april 4-7, 2016 | silicon valley s6385 - what is cloud and...
TRANSCRIPT
April 4-7, 2016 | Silicon Valley
Matt Coppinger Director, Technical Marketing, End User Computing, VMware
Stephane Asselin Senior EUC Architect, Technical Marketing, End User Computing, VMware
Jeff Weiss GRID Solutions Architect Manager, NVIDIA
S6385 - WHAT IS CLOUD AND WHAT CAN IT DO FOR YOUR DESKTOP WORKLOADS
2
AGENDA
Why GPU in the Cloud?
Key technologies for GPU in the Cloud
Performance & Sizing Best Practices
Deployment Quick Start
Summary
Resources & Next Steps
3
AGENDA
Why GPU in the Cloud?
Key technologies for GPU in the Cloud
Performance & Sizing Best Practices
Deployment Quick Start
Summary
Resources & Next Steps
4
MAJOR MARKETPLACE DEMAND FOR CLOUD-BASED WORKSTATIONS
4
Large datasets are harder to share
Data leakage & corporate security
More distributed than ever
Contractors need revocable access
5
TRYGRID DEMO
6
Desktop Virtualization with Server-side GPU
Improves the Delivery of Graphics
IT Manages Centralized Virtual Desktops
Users Remotely Access from Various Devices
DESKTOP, APP + GPU VIRTUALIZATION ENABLES CLOUD-BASED WORKSTATIONS
6
Security &
Compliance
Anywhere
Access
Business Agility
Mobile & BYOD
Access
OpEx Savings
Highly Efficient
Remote Protocol
+
7
DESIGNER
POWER USER
TASK WORKER
KNOWLEDGE WORKER
Source: Gartner & NVIDIA 2012 Research
WHO BENEFITS THE MOST FROM GPU?
7
8
THE IMPORTANCE OF GPUHIGH-PERFORMANCE GRAPHICS APPLICATIONS
8
Media Engineering & DesignPLM & Volume Design
9
Heavy Users
Many Applications
Task
Worker
Knowledge Worker Power User Designer
Image
Quality
Interactivity
Cost/Seat
2D / 3D
Light Users
Fewer Applications
HIGH-PERFORMANCE GRAPHICS USE CASES
9
Aero
Productivity Apps with Graphics
Diagnostics
WebGL
3D Viewers
Lightweight CAD
Oil & Gas
Engineering
Architecture Design
Media & Graphics
GIS
10
Soft 3D
Software 3D
vSGA
Shared Graphics
Acceleration
vDGA
Dedicated
Graphics
Acceleration
NVIDIA GRID VGPU
Mediated Passthrough
Heavy Users
Many Applications
Image
Quality
Interactivity
Cost/Seat
2D / 3D
Light Users
Fewer Applications
DELIVERING A COMPLETE PORTFOLIO, FOR EVERY WORKER, EVERY USE CASE
1
0
Aero
Productivity Apps with Graphics
Diagnostics
WebGL
3D Viewers
Lightweight CAD
Oil & Gas
Engineering
Architecture Design
Media & Graphics
GIS
Task
Worker
Knowledge Worker Power User Designer
11
VIRTUAL BATTLESPACE IN THE CLOUD
12
13
AGENDA
Why GPU in the Cloud?
Key technologies for GPU in the Cloud
Performance Best Practices and Results
Deployment Quick Start
Review of Cloud Desktop Components
Resources
14
KEY TECHNOLOGY: VMWARE HORIZONPrivate Cloud infrastructure for delivering graphics rich desktops in the GPU enabled datacenter
15
WHY CHOOSE VMWARE FOR CLOUD-BASED WORKSTATIONS?
