Fujitsu VDI / vgpu Virtualization Antti Sirkiä Service Partner Manager, Certified Trainer Fujitsu, Product Business Unit
Why Virtualization / Graphics Virtualization? :: GRAPHICS VIRTUALIZATION :: Multiple users per workstation (shared resource) Remove heat/noise from offices Enhanced Data Security Increased data security Protection against theft or damage Separate workstation and user Hardware Protection Resource sharing Free choice of client Workstation Consolidation Virtual Desktop / Virtual App w. / wo. GPU Windows 10 Rollout Working remotely Shared virtual workstations (Rendering unit etc.) One Administrator for multiple locations No need of SCCM / Miradore etc. Separate Workstation and User
To whom? Task worker VDI without vgpu Knowledge Worker MS Office, Photoshop Power User Power User PLM, Showcase, Medical imaging The users of VDI / vgpu Designer / Engineer Designer / Engineer NX, CATIA, CS6, Inventor A user of visual data (3D images, and 2D graphs and line charts). Often uses a specialized application beyond the typical Office suite and web tools. Creates and works with complicated datasets using graphics intensive applications Industry, media & entertainment, medical imaging, oil & gas, visualization, rendering Copyright 2013 FUJITSU
Fujitsu VDI-Solution for Enterprise
Backbone Storage (Water Tower) Server (water) SAN (funnel 1) Bandwidth (funnel 2) End User Experience (bottle)
Remote Graphics Solutions from Fujitsu Technical overview RemoteAccess (1:1) Pass-through / direct assigned GPU Hardware virtualized GPU Bare metal Installation of OS on host direct usage of GPU (Futro L + host) 1:1 relation 1 Host : 1 User (at the same time) Cost-effective Best graphics performance Native driver High bandwidth VMware Horizon for brokering User Apps OS Translation on special card 1:1 relation of VM with GPU (or GPU core) High-end 3D Graphics and workstation class performance for virtual desktops. Highest Performance to density Vmware / Citrix / HyperV (2016) User Apps OS Driver API Hypervisor User Apps OS Driver VMs share GPU on the host Highest flexibility through various vgpu profiles Highest costs (License, Tesla) Vmware / Citrix / HyperV (2016) User Apps OS Driver API Hypervisor User Apps OS Driver LAN WAN
Remote Graphics Solutions Products in the Data Center (Backend for VDI / vgpu) FUJITSU REMOTE GRAPHICS SOLUTION OFFERING FUJITSU CELSIUS C740 FUJITSU PRIMERGY RX2540 M4 FUJITSU PRIMERGY CX400 M4 The Fujitsu Portfolio 1U flexible workstation for 1:1, GPU pass-through or vgpu High performance 2U server with up to 2 NVIDIA GRID GPUs High density, flexible 2U server with up to 4 NVIDIA GRID GPUs Form Factor / function 1U Rack Workstation 2U Server 2U Server (with 2 x PRIMERGY CX2570 M4 HPC/VDI Server Nodes) CPUs Intel Xeon E5-1600 v4 (up to 8 cores or 3.70 GHz) or Intel Xeon E5-2600 v4 (up to 22 cores or 2.60 GHz) 1x or 2x Intel Xeon Scalable Family CPUs with 28 cores Up to Intel Xeon Scalable Family CPUs with 28 cores Memory Up to 256 GB DDR4, up to 2400 MHz registered ECC Up to 3 TB DDR4 registered ECC (2,666MHz) Up to 1 TB DDR4 registered ECC (2,666MHz) Graphics 1x NVIDIA TESLA M60 or NVIDIA Quadro M5000 or 2x NVIDIA Quadro M4000 (or below) Up to 2x NVIDIA Tesla M60 (Tesla P40 Q1 2018) Up to 4x NVIDIA Tesla M60 (Tesla P40 Q1 2018) Storage Up to 4x cold plug 2.5 drives (SSDs up to 1,024 GB/HDD SAS 10k) Up to 27x 2.5 hot-plug SAS/SATA Up to 12x 2.5 hot-plug SAS/SATA
Fujitsu Server Primergy CX2570 M4 (HPC/VDI node) Two PRIMERGY CX2570 M4 server nodes, each with latest Intel Xeon Processor Scalable Family, can be smartly packaged into a condensed 2U rack enclosure. Half-width, two-socket server node for PRIMERGY CX400 M4 chassis enabling highest computing density. Intel Xeon Processor Scalable Family with up to 28 cores relying on Intel UltraPath Interconnect for an increased data rate between the CPUs. Support for up to four high-end computational and graphic cards. 2X Tesla M60, P100, P40 (PCIe) or 4x Tesla P100, P4 (SXM2,NVLINK) http://www.fujitsu.com/fts/products/computing/servers/primergy/scale-out/cx2570m4/ http://www.fujitsu.com/fts/products/computing/servers/primergy/scale-out/cx400m4/
Frontend - Many possibilities - BYOD STYLISTIC Tablets LIFEBOOK 2 in 1 FUTRO Thin Clients ipad / iphone ESPRIMO Desktops CELSIUS Mobile Workstations
