accelerated cloud graphicson-demand.gputechconf.com/gtc/2013/presentations/s3543-accelerating... ·...
TRANSCRIPT
![Page 1: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/1.jpg)
Accelerated Cloud Graphics
Franck DIARD, Ph.D.
Chief SW Architect, NVIDIA
![Page 2: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/2.jpg)
[GRID] … truly appears to be prepared to change
the way we play games from top to bottom.
… the GRID could be the long-awaited answer to
cloud gaming. “
”
“ ”
Convenience wins. Putting movies in the cloud took sales
revenue from $50B (packaged DVDs) to more than $250B
(on demand). Cloud gaming is next.
![Page 3: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/3.jpg)
Scope
System Architectures
Kinds of GPU “virtualization”
HW: potential and limits
SW: SDK, techniques
Streaming
New content / Apps
![Page 4: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/4.jpg)
GRID Server Market
All cloud gaming platforms from our partners
G-cluster, Playcast, Ubitus, Agawi, CyberCloud, Cloud Union
Desktop/App remoting for VCA (Visual Computing Appliance)
New cloud game engines
Video GRID servers
vSTB: Live TV, PVR, VOD
Transcoding farms
![Page 5: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/5.jpg)
GRID Cloud Gaming Ecosystem
Partner middleware
They deliver full solution with
content, writing their middleware
They buy simple “blank” gaming
servers
They use our Monterey SDK to
write their SW stack
They fit our HW with their goals,
QOS and content with our SDK
Partners Middleware
NVIDIA HW Servers
Monterey Server
Side SDK
![Page 6: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/6.jpg)
Comparison
![Page 7: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/7.jpg)
Cloud Gaming
CLIENT
Decode Render
Kybd/Mse
SERVER
Render
Capture
Encode
GRID
30 ms
2 Frames
Network
30 ms
2 Frames
GRID Client
<16ms
IP Network
CPU NIC
![Page 8: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/8.jpg)
GRID Gaming
What is Cloud gaming ?
AAA games? Casual games?
What is the experience?
What is the QOS?
That is a lot of things, the spectrum is VERY large
What is the right architecture to do all of it?
Well, it depends…
![Page 9: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/9.jpg)
Some Background on Platforms
![Page 10: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/10.jpg)
SW Architecture
Windows 7+
GPU
Monterey SDK
Windows 7+
Low Latency Frame Buffer Capture
NvFBC
Low Latency Render Target Capture
NvIFR
NVENC Low latency
Encoder
NVMOS Platform Virtualization
Dedicated GPU
Ad-hoc API Shimming DirectX
GPU
Windows 7+
Game Game
GPU GPU
Game Game
Game Game
![Page 11: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/11.jpg)
Architectures: Bare Metal OS No virtualization per se
Dx/OGL API shimming can be used
Windows
GPU GPU GPU GPU
![Page 12: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/12.jpg)
Multiple DX/OpenGL Contexts
Windows
GPU GPU GPU GPU
![Page 13: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/13.jpg)
Bare Metal OS/Shimming
Advantages
One OS, multiple game
instances
Load balancing is dynamic
depending on content
Licencing cost (one OS
licence)
High CCU capability for casual
gaming
Disadvantages
Require game modification
for input, frame output
encoding
Today limited by CPU
encoding on previous
middleware partners solution
![Page 14: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/14.jpg)
Windows Linux Windows
Architectures: XEN NMOS
The platform is virtualized, not the GPUs
Windows
GPU GPU GPU GPU
XEN
VM VM VM VM
![Page 15: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/15.jpg)
XEN / Hypervisor for Gaming
Advantages
Complete separate OSes, per direct attach GPU
Security
Deployment
Basically, a real desktop
complete high performance of DirectX, OpenGL
Each game can go full screen
Problems
CPU tax, performance
Needs more resources
Not well suited for high CCUs
License costs
![Page 16: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/16.jpg)
GRID HW
![Page 17: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/17.jpg)
Kepler, first Cloud GPU
High performance per watt
Integrated hardware encoder
Low-latency frame buffer reads
28nm
![Page 18: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/18.jpg)
GRID Boards
K340
4 GK107 (GT650)
Only 8x Gen3 between PLX and GPU
1GB per GPU
Low 3D perf, but more encoders
K520 (0.8 GTX680)
2 GK104
4GB per GPU
Lot more 3D perf, but less encoding
For Dx gaming, GeForce driver is king
GRID GPUs need best perf
Profiles, optimizations
![Page 19: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/19.jpg)
Systems
![Page 20: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/20.jpg)
HW: Systems
Requirements
Pack many GPUs
2-4 slots
VT-D Intel for direct attach/pass-though
HW requirements
Cooling
Rack space
Power
SuperMicro
![Page 21: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/21.jpg)
GRID Gaming Systems:
SuperMicro 2027 2U
3 or 4 GPU boards
2 Xeons E5 Sandy Bridge
Full lanes of PCI 3.0
Supports VT-d well
6-8 GK104
12-16 GK107
Some other form factors
VCA (TYAN, 4U)
Balance of power, rack space, cooling
![Page 22: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/22.jpg)
1x Chipset 2P, 16 cores
![Page 23: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/23.jpg)
Monterey SDK: Server Side
Capture and Stream