Market leading hypervisor
Proven, industry leading platform, fully integrated end-to-end solution
Supports huge range of use cases
Deliver a superior experience with state of the art protocols
Choice of GPU technologies
—vGPU, vDGA and others
1
5
161
6
AnyApplication
Traditional Applications
Modern, Cloud Applications
OneCloud
Cloud Management
Build-Your-OwnConverged
InfrastructureHyper-Converged
Infrastructure
AnyDevice
Business Mobility: Applications | Devices | Content
HYBRID CLOUD
PRIVATE
YourData Center
PUBLIC
vCloud Air
MANAGED
vCloud AirNetwork
Virtualized Compute, Network, Storage
vGPU
VMWARE NVIDIA CLOUD ARCHITECTURE
17
VMWARE HORIZON: CLOUD-BASED WORKSTATION SOLUTION FOR ALL USE CASES
MOBILE
DESIGNERS/ENGINEERS NON-EMPLOYEE DEVELOPERSREMOTE OFFICEMAIN OFFICE
NEWNEWNEWNEW
POINT OF SALE
NEW
Virtual desktop and app delivery
Horizon 7
Desktop and app delivery from public
cloud
Horizon Air
Secure containerized desktops
Simplifying App ManagementAPP Horizon App Management
1
7
Horizon FLEX
18
PRIVATE CLOUD VIRTUAL DESKTOP INFRASTRUCTURE
1
8
• Physical server host(s)
• Hypervisor Software
• Virtual machine management software
• Virtual desktop management system
• Virtual desktop connection broker
• Access and authentication systems
• Display protocol
• Application Delivery mechanisms
• User Persona/Profile management
• Clever “stuff” that simplifies it all for you….
19
HORIZON 7: FIRST-TO-MARKET INNOVATIONS TO MAKE THE CLOUD A REALITY
19
GPU Virtualization
Deployment Options
Blast Extreme Experience
Enterprise Scale & Security
App Lifecycle Management
Smart Policies
CONFIDENTIAL
20
INTRODUCING BLAST EXTREME - THE MOBILE CLOUD PROTOCOL
2
0
Use Less Bandwidth:Deliver better video to more users
Broad Client Support: H264 available in most devices
GRID Optimized:Deliver more 3D users
Network Friendly: TCP or UDP works in more places
Built for the Cloud:Adapts better to lossy networks
Better Battery Life:Optimized for mobile workers
21
Horizon 3D Virtual Desktops
VMWARE AND NVIDIA REFERENCE ARCHITECTURE
VMware Access PointHorizon Connection Servers
Horizon Virtual Desktops
Identity Manager
vCenter Operations
Manager for Horizon
NVIDIA
Lic. Svr
RADIUS
Server
View
Composer
vCenter
Shared Storage (Optional)
Active
DirectoryMS
SQL
SSD SSD
HTTPS/PCoIP/Blast Extreme
DMZ (HTTPS/PCoIP/Blast)
PCoIP or Blast Extreme
ESX, vCenter, Horizon, SQL, AD traffic
Shared Storage
3D DESKTOP CLUSTER DESKTOP CLUSTER
MANAGEMENT CLUSTER
iOS
AndroidMacOS Thin Client PC
KioskHorizon Clients
File Print
Server
App Volumes
CAD/CAx
“Server”
22
KEY TECHNOLOGY: GPU VIRTUALISATIONHow NVIDIA GRID is bringing GPU power to cloud based workstations
23
Server
Hypervisor
Virtual
Desktop
Virtual
Desktop
Virtual
Desktop
Virtual
Desktop
HOW DOES NVIDIA GRID WORK?Virtual
Desktop
Virtual
Desktop
CPUs
Hard
ware
Vir
tualizati
on L
ayer
24
Server
Hypervisor
Virtual
PC
Virtual
Workstation
Virtual
PC
Virtual
Workstation
HOW DOES NVIDIA GRID WORK?Virtual
PC
NVIDIA
GPU
H.264 Encode
Virtual
Workstation
NVIDIA Graphics
Driver
NVIDIA Quadro
Driver
NVIDIA GRID vGPU manager
NVIDIA Graphics
Driver
NVIDIA Graphics
Driver
NVIDIA Quadro
Driver
NVIDIA Quadro
Driver
vGPU vGPUvGPU vGPU vGPU vGPU
CPUsNVIDIA
GPU
Hard
ware
Vir
tualizati
on L
ayer
25
NVIDIA GRID PLATFORMCentralized virtual graphics capabilities for the enterprise
Support, Updates & Maintenance
NVIDIA Tesla (Datacenter GPU)
