NVIDIA GPU CLOUD PLATFORM UPDATE NVIDIA 서완석이사 wseo@nvidia.com
THEIR OPPORTUNITIES & CHALLENGES More complex designs Larger data (4K & beyond) New Cloud workflows Follow-the-Sun design teams BYOD & Mobile capabilities BIG DATA THE CLOUD COLLABORATION MOBILE
Visualize Larger Data Sets LARGER GPU MEMORY + HIGHER PERFORMANCE Implement New Cloud Workflows OPTIMIZED FOR REMOTE RENDERING Collaborate Across Teams
3D Graphic VDI with NVIDIA GRID + vgpu HP Z Virtual Workstation
NVIDIA GRID 클라우드 GPU NVIDIA GRID K2 디자이너 NVIDIA GRID K1 파워유저 GPU 4 Kepler GPUs 2 High End Kepler GPUs CUDA Cores 768 (192/GPU) 3072 (1536/GPU) Memory Size 16GB DDR3 (4GB/GPU) 8GB GDDR5 (4GB/GPU) 일반사무직 Max Power 130 W 225 W 동일수준 Quadro Quadro K600 (entry) Quadro K5000 (high end) 1 Number of users depends on software solution, workload, and screen resolution
NVIDIA GRID 그래픽옵션
VDI Virtual Workstation NVDIA GRID 와 VMware 로드맵 Q1 2013 Q4 2014 Q1 2015 VMware vdga Dedicated GPU DESIGNER View 5.3 or higher vsphere 5.1 or higher VMware vgpu View 6 or higher vsphere 6 or higher POWER USER VMware vsga Software Virtualization View 5.2 or higher vsphere 5.1 or higher KNOWLEDGE WORKER
VMware EUC vgpu 구성요소 GRID GPU 가상화 S/W 서버플랫폼 VMware vsphere 6.0 or Higher NVIDIA GRID K1 VMware Horizon View 6.0 or Higher NVIDIA GRID vgpu Manager NVIDIA GRID vgpu Driver NVIDIA GRID K2
GRID K2 GRID K1 VMware EUC vgpu Profile GPU 구분 vgpu 프로필 최대사용자 최대해상도 비디오메모리 K180Q 4 2560x1600 4GB K160Q 8 2560x1600 2GB K140Q 16 2560x1600 1GB K120Q 32 2560x1600 512MB K100 32 1920x1200 256MB K280Q 2 2560x1600 4GB K260Q 4 2560x1600 2GB K240Q 8 2560x1600 1GB K220Q 16 2560x1600 512MB K200 16 1920x1200 256MB
vgpu Early Access Program VMware vsphere 의 NVIDIA GRID vgpu 를출시전미리경험해보실수있는 Early Access Program 이제공됩니다. 이프로그램은우선엑세스요구사항을충족하는고객에게만한정적으로제공됩니다. 요구사항은다음과같습니다. VMware 계정보유 NVIDIA GRID K1/K2에서 vsga 혹은 vdga 사용중 VMware Horizon View ( 정식 / 평가판 ) 사용중 vsphere 2015 Beta Access http://www.nvidia.co.kr/object/vmware -vgpu-early-access-kr.html
CHALLENGES Photo Realistic Rendering Is Computationally Expensive Long rendering time Constrains creativity Reduces productivity Limited accessibility
GTC Keynote Recap Honda At Honda Japan: Early Iray IQ adopter 25 8xGPU servers close in spec to the Iray VCA Used for design inspection on original automotive data (typ. 40-50m triangles)
GPU RENDERING In real time on 4K screen
NVIDIA VCA (VISUAL COMPUTING APPLIANCE) Fastest GPU rendering on a single node; scalability across multiple nodes Turnkey appliance engineered by NVIDIA expressly for rendering Network availability for multiple users Rendered in Iray / Vray-RT on VCA
TIME IN SECONDS TO RENDER IMAGE Base: Xeon X5650 CPU 22 MINUTES 1291.3 Entry: Base + Quadro K4000 284.5 Mid: Base + Quadro K6000 119.4 Base: Xeon X5650 CPU Entry: Base + Quadro K4000 22.5 minutes 4.5x faster VCA 11.6 Mid: Base + Quadro K6000 11x faster 16 VCA 0.7 VCA 111x faster 16 VCAs 1800x faster 0.0 200.0 400.0 600.0 800.0 1000.0 1200.0 1400.0
Iray IQ Workflow Flexibility Image Streams Incremental Changes 3D Model Design done by HP Zbook Mobile workstation as it is, then Use VCA as render machine No Networking Expertise Required Easy to manage VCA cluster through network Iray VCAs VCA Cluster Manager Remote Clients
V-RAY RT FOR FINAL FRAME RENDERING V-Ray RT 3.0 New Features Render Elements, allowing layers of final frames to be individually rendered for assembly in compositing application 10-20 X performance acceleration over CPU-based V-Ray
GPU RENDERING FOR POPULAR 3D APPLICATIONS Autodesk 3ds Max Autodesk Maya McNeel Rhino Trimble SketchUp Dassault Systemes Bunkspeed