keyadv--fetch-keyshttps://developer.download.nvidia.com/compute/cuda/repos/ubuntu1804/x86_64/7fa2af80.pubsudoaptupdatewgethttp://developer.download.nvidia.com/compute/machine-learning/repos/ubuntu1804/x86_64/nvidia-machine-learning-repo-ubuntu1804_1.0.0-1_amd64.debsudoaptinstall./nvidia-machine-learning-repo-ubuntu1804_1.0.0-1_amd64.debsudoaptupdate1.安裝NVIDIADriver首先要查看一下有沒有內建可裝的driverubuntu-driversdevices輸出結果==/sys/devices/pci0000:00/0000:00:01.1/0000:02:00.0==modalias:pci:v000010DEd00001B06sv00003842sd00006696bc03sc00i00vendor:NVIDIACorporationmodel:GP102[GeForceGTX1080Ti]driver:nvidia-driver-390-distronon-freedriver:nvidia-driver-410-third-partyfreedriver:nvidia-driver-418-third-partyfreerecommendeddriver:xserver-xorg-video-nouveau-distrofreebuiltin由此可知前面成功加入NVIDIArepository。
如果失敗的話會看到只有`nvidia-driver-390`這一項,這樣是沒辦法安裝CUDA10的。
開始安裝sudoubuntu-driversautoinstall輸出結果(前略)Thefollowingpackageshaveunmetdependencies:nvidia-driver-418:Depends:xserver-xorg-video-nvidia-418(=418.39-0ubuntu1)butitisnotgoingtobeinstalledRecommends:libnvidia-compute-418:i386(=418.39-0ubuntu1)butitisnotinstallableRecommends:libnvidia-decode-418:i386(=418.39-0ubuntu1)butitisnotinstallableRecommends:libnvidia-encode-418:i386(=418.39-0ubuntu1)butitisnotinstallableRecommends:libnvidia-ifr1-418:i386(=418.39-0ubuntu1)butitisnotinstallableRecommends:libnvidia-fbc1-418:i386(=418.39-0ubuntu1)butitisnotinstallableRecommends:libnvidia-gl-418:i386(=418.39-0ubuntu1)butitisnotinstallableE:Unabletocorrectproblems,youhaveheldbrokenpackages.表示缺少xserver-xorg-video-nvidia-418,於是安裝它sudoaptinstallxserver-xorg-video-nvidia-418輸出結果Thefollowingpackageshaveunmetdependencies:xserver-xorg-video-nvidia-418:Depends:xserver-xorg-core(>=2:1.19.6-1ubuntu2~)E:Unabletocorrectproblems,youhaveheldbrokenpackages.表示缺少xserver-xorg-core,於是安裝它sudoaptinstallxserver-xorg-core安裝完成以後再輸入一次sudoubuntu-driversautoinstall安裝完成以後重新開機sudoreboot重新開機以後測試一下nvidia-smi輸出結果MonMar1121:20:522019+-----------------------------------------------------------------------------+|NVIDIA-SMI418.39DriverVersion:418.39CUDAVersion:10.1||-------------------------------+----------------------+----------------------+|GPUNamePersistence-M|Bus-IdDisp.A|VolatileUncorr.ECC||FanTempPerfPwr:Usage/Cap|Memory-Usage|GPU-UtilComputeM.||===============================+======================+======================||0GeForceGTX108...On|00000000:01:00.0Off|N/A||0%38CP817W/280W|1MiB/11175MiB|0%Default|+-------------------------------+----------------------+----------------------+|1GeForceGTX108...On|00000000:02:00.0Off|N/A||0%33CP811W/280W|1MiB/11178MiB|0%Default|+-------------------------------+----------------------+----------------------++-----------------------------------------------------------------------------+|Processes:GPUMemory||GPUPIDTypeProcessnameUsage||=============================================================================||Norunningprocessesfound|+-----------------------------------------------------------------------------+可以正常顯示,看來是沒問題了。