NVIDIA GRID Software
26
DESKTOPS & VIRTUAL APPS
Apps
NVIDIA GRID Virtual Applications
NVIDIA GRID Virtual PC
NVIDIA GRID Virtual Workstation
29
VM
A quick recap …
Tesla GPU
NVIDIA driver
AppsHypervisor
GRID vGPU Manager
GRID vGPU Architecture 101
30
Graphics
Compute
Copy
Engine
Video
Encode
Video
Decode
Tesla GPU
Framebuffer
Graphics And Compute
Copy Engine
Video Encode
Video Decode
Framebuffer (FB)
GPU engines
GRID vGPU Architecture 101
VM
NVIDIA driver
AppsHypervisor
GRID vGPU Manager
31
Tesla GPU
Framebuffer
Base Address Register (BAR)
A “window” into the GPU
Used to access control registers and framebuffer
All driver accesses to GPU resources go through a BAR
We assign the VM a subset of the GPU’s BAR – a virtual BAR
GPU Base Address Register
GPU access
GRID vGPU Architecture 101
VM
NVIDIA driver
AppsHypervisor
GRID vGPU Manager
Virtual BAR
Graphics
Compute
Copy
Engine
Video
Encode
Video
Decode
32
Tesla GPU
Framebuffer
Access to BAR controlled by CPU’s Memory Management Unit (MMU)
Hypervisor creates MMU page tables that allow VM to access its Virtual BAR
Blocks access to locations outside the Virtual BAR
GPU Base Address Register
Access protection
GRID vGPU Architecture 101
VM
NVIDIA driver
AppsHypervisor
GRID vGPU Manager
CPU’s MMU
Page Table
Translation
Virtual BAR
Graphics
Compute
Copy
Engine
Video
Encode
Video
Decode
33
Tesla GPU
Framebuffer
VMs post work to GPU via channels
Each channel is a queue of work from an app to the GPU engines
Each VM has dedicated channels, exposed inside VM’s Virtual BAR
GPU BAR
Channels
Work posting
GRID vGPU Architecture 101
VM
NVIDIA driver
AppsHypervisor
GRID vGPU Manager
Graphics
Compute
Copy
Engine
Video
Encode
Video
Decode
34
Tesla GPU
Framebuffer
GPU’s hardware scheduler places work from active channels onto GPU engines
Engines sequentially execute work, signal work completions to NVIDIA driver
Scheduler immediately loads engine with next work item
GPU BAR
Channels
Scheduling
vGPU Architecture 101
Scheduler
VM
NVIDIA driver
AppsHypervisor
GRID vGPU Manager
Graphics
Compute
Copy
Engine
Video
Encode
Video
Decode
35
Tesla GPU
Framebuffer
All GPU engines access memory
Each vGPU is assigned fixed range of FB for its exclusive use
GPU’s MMU manages all engine accesses to FB and system memory
Prevents engines accessing outside of vGPU’s FB, or the guest VM’s system memory
Memory protection
GRID vGPU Architecture 101
Memory Management Unit (MMU)
VM
NVIDIA driver
Hypervisor
GRID vGPU Manager
Sysmem resource
Guest VM’s FBGraphics
Compute
Copy
Engine
Video
Encode
Video
Decode
36
• vGPU dedicated resources: virtual PCIe BAR, framebuffer, channels
• Guarantees uncontested access when posting work, accessing FB
• vGPU shared resources: Graphics/Compute, Video Encode/Decode, Copy Engines
• GPU engines are timeshared, execute in parallel
• GPU hardware scheduler places work on engines
• GPU MMU protects engine accesses to framebuffer and system memory
Summary
GRID vGPU Architecture 101
37
QUADRO: THE CHOICE OF PROFESSIONALS
Enables designers to bring their visions to life
Quadro: 80%+ workstation share
Suite of algorithms incorporated in major design tools
Source: IDC
38
QUADRO IN THE DATACENTERNVIDIA GRID delivers Quadro features anywhere, on any device
Realistic Models Application Certification Improved ProductivityPerformance Enhancements
39
AGENDA
Why GPU in the Cloud?
