주식회사미루웨어 deep learning 개발머쉰 미루웨어는 NVIDIA GPU Computing / GPU 가상화분야솔루션제공공식파트너사입니다. http://www.miruware.com / miruware@miruware.com T : 02-562-8993 / F : 02-562-8994
Deep Learning 개발환경
Unutu 장점 ( 개발머쉰 ) 캐노니컬은 6개월마다새로운배포판을출시핚다는일정을유지다양핚바이러니패키지제공최신의라이브러리제공쉬운패키지관리및시스템관리 (apt-get) 다양핚장비의지원 (PPC, ARM, 안드로이드.)
GPU 선택 NVIDIA GPU Model GeForce GTX itan X Tesla K40 Double-precision (64-bit) Floating Point Performance 0.2 TFLOPS 1.68 TFLOPS - GPU Direct RDMA - The Direct Memory Access (DMA) - Hyper-Q - GPU Health Monitoring and Management Capabilities (NVML/nvidia-smi, OOB(out of band monitoring via IPMI), TCC, ECC - GPU Boost - 연산오류보장
GPU 선택 Feature Tesla Geforce PCI-Express Generation (e.g., 2.0 vs 3.0) yes - PCI-Express Link Width (e.g., x4, x8, x16) yes - Current Performance State yes - Clock Throttle Status yes - Current GPU Usage (percentage) yes - Current Memory Usage (percentage) yes yes GPU Boost Capability yes yes^ ECC Error Detection/Correction Support yes - Current Power Draw yes - Set Power Draw Limit yes - Current GPU Clock Speed yes - Current Memory Clock Speed yes - Show Available Clock Speeds yes - Show Available Memory Speeds yes - Set GPU Boost Speed (core clock and memory clock) yes - Show Current Compute Processes yes - Card Serial Number yes -
GPU-Direct RDMA
GPU-Direct RDMA
GPU-Direct RDMA
Multi GPU
CAFFE 란무엇인가? Deep learning 을위핚묘듈화된프레임워크 Berkeley Vision and Learning Center 에서개발및제공 UC 버클리에서박사학위동안프로젝트로재발. BSD 라이선스로배포 GPU 가속지원
CAFFE 란무엇인가?
Nvidia Driver 설치 필수패키지설치 - apt-get install default-jre default-jre-headless fonts-dejavu-extra Nouveau Blacklist 추가 - etc/modprobe.d/blacklist-nouveau.conf 수정 blacklist nouveau options nouveau modeset=0 CUDA SDK 설치 - /etc/init.d/lightdm stop - Nvidia Driver 설치 (sh./cuda-7.0~~~~~.run) Cuda 환경변수설정 PATH=$CUDA_HOME/bin:$PATH LD_LIBRARY_PATH=$CUDA_HOME/lib64:$LD_LIBRARY_PATH
Caffe 설치 필수패키지설치 - sudo apt-get install python-numpy python-scipy python-dev python-pip\ python-nose g++ libopenblas-dev git libopenblas-dev Caffe 설치준비 - wget https://github.com/bvlc/caffe/archive/master.zip - wget http://developer.download.nvidia.com/~~~cudnn-6.5-linux-x64-v2.tgz options nouveau modeset=0 - libcudnn, cudnn.h 각각 cuda 의 lib 와 include 에복사 Caffe 설치 - pip install -r /home/digits/scr/caffe/caffe-master/python/requirements.txt - Makefile.config 수정
Nvidia Digits 설치 https://developer.nvidia.com/digits $) tar xvf digits-1.0.tgz $) cd digits-1.0 $)./one_time.sh $)./runme.sh
의존성패키지설명 libatlas-base-dev libboost-all-dev libopencv-dev libprotobuf-dev libgoogle-glog-dev libgflags-dev libhdf5-dev libleveldb-dev liblmdb-dev libsnappy-dev 선형대수학알고리즘 C++ source libraries opencv Protocol Buffers compiler for C++ headers and libraries. 애플케이션레벨의라이브러리구현 commandline flags module for C++ 과학적데이터저장하는포맷 google의가벼운유라이브러리 Memory-Mapped Database development files 압축 / 압축해제라이브러리
Nvidia Digits BOX - Four TITAN X with 12GB of memory per GPU - 64GB DDR4 - with 4-way PCI-E Gen3 x16 support - 6 Core desktop processor - 1500W ~ 1600W Power Supply Unit from - Ubuntu 14.04 - NVIDIA-qualified driver - NVIDIA CUDA Toolkit 7.0 - NVIDIA DIGITS SW - Caffe, Theano, Torch, BIDMach
Nvidia Digits BOX
Nvidia Digits BOX
분야별마케팅및젂략 NVIDIA Tesla GPU Clusters - Large Cluster Scaling for GPU-Accelerated Applications MW- Tesla130-1000 3x NVIDIA Tesla GPUs 2x Intel Xeon E5-2600 v3 CPUs 1U Server, Up to 1TB DDR4 MW- Tesla140-2560 4x NVIDIA Tesla GPUs 2x Intel Xeon E5-2600 v3 CPUs 1U Server, Up to 256GB DDR4 Compute Node MW- Tesla240-5120 4x NVIDIA Tesla GPUs 2x Intel Xeon E5-2600 v3 CPUs 1U Server, Up to 512GB DDR4 MW- Tesla240-1000 4x NVIDIA Tesla GPUs 2x Intel Xeon E5-2600 v3 CPUs 2U Server, Up to 1TB DDR4 MW- Tesla440-5120 4x NVIDIA Tesla GPUs 2x Intel Xeon E5-2600 v2 Ivy Bridge CPUs 4U Server, Up to 512GB DDR3 MW- Tesla480-7680 8x NVIDIA Tesla GPUs 2x Intel Xeon E5-2600 v3 CPUs 4U Server, Up to 768GB DDR3
감사합니다. ^^ 주식회사미루웨어