NVIDIA CUDA cuDNN Caffe vs. Intel CPUs On Linux Run NVIDIA CUDA Caffe build versus CPU-only Caffe build with both AlexNet and GoogleNet. Tests by Michael Larabel and all tests on same system sans switching out CPUs for other CPU-only runs.
HTML result view exported from: https://openbenchmarking.org/result/1610067-STYL-160829608&grw .
NVIDIA CUDA cuDNN Caffe vs. Intel CPUs On Linux Run Processor Motherboard Chipset Memory Disk Graphics Audio Network OS Kernel Desktop Display Server Display Driver OpenGL Vulkan Compiler File-System Screen Resolution GeForce GTX TITAN X GeForce GTX 1080 GeForce GTX 950 GeForce GTX 1080 Zotac Intel Xeon E5-2609 v4 @ 1.70GHz (8 Cores) MSI X99A WORKSTATION (MS-7A54) v1.0 Intel Xeon E7 v4/Xeon 16384MB 3 x 120GB TOSHIBA-TR150 NVIDIA GeForce GTX TITAN X 12288MB (1001/3505MHz) Realtek ALC1150 Intel Connection Ubuntu 16.04 4.4.0-34-generic (x86_64) Unity 7.4.0 X Server 1.18.3 NVIDIA 367.35 4.5.0 1.0.8 GCC 5.4.0 20160609 + CUDA 8.0 ext4 2560x1440 NVIDIA GeForce GTX 1080 8192MB (1604/5005MHz) eVGA NVIDIA GeForce GTX 950 2048MB (1202/3304MHz) Intel Core i7-6700 @ 4.00GHz (8 Cores) Gigabyte H170-Gaming 3 Intel Sky Lake 63488MB 1000GB Seagate ST1000DM003-1SB1 + Samsung SSD 950 PRO 256GB Zotac NVIDIA Device 1b80 Qualcomm Atheros Killer E220x Gigabit 4.4.0-38-generic (x86_64) 800x600 OpenBenchmarking.org Compiler Details - --build=x86_64-linux-gnu --disable-browser-plugin --disable-vtable-verify --disable-werror --enable-checking=release --enable-clocale=gnu --enable-gnu-unique-object --enable-gtk-cairo --enable-java-awt=gtk --enable-java-home --enable-languages=c,ada,c++,java,go,d,fortran,objc,obj-c++ --enable-libmpx --enable-libstdcxx-debug --enable-libstdcxx-time=yes --enable-multiarch --enable-multilib --enable-nls --enable-objc-gc --enable-plugin --enable-shared --enable-threads=posix --host=x86_64-linux-gnu --target=x86_64-linux-gnu --with-abi=m64 --with-arch-32=i686 --with-arch-directory=amd64 --with-default-libstdcxx-abi=new --with-multilib-list=m32,m64,mx32 --with-tune=generic -v Processor Details - GeForce GTX TITAN X: Scaling Governor: intel_pstate performance - GeForce GTX 1080: Scaling Governor: intel_pstate performance - GeForce GTX 950: Scaling Governor: intel_pstate performance - GeForce GTX 1080 Zotac: Scaling Governor: intel_pstate powersave
NVIDIA CUDA cuDNN Caffe vs. Intel CPUs On Linux Run caffe: 200 - AlexNet caffe: CPU AlexNet caffe: 200 - Googlenet caffe: CPU Googlenet GeForce GTX TITAN X GeForce GTX 1080 GeForce GTX 950 GeForce GTX 1080 Zotac Xeon E5-2687W v3 Core i7-5960X Xeon E5-2609 v4 2297.70 6410.67 1990.28 5391.08 6141.34 14154.97 1793.67 357507 4507.88 707479 427198 427198 825606 825606 424764 424764 814653 814653 831629 831629 1603103 1603103 OpenBenchmarking.org