Key technologies for GPU in the Cloud
Performance & Sizing Best Practices
Deployment Quick Start
Summary
Resources & Next Steps
40
SIZING YOUR VMWARE INFRASTRUCTURE FOR HIGH PERFORMANCE GRAPHICS
41
HIGH PERFORMANCE VIRTUAL DESKTOP IS NOT YOUR AVERAGE VDI
4
1
Density
CPU
Memory
IOPS
Rich Graphics
Trade-offs
Trade-offs
42
SIZING ESX FOR RICH GRAPHICS USERS: COMPUTEVMware HCL & the NVIDIA GRID Certified Server
—2 Display Adapters (onboard + GFX card)
—Virtual Technology for Directed I/O (VT-d)
—PCIe x16
Host CPU
—Heavy 3D applications are often mono-threaded
—CPU frequency is more important than the number of cores
—Workloads will still require multiple vCPUs
OS, Protocol, Improved memory performance
Multi threaded rendering applications
4
2
• 3.1 Ghz• 10 core• 160W
ESXi Host
43
SIZING ESX FOR RICH GRAPHICS USERS: MEMORY
Host Memory
—Audit what your workstations use today
—Highest frequency memory matters
—3D workloads likely require (per user):
8GB (medium workload)
16GB (most common)
64GB (heaviest workload)
4
3
• 16 x 16GB• 2133Mhz
ESXi Host
44
SIZING ESX FOR RICH GRAPHICS USER: STORAGE
Storage will be the CRITICAL factor in performance
CAD users >1TB of storage per desktop
SIZE and PERFORMANCE matter now
Virtual SAN – SSD based local storage
—Understand local storage capacity
All Flash based SAN / VSAN
4
4
1,536
45
SIZING ESX FOR RICH GRAPHICS USERS: STORAGE
4
5
Test Conditions Distributed
Solver PCG Solver Modal Analysis
Number of Nodes 1340453
Number of Contact Elements 21763
Number of Solid Elements 833530
Number of Total Elements 862707
10843
9181
1483
1444
0 2000 4000 6000 8000 10000 12000
Elapsed Time (in sec) Lower is better
Modal Benchmark –Physical vs. Virtual Environment
Optimized (Non-Virtual) Optimized (Virtual)
Non-Optimized (Physical) Non-Optimized (Virtual)
0.02%
Difference
15%
Difference
46
SIZING ESX FOR RICH GRAPHICS USERS: GPU
GPU dictates interactive user experience –rotation/scaling etc
GPU dictates performance when handling complex models (no. of polygons)
GPU critical for high-performance graphics
NVIDIA GRID – most options for vSGA, vDGA and vGPU
Maximum 8 GPU cards per ESX host (2-4 typical)
4
6
GRID M6 GRID M60
Number of GPUs 1 x entry Maxwell GPUs 2 x Maxwell GPUs
Total NVIDIA CUDA cores 1536 4096
Total memory size 8 GB DDR3 16 GB GDDR5
Max users or vGPUs 16 32
Card Equivalent Quadro Quadro
47
SIZING ESX FOR RICH GRAPHICS USERS: NVIDIA GRID VGPU
M60
Framebuffer GRID
Virtual
Applications
GRID
Virtual PC
GRID
Virtual WorkstationVMs per
GPU
VMs per
M60
512 MBM60-0B M60-0Q 16 32
1 GBM60-1A M60-1B M60-1Q 8 16
2 GBM60-2A M60-2Q 4 8
4 GBM60-4A M60-4Q 2 4
8 GBM60-8A M60-8Q 1 2
Pass-through 8GB 1 2
48
HOMOGENOUS VIRTUAL GPUS
Tesla M60
Physical GPU 0 Physical GPU 1
M60-2Q M60-2Q M60-2Q M60-2Q M60-4Q M60-4Q
M60
-1B
M60
-1B
M60
-1Q
M60
-1Q
M60
-1B
M60
-1B
M60
-1B
M60
-1BM60-2Q M60-2Q M60-2Q M60-2Q
M60-2Q M60-2Q M60-4QM60
-1Q
M60
-1Q
M60
-1Q
M60
-1Q
M60
-1Q
M60
-1Q
M60
-1Q
M60
-1Q
Valid configuration with K240Q vGPUs on GPU 0, K260Q vGPUs on GPU 1:
Valid configuration with K200 vGPUs on GPU 0, K240Q vGPUs on GPU 1:
Invalid configuration with mixed vGPU types on GPU 0:
49
SIZING YOUR VIRTUAL DESKTOP FOR HIGH PERFORMANCE GRAPHICS
4
9
50
SIZING FOR RICH GRAPHICS WORKLOADS –VIRTUAL MACHINE
5
0
Virtual Machine
• Hardware v11
• VMXNET3 NIC
• vGPU
– Configure
Profile
• vDGA
– Enable Pass-
through
vCPU
• 2 vCPUs:
– 1 vCPU - OS
– 1 vCPU - PCoIP
• Recommend at
least 4 vCPUs for
Power Users
vRAM
• 4GB
– Power User/
Entry Level Eng
• 8GB
– Mid Eng./Video
• 16GB
– Advanced Eng.