2.安裝CUDA10&cuDNN7.4這一步幾乎直接抄官網的。
sudoaptinstall--no-install-recommendscuda-10-0libcudnn7=7.4.1.5-1+cuda10.0這樣就沒問題了。
其中,安裝cuDNN會比較容易一點,也可以拆成兩步:2.1.安裝CUDA10sudoaptinstall--no-install-recommendscuda-10-02.2.安裝cuDNN從官網https://developer.nvidia.com/rdp/cudnn-download下載,挑選RuntimeLibrary"cuDNNRuntimeLibraryforUbuntu18.04(Deb)"只要版本大於7.4.1都可以,因為cuDNN有向後相容,可以盡量安裝最新的沒煩惱。
sudodpkg-i
3.安裝TensorRTsudoaptupdate&&sudoapt-getinstallnvinfer-runtime-trt-repo-ubuntu1804-5.0.2-ga-cuda10.0&&sudoaptupdate&&sudoaptinstall-y--no-install-recommendslibnvinfer-dev=5.0.2-1+cuda10.0毫無反應,就只是安裝成功。
...4.安裝TensorFlow1.13我打算用Miniconda安裝,所以先安裝Miniconda4.1安裝Minicondawgethttps://repo.anaconda.com/miniconda/Miniconda3-latest-Linux-x86_64.shsudobashMiniconda3-latest-Linux-x86_64.sh出現license選yesMiniconda3willnowbeinstalledintothislocation:/home/
最後出現這個DoyouwishtheinstallertoinitializeMiniconda3inyour/home/
登出之後重新登入,然後輸入conda輸出usage:conda[-h][-V]command......(略)看來是沒問題了。
4.2建立虛擬環境condacreate-ntfpython=34.3啟動虛擬環境condaactivatetf4.4接下來按照官網提示安裝Tensorflow:pipinstall--upgradetensorflow-gpu我做到這邊一路沒發生什麼問題。
接下來就測試一下有沒有問題:python...>>>importtensorflowastf>>>tf.test.gpu_device_name()這邊如果沒有出現錯誤,那就成功了。
如果有任何問題,歡迎在底下留言討論。
MorefromManiac.twFollowLovepodcastsoraudiobooks?Learnonthegowithournewapp.TryKnowableGetstartedManiac.tw17FollowersFollowRelatedInstallingCUDAonheadlessUbuntuserverPrerequisites:NewlyinstalledUbuntu20.04onmachinewithNVidiaGPU.CUDAonSolusSolusisanindependentcuratedrollingreleaselinuxdistribution.Onceinstalled,itisoneofthemosthasslefreelinuxdistributionsI…TheRustProgrammingLanguage — Chapter3ThesearemynotesforChapter3 — CommonProgrammingProjectsWindowsiteratorIcontinuetocrawlthroughleetcodeproblems,andIfoundthatIusewindowsinabout30%ofproblemsIsolve.(donotconfuseitwitha…HelpStatusWritersBlogCareersPrivacyTermsAboutKnowable
延伸文章資訊
- 1CUDA Toolkit 11.0 Download | NVIDIA Developer
CUDA Toolkit 11.0 Download ... By downloading and using the software, you agree to fully comply w...
- 2CUDA Toolkit 11.1 Update 1 Downloads | NVIDIA Developer
CUDA Toolkit 11.1 Update 1 Downloads. Please Note: We advise customers updating to Linux Kernel 5...
- 3CUDA Toolkit Archive | NVIDIA Developer
- 4Start Locally | PyTorch
CUDA 10.2. CUDA 11.3. ROCm 4.2 (beta). CPU. Run this Command: conda install pytorch torchvision t...
- 5CUDA Toolkit 11.6 Update 1 Downloads | NVIDIA Developer
Resources CUDA Documentation/Release NotesMacOS Tools Training Sample Code Forums Archive of Prev...