Nvidia GRID: Maximum power for vgpu NVIDIA Tesla-card
NVIDIA vgpu GPU Manager as HyperVisor Add-on in Dom0 1 to 1 combination GPU Manager + Graphic driver in VM vgpu profiles depending on user needs (different type of graphics) Splitting of GPU power depending vgpu profile + concurrent usage GRID 2.0: no more Grid card Tesla M60 will be used in Graphic mode therefore license and License Management server is needed. GRID 5.0: No support for GRID K-series! http://docs.nvidia.com/grid/ Virtual Machine Virtual Machine Virtual Machine Virtual Machine * Grid 2.0 OS ( Linux */Windows) Applications Graphics Driver OS ( Linux */Windows) Applications Graphics Driver OS ( Linux */Windows) Applications Graphics Driver OS ( Linux */Windows) Applications Graphics Driver Virtual Machine Virtual Machine Virtual Machine Virtual Machine OS ( Linux */Windows) Applications Graphics Driver OS ( Linux */Windows) Applications Graphics Driver OS ( Linux */Windows) Applications Graphics Driver OS ( Linux */Windows) Applications Graphics Driver Nvidia Grid Dom0 Manager HyperVisor: Citrix / Microsoft Server 2016 / VMware Server/WS http://www.nvidia.com/object/grid-technology.html
NVIDIA Tesla M60 vgpu solution since GRID 2.0: NVIDIA Tesla M60 card and NVIDIA GRID software Successor of GRID K2 GRID cards are discontinued. Software is still named GRID. Since now Tesla is for vgpu and Supercomputing NVIDIA GRID gives three different licensing models: NVIDIA GRID Virtual Applications, Virtual PC, Virtual Workstation Cuda cores: 4096 (2x 2048) VRAM: 16GB GDDR5 (2x 8GB) vgpu profiles: 1,2,4,8GB Framebuffer x2 Data sheet: http://images.nvidia.com/content/pdf/grid/data-sheet/nvidia-grid20-ds.pdf
NVIDIA Tesla P40 Successor of Tesla M60, available Q1 2018 GRID 5.0 -> no support for GRID K-cards! vws -> Quadro vdws (Nvidia Quadro virtual datacenter workstation) GPU: 1x Nvidia Pascal GPU vgpu profiles: 1,2,3,4,6,8,12 and 24GB Framebuffer Cuda cores: 3,840 VRAM: 24GB Datasheet: http://images.nvidia.com/content/pdf/grid/data-sheet/nvidia-p40- datasheet.pdf
vgpu type Change through XenCenter / vsphere Web Client All supported types available. GRID and Tesla M60 can be in same Server There are no such a GUI in vsphere environment! Check through console with nvidia-smi-command
Remote Graphics Solutions Different vgpu scenarios (1/3) Celsius 1x NVIDIA Tesla M60 CELSIUS C740 OR CELSIUS R940 Top 5 Benefits Performance depending on your needs (CPU, GPU) Scalability depending on your needs Flexible Implementation Local data stored on the SSD C740: 1U / 1 Tesla / 1 fast CPU / max. 256GB RAM R940: Tower / 3 Tesla / 2 fast CPU / max. 1,024GB RAM 3x NVIDIA Tesla M60 NVIDIA Tesla M60 (2 physical GPUs) vgpu Profile Graphics Memory Amount of vgpus per phys. GPU Max. Users per GRID Card (3x Tesla M60) M ax. Resolution M60-8Q 8192 MB 1 6 3840x2160 M60-4Q 4096 MB 2 12 3840x2160 M60-2Q 2048 MB 4 24 2560x1600 M60-1Q 1024 MB 8 48 2560x1600
Remote Graphics Solutions Different vgpu scenarios (2/3) Primergy PRIMERGY RX2540M2 2x NVIDIA Tesla M60 Top 5 Benefits Performance depending on your needs (CPU, GPU) Scalability depending on your needs Flexible Implementation Redundancy (LAN/Power Supply) 2U / 2 Tesla / 2 fast CPU / max. 3TB RAM NVIDIA Tesla M60 (2 physical GPUs) vgpu Profile Graphics Memory Amount of vgpus per phys. GPU Max. Users 2x GRID Card M ax. Resolution M60-8Q 8192 MB 1 4 3840x2160 M60-4Q 4096 MB 2 8 3840x2160 M60-2Q 2048 MB 4 16 2560x1600 M60-1Q 1024 MB 8 32 2560x1600
Remote Graphics Solutions Different vgpu scenarios (3/3) Primergy PRIMERGY CX400 M1 Top 3 Benefits Highest density (2U) Rack Server Ultra flexible e.g. Up to 128 VM on 2U with K620 graphics 2U / 4 Tesla / 4 nodes 4x NVIDIA Tesla M60 NVIDIA Tesla M60 (2 physical GPUs) vgpu Profile Graphics Memory Amount of vgpus per phys. GPU Max. Users 4x GRID Card M ax. Resolution M60-8Q 8192 MB 1 8 3840x2160 M60-4Q 4096 MB 2 16 3840x2160 M60-2Q 2048 MB 4 32 2560x1600 M60-1Q 1024 MB 8 64 2560x1600
How to prove that virtual solutions really works? Demo environment (Helsinki) Proof Of Concept aka. POC Customer own environment with our HW
Nvidia GRID vgpu demo environment Solution Center Fujitsu Finland Primergy RX2540M2 Nvidia Grid K2 Citrix XenServer 7.1 XenApp/Desktop 7.15 VPN-tunnel to the Solution Center Primergy RX2540 M2 2xTesla M60 Vmware vsphere 6.5u1 Horizon 7.3 / Xen 7.15