• 32GB
– CAD/CAM
• 64GB
– Digital Mock Up
OS
• vDGA
– 64bit Windows
• vGPU
– 32/64 bit
Windows
• Install
– VM Tools
– View Agent
– NVIDIA driver
(vGPU)
VMware OS
Optimization Tool
• MASSIVE 40%
saving!
• Download from
labs.vmware.com
51
VCPU IMPACT ON MEMORY PERFORMANCE
5
1
Read (in MB/s) Write (in MB/s) Copy (in MB/s) Latency (in ns)
VM - 2 Cores/16GB vs
VM - 4 Cores/16GB44% 49% 86% 0%
44%49%
86%
0%0%
10%
20%
30%
40%
50%
60%
70%
80%
90%
100%4 C
ore
s -
% Incre
ase
Memory Performance Increase from 2 vCPU to 4 vCPU
52
Entry Level
Engineer/Reviewer
•M60-1Q
•4 GB RAM
•2 VCPUs (2.6 GHz)
•1 GB video RAM
•2 2560x1600
displays
Mid Level
Engineer/Editor
•M60-2Q
•8 GB RAM
•2-4 VCPUs (3.1 GHz)
•2 GB video RAM
•4 2560x1600
displays
Advanced Engineer
•M60-4Q
•16 GB RAM
•4-8 VCPUs (3.1 GHz)
•4 GB video RAM
•4 2560x1600
displays
• 4 x 4K displays
SIZING FOR RICH GRAPHICS USERS: NVIDIA GRID VGPU RECOMMENDATIONS
5
2
53
SIZING FOR RICH GRAPHICS USERS: CLIENT DEVICE3D and multimedia application can drive high FPS at high resolution
GPU throughput uncompressed over 2 Gbps
Protocol optimization, compression and bandwidth savings
—Only send pixels that change etc
Plan for 2-2.5 Mbps for CAD users (average)
Burst to 5-7 Mbps, can be higher in LAN scenarios (20Mbps-200Mbps depending on hardware)
Client device must be capable of decoding PCoIPthroughput or H264 offload
—Tera2 client or suitable spec Client
—Check your device specifications
5
3
55
VMWARE BLAST EXTREME ACCELERATION WITH NVIDIA GRID
56
CHALLENGES FOR VIRTUAL GRAPHICSProfessional graphics workloads require great user experience
UX
Require “snappy”
experience
Rely on heavy
encoding and
decoding
User density is
limited by CPU
bottleneck
57
Up to
51mslatency
reduction
NVIDIA BLAST EXTREME ACCELERATIONIncreased scalability and performance for graphics workloads
Up to
18%increase
in density
Up to
89%bandwidth reduction
Source: NVIDIA GRID Performance Engineering Lab
58
HOW IT WORKS TODAY: PCoIP
CLIENT
Render
Kybd/Mse
SERVER with GRID GPU
Capture
Encode
IPNetwork
CPU NIC
GRID GPU WORKLOAD
NON GPU WORKLOAD
Decode
Render
59
NVIDIA BLAST EXTREME ACCELERATION
CLIENT
Render
Kybd/Mse
SERVER with GRID GPU
Capture
Encode
IPNetwork
CPU NIC
GRID GPU WORKLOAD
NON GPU WORKLOAD
Decode
Render
60
NVIDIA BLAST EXTREME ACCELERATION
• Reduces overall latency
• Offloads CPU workload to
GPU
• Increases scalability
• Improves user experience
• Lowers N/W bandwidth
demand
GRID GPU
3D
HW Encoder
Framebuffer
AppsApps
Apps
Graphics
commands
Context/Display Capture
Render Target
Front Buffer
H.264 /
H.265
streams
Remote Client
61
• Scalability
• 18% higher user density for ESRI ArcGIS Pro 1.1
• User Experience
• 6-13% higher average FPS
• 27-51ms less latency
• 48-89% bandwidth reduction with TCP and 19% with UDP
You don’t have to chose, you get all at the same time
NVIDIA BLAST EXTREME ACCELERATION IMPROVES USER EXPERIENCE AND REDUCES TCO!