Caffe AlexNet Iterations: 200 - Build: AlexNet OpenBenchmarking.org Milli-Seconds, Fewer Is Better Caffe AlexNet 2016-06-11 Iterations: 200 - Build: AlexNet Xeon E5-2687W v3 Core i7-5960X Xeon E5-2609 v4 GeForce GTX TITAN X GeForce GTX 1080 GeForce GTX 950 GeForce GTX 1080 Zotac 200K 400K 600K 800K 1000K SE +/- 561.17, N = 3 SE +/- 627.11, N = 3 SE +/- 4812.82, N = 3 SE +/- 4.23, N = 3 SE +/- 3.77, N = 3 SE +/- 7.74, N = 3 SE +/- 0.66, N = 3 427198.00 424764.00 831629.00 2297.70 1990.28 6141.34 1793.67 1. (CXX) g++ options: -pthread -fPIC -O2 -lcaffe -lglog -lgflags -lprotobuf -lboost_system -lboost_filesystem -lm -lhdf5_hl -lhdf5 -lleveldb -lsnappy -llmdb -lopencv_core -lopencv_highgui -lopencv_imgproc -lboost_thread -lstdc++ -lcblas -latlas
Caffe AlexNet Build: CPU AlexNet OpenBenchmarking.org Milli-Seconds, Fewer Is Better Caffe AlexNet 2016-06-11 Build: CPU AlexNet Xeon E5-2687W v3 Core i7-5960X Xeon E5-2609 v4 GeForce GTX 1080 Zotac 200K 400K 600K 800K 1000K SE +/- 561.17, N = 3 SE +/- 627.11, N = 3 SE +/- 4812.82, N = 3 SE +/- 235.33, N = 3 427198 424764 831629 357507 1. (CXX) g++ options: -pthread -fPIC -O2 -lcaffe -lglog -lgflags -lprotobuf -lboost_system -lboost_filesystem -lm -lhdf5_hl -lhdf5 -lleveldb -lsnappy -llmdb -lopencv_core -lopencv_highgui -lopencv_imgproc -lboost_thread -lstdc++ -lcblas -latlas
Caffe AlexNet Iterations: 200 - Build: Googlenet OpenBenchmarking.org Milli-Seconds, Fewer Is Better Caffe AlexNet 2016-06-11 Iterations: 200 - Build: Googlenet Xeon E5-2687W v3 Core i7-5960X Xeon E5-2609 v4 GeForce GTX TITAN X GeForce GTX 1080 GeForce GTX 950 GeForce GTX 1080 Zotac 300K 600K 900K 1200K 1500K SE +/- 1923.36, N = 3 SE +/- 737.95, N = 3 SE +/- 2962.73, N = 3 SE +/- 17.08, N = 3 SE +/- 3.61, N = 3 SE +/- 6.83, N = 3 SE +/- 5.49, N = 3 825606.00 814653.00 1603103.00 6410.67 5391.08 14154.97 4507.88 1. (CXX) g++ options: -pthread -fPIC -O2 -lcaffe -lglog -lgflags -lprotobuf -lboost_system -lboost_filesystem -lm -lhdf5_hl -lhdf5 -lleveldb -lsnappy -llmdb -lopencv_core -lopencv_highgui -lopencv_imgproc -lboost_thread -lstdc++ -lcblas -latlas
Caffe AlexNet Build: CPU Googlenet OpenBenchmarking.org Milli-Seconds, Fewer Is Better Caffe AlexNet 2016-06-11 Build: CPU Googlenet Xeon E5-2687W v3 Core i7-5960X Xeon E5-2609 v4 GeForce GTX 1080 Zotac 300K 600K 900K 1200K 1500K SE +/- 1923.36, N = 3 SE +/- 737.95, N = 3 SE +/- 2962.73, N = 3 SE +/- 363.75, N = 3 825606 814653 1603103 707479 1. (CXX) g++ options: -pthread -fPIC -O2 -lcaffe -lglog -lgflags -lprotobuf -lboost_system -lboost_filesystem -lm -lhdf5_hl -lhdf5 -lleveldb -lsnappy -llmdb -lopencv_core -lopencv_highgui -lopencv_imgproc -lboost_thread -lstdc++ -lcblas -latlas
Phoronix Test Suite v10.8.5