Source: NVIDIA GRID Performance Engineering Lab
62
PERFORMANCE RESULTS
63
16 16
8
4
32 32
16
8
0
5
10
15
20
25
30
35
Standard 2vCPU16GB RAM
1GB Frame Buffer(K240Q / M60-1Q)
Moderate 4vCPU32GB RAM
1GB Frame Buffer(K240Q / M60-1Q)
Power User 4vCPU64GB RAM
1GB Frame Buffer(K240Q / M60-1Q)
Analyst 4vCPU128GB RAM
1GB Frame Buffer(K240Q / M60-1Q)
K2 M60Heavy Users
Many Applications
Number of
Concurrent
users
on a single
server
Light Users
Fewer Applications
M60 OFFERS PERFORMANCE IMPROVEMENTSNVIDIA GRID DASSAULT CATIA V5/V6 SCALABILITY GUIDE
6
3
Task Worker Knowledge Worker Power User Designer
Aerospace Manufacturing
Production CATIA workloads
Satellite offices, over WAN links
GRID K2 GPUs and M60 GPUs
Four different user profiles
Frame Rate Limiting (FRL) turned on
Results
Doubled the number of users per host
Observed a 70% productivity improvement
File proximity/transfer time savings
No reported degradation to end user
Even LAN users reported 30% productivity gain
** 2.6 GHz+ Intel Xeon
E5 v3 or greater
64
VIRTUAL BATTLESPACE RESULTS
4/4/2016
65
66
AGENDA
Why GPU in the Cloud?
Key technologies for GPU in the Cloud
Performance Best Practices and Results
Deployment Quick Start
Summary
Resources & Next Steps
67
GETTING STARTED
Deployment Requirements
Product Activation Keys
(PAK)
Download GRID SW (host driver,
guest driver)
License server creation,
installation and management
The GPU Mode Switch Utility
Install host driver
Attach a vGPU profile to a VM
Install the NVIDIA driver
into a VM
Configure a VM for a NVIDIA GRID vGPU
license
Key concepts to get you going
68
DEPLOYMENT REQUIREMENTS
NVIDIA GRID enterprise software is designed to
run on select NVIDIA Tesla GPUs.
Ensure you have the following:
NVIDIA GRID Certified Servers:
http://www.nvidia.com/object/grid-certified-
servers.html
Required GPU hardware: Tesla M6 or M60
A supported virtualization software stack
Read the release notes!
Correct build numbers for the virtualization
software that have been tested for the
specific NVIDIA GRID Software release
Other important version-specific information
Note: You will need to register first to get access to the release notes.
69
QUICK START GUIDE
Deployment Requirements
Product Activation Keys
(PAK)
Download GRID SW (host driver,
guest driver)
License server creation,
installation and management
The GPU Mode Switch Utility
Install host driver
Attach a vGPU profile to a VM
Install the NVIDIA driver
into a VM
Configure a VM for a NVIDIA GRID vGPU
license
Overview of concepts to get you going
70
PRODUCT ACTIVATION KEYS (PAK)Order email template
Email content for GRID SW Order:
• Product Activation Key (PAK) is dynamically added to order fulfillment email template
• Blue underlined content are links to those destinations for registering, logging in or support
• Enterprise support portal and email address is provided for help
• Email generated from [email protected] (advise on checking spam folders as needed)
Sample email
71
NVIDIA LICENSING PORTALRegistration Page
• Customers will click on the Register or Log in link within the order fulfillment email to navigate to the NVIDIA Licensing Portal
• First time customers will be required to register their information
• A link to the Terms and Conditions is provided (End User License Agreement)
• Customers must accept the terms and conditions to download licenses and software
• Returning customers will not have to re-register but can login to their account
72
QUICK START GUIDE
Deployment Requirements
Product Activation Keys
(PAK)
Download GRID SW (host driver,
guest driver)
License server creation,
installation and management
The GPU Mode Switch Utility
Install host driver
Attach a vGPU profile to a VM
Install the NVIDIA driver
into a VM
Configure a VM for a NVIDIA GRID vGPU
license
Key concepts to get you going
73
DOWNLOAD GRID SOFTWAREThe Product Download Page contains…
NVIDIA GRID Software: Select based on the
hypervisor brand and version you will run. NVIDIA
GRID vGPU supports versions of VMware vSphere
ESXi and Citrix XenServer
GPU mode change utility: The Tesla M60/M6
boards are shipped in Compute Mode. NVIDIA GRID
vGPU requires Graphics Mode. The GPU mode
change utility is used to switch modes.
NVIDIA License Manager Software: Enables Guest
OS to utilize the NVIDIA GRID vGPU profiles by
checking out a license at boot time
All downloads are in .zip format and contain
release notes and user guides along with the
software.
74
QUICK START GUIDE
Deployment Requirements
Product Activation Keys
(PAK)
Download GRID SW (host driver,
guest driver)
License server creation,
installation and management
The GPU Mode Switch Utility
Install host driver
Attach a vGPU profile to a VM
Install the NVIDIA driver
into a VM
Configure a VM for a NVIDIA GRID vGPU
license
Key concepts to get you going
75
THE NVIDIA GRID LICENSE SERVER
NVIDIA GRID vGPU profiles are licensed features on NVIDIA Tesla GPUs
• Licenses generated and retrieved from NVIDIA’s Licensing Portal
• Installed onsite, enables NVIDIA GRID licensed products
• Manages a pool of floating licenses• License acquired during boot time of the
Guest OS • The license returned when the Guest OS
shuts down
Don’t impact workUse for license management
Enable more use cases
UX
77
QUICK START GUIDE
Deployment Requirements
Product Activation Keys
(PAK)
Download GRID SW (host driver,
guest driver)
License server creation,
installation and management
The GPU Mode Switch Utility
Install host driver
Attach a vGPU profile to a VM
Install the NVIDIA driver
into a VM
Configure a VM for a NVIDIA GRID vGPU
license
Key concepts to get you going
78
MODE SWITCHING
Compute Mode
Shipped in this mode
Graphics Mode
Required for NVIDIA GRID
Tesla M60 Tesla M6
gpumodeswitchBootable Linux or Windows
Temporarily boot Linux on the server to change mode; the
gpumodeswitch release package includes bootable Linux images
for this purpose
79
GPUMODESWITCH EXAMPLEList the current mode of all GPUs use --listgpumodes
Switch a GPU’s mode use--gpumode graphics
--gpumode compute
After a GPU mode switch, the server is rebooted to ensure that the modified resources are correctly accounted for by the OS or hypervisor
80
QUICK START GUIDE
Deployment Requirements
Product Activation Keys
(PAK)
Download GRID SW (host driver,
guest driver)
License server creation,
installation and management
The GPU Mode Switch Utility
Install host driver
Attach a vGPU profile to a VM
Install the NVIDIA driver
into a VM
Configure a VM for a NVIDIA GRID vGPU
license
Key concepts to get you going
81
NVIDIA GRID WITH CONFIGURATION PROCESS
For more details consult the GRID Deployment Guide and additional online training
• Install NVIDIA GRID software into ESXiHypervisor
• Build a Gold Master VM Image with NVIDIA GRID vGPU Profile
• Install NVIDIA Windows Driver into Gold Master Image
• Configure vGPUenabled VM pools in Horizon Environment
82NVIDIA CONFIDENTIAL. DO NOT DISTRIBUTE.
INSTALLING AND TESTING VIB
$ esxcli system maintenanceMode set –e true
$ esxcli software vib install -v /store/NVIDIA.vib
$ reboot
Installation Result:
Message: Operation finished successfully. Reboot Required: false
VIBs Installed: NVIDIA_bootbank_NVIDIA-vgx-
VMware_vSphere_6_Host_Driver_346.27-1OEM.600.0.0.2159203
VIBs Removed:
VIBs Skipped:
$ esxcli system maintenanceMode set –e false
$ esxcli software vib list | grep -i nvidia
$ nvidia-smi
83
WINDOWS VM: VGPU ASSIGNMENT
In vCenter, edit VM settings, add “Shared PCI Device”, then select “Add”
Select GPU profile, then select “OK”
Start VM
84
NVIDIA DRIVER INSTALL
85
HORIZON POOL CREATION
86
QUICK START GUIDE
Deployment Requirements
Product Activation Keys
(PAK)
Download GRID SW (host driver,
guest driver)
License server creation,
installation and management
The GPU Mode Switch Utility
Install host driver
Attach a vGPU profile to a VM
Install the NVIDIA driver
into a VM
Configure a VM for a NVIDIA GRID vGPU
license
Key concepts to get you going
87
CONFIGURE NVIDIA GRID VGPU LICENSE
Configure the Manage License Feature in the NVIDIA Control Panel
• Enter the license server address and port
• These settings can be preloaded through register keys.
• The license settings will persist across reboots.
• A license is obtained prior to the user logging into the VM.
Enables a VM access to an NVIDIA GRID vGPU
April 4-7, 2016 | Silicon Valley
L6117 - NVIDIA GRID DEPLOYMENT HANDS ON LAB
Room 210C 4/7/2016 13:00 - 16:00
89
AGENDA
Why GPU in the Cloud?
Key technologies for GPU in the Cloud
Performance Best Practices and Results
Deployment Quick Start
Summary
Resources & Next Steps
90
VMWARE HORIZON: DESKTOP TRANSFORMATION SOLUTION FOR EVERY USER
MOBILE
DESIGNERS/ENGINEERS NON-EMPLOYEE DEVELOPERSREMOTE OFFICEMAIN OFFICE
NEWNEWNEWNEW
POINT OF SALE
NEW
Virtual desktop and app delivery
Horizon 7
Desktop and app delivery from public
cloud
Horizon Air
Secure containerized desktops
Simplifying App ManagementAPP Horizon App Management
9
0
Horizon FLEX
91
NVIDIA GRID PLATFORMCentralized virtual graphics capabilities for the enterprise
Support, Updates & Maintenance
NVIDIA Tesla (Datacenter GPU)
NVIDIA GRID Software
929
2
Streamlined end
user access and supplier
connectivity
Real-time collaboration for
engineers in
29 offices
across 7countries
When I use
SolidWorks on
virtual desktops,
it’s seamless…
from my phone, my
iPad, and my
computer. 33% productivitygain, access to
radiology images on a variety of endpoints
We are confident
that it is now
finally
economically viable
for CH2MHILL to
provide a VDI farm
for graphically
demanding
workloads across
the globe. This
could truly
transform the way
we do business.– Mechanical Engineering
StudentNorth Carolina State
University
– Craig FletcherCTO
CH2MHILL
“
”
“
”US Based Healthcare
Provider
93
AGENDA
Why GPU in the Cloud?
Key technologies for GPU in the Cloud
Performance Best Practices and Results
Deployment Quick Start
Summary
Resources & Next Steps
94
TEST DRIVE NVIDIA GRID TODAYhttp://www.nvidia.com/trygrid
Instant access to NVIDIA GRID on VMware vSphere and Horizon
Pre-installed apps such as AutoCAD, SolidWorks, ArcGIS, NX, etc.
HTML demos and HD videos
95
KEYS TO A SUCCESSFUL POC
Important things to remember during your POC:
Define “acceptable” user experience
Render times, smoothly panning, menu to appear
Compare real world workloads
With “think time” included, opposed to peak workload benchmarks
Test with real users
3D designer in the application, ultimate test of success
96
DETAILED INSTRUCTIONSDeployment guides
www.nvidia.com/gridresources
97
OTHER NVIDIA GRID SESSIONS TO ATTEND
S6504 - A Data-Driven Methodology for NVIDIA GRID™ vGPU™ Sizing
S6392 - AEC Project Execution Using GRID vGPU Enhanced Virtualization
S6859 - Unveiling the Impact of Time Slicing with NVIDIA GRID™ vGPU for Realistic ROI/TCO
Analysis
S6210 - NVIDIA GRID™ and Dassault Catia from Proof of Concept to Production
S6194 - Delivering Graphics-Intensive Applications to Computing Labs and BYOD in Education
S6833 - Delivering GPU-Accelerated Applications from your Private Cloud
S6209 - A Look at Real World Performance Capabilities of NVIDIA GRID™ 2.0
S6198 - The Latest in High Performance Desktops with VMware Horizon and NVIDIA GRID™
vGPU
S6200 - Delivering 3D Workstations with VMware Horizon and NVIDIA GRID
L6117 - NVIDIA GRID Hands-on Lab
S6332 - Training and Simulation in a Secure Cloud Environment
April 4-7, 2016 | Silicon Valley
THANKS – Q&A
99
PHOTO CAPTION