![Page 24: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/24.jpg)
Monterey SDK
Core components
NvFBC: whole display grab: 1 OS = 1 user = 1 stream
NvIFR: render buffer grab: 1 OS = n users = n streams
Doc
Header files, 2 dlls
To system memory
CSC conversion
Partners running CPU encode
Bus contention
To H264 HW encoder
CSC
![Page 25: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/25.jpg)
NvFBC / NvIFR
Frame Buffer Capture (NvFBC) Simple onboarding, low CCU
A display remote display
1 OS = 1 user = 1 GPU = 1 virtual display = 1 stream
Orthogonal to the rest of everything that puts pixels on the screen
Supports multiple heads
Inband Frame Readback (NvIFR) Complex onboarding, high CCU for small games
Injected into apps API calls or modified apps
1 OS = n users = n apps = n streams
Can use several GPUs mapped into OS with Dx/OpenGL affinity
![Page 26: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/26.jpg)
NvFBC
Overlay, mouse, GDI, Dx, OGL, video
0 CPU cycles
Compressor process comes async to grab content
Negative latency:
NvFBC grabs the display back buffer
If Window Manager runs VSYNC’d, NvFBC grabs bits before they
make it to the local screen (if any)
![Page 27: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/27.jpg)
Windows
DX
NvFBC Usage case
VCA, VDI solutions, high-end gaming
One VM on NMOS has a full OS and will run only one game/desktop per VM
So it can go full screen and take ownership of the single display attached to VM
Game’s behavior is not modified, at all
Only one game can run
A second background process uses NvFBC to capture what is on the screen, gets back the H264 bits and streams it out
Complete asynchronous, pure GPU/DMA
Game process
DX
Streamer
process
NvFBC
NVEnc
3D CSC
Shader
3D
![Page 28: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/28.jpg)
NvIFR
SDK to use with API shimming
Render target read back: Dx9, Dx10, Dx11 (OGL in testing)
Format conversion, scaling
In band with GFX API: Present() call
Page locked sysmem
H.264 interface
Asynchronous Event Signaling, CPU friendly, interrupt driven Not blocking main render loop
Texture sharing extension for DX9
Linux Support
![Page 29: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/29.jpg)
Windows
GPU
NvIFR Usage case
Game is injected/hooked
DX device creation is forced
offscreen, window hidden
n smaller games can run at
once, generating n streams
Game
process
DX
NVEnc
Shim
NvIFR
Game
process
DX
Shim
NvIFR
Game
process
DX
Shim
NvIFR
GPU CSC
Shader
![Page 30: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/30.jpg)
NvENC: Fast pathed from NvIFR/NvFBC
Completely separate GPU unit: <2 watts
PSNR comparable to x264
up to 16 encoding contexts, per GPU
High Profile
Single pass or dual pass 720p: 4 ms
1080p: 8 ms
4 720p streams per GPU at 30 Hz
Constrained VBV buffer size network packet framing for real time delivery
CBR, VBR, Min QP
CUDA Interoperability
I-frame on-demand
Reference picture invalidation logic API for packet loss
4Kx4K support
Stereo MVC Encoding
![Page 31: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/31.jpg)
Baremetal Shimming Techniques
![Page 32: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/32.jpg)
Windows DirectX Gaming Limitations…
If Windows is great…
DirectX and OpenGL games run great on our GPUs
It can’t do more than one game fullscreen at once
So to run several games on the same baremetal OS
So games they must be injected/hooked to prevent them
from taking ownership of the display and preventing other
games from launching/rendering
![Page 33: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/33.jpg)
Shim layer DX affinity
N GPUs attached to Windows desktop
N D3D adapters
Apps create D3D device on default adapter (DX ordinal 0)
Apps have to be injected and hooked so
The CreateDevice() ordinal parameter is overriden to DX ordinal n
DXVA follows the Dx device
NvIFR uses the same device so the HW encoder will be on the same GPU
3 2 1 0
![Page 34: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/34.jpg)
Monterey SDK: Client Side
Decode and Display
![Page 35: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/35.jpg)
Decoders
Low latency
Power efficient
H264 packets, fastest path to display
Windows
Android, Surface RT
Mac
LGTV
![Page 36: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/36.jpg)
HW decoding with nvCUVID/DX
Popping off h264 frames from network socket
Feeding elementary stream to nvCUVID -> DXVA
YUV comes out in a CUDA buffer
Run CUDA program to convert YUV to ARGB CUDA buffer
Map cuda ARGB buffer into ARGB texture
Draw quad with ARGB texture
![Page 37: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/37.jpg)
Android Decode
Low latency decode/display
Bypass of OS traditional stack, this is not streaming
Java frame work calls jni with native window handle
Set output of decoder directly into native window
SDK:
HoneyComb/ICS and up, native JNI
Decoder: 8ms 720p
Tear free display
720p / 60 FPS
1080p / 30 FPS
![Page 38: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/38.jpg)
New Type of Content
![Page 39: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/39.jpg)
Multi-View game engine
![Page 40: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/40.jpg)
Cloud Game Engine
Cloudgine, EA
In house prototype
New type of games
One single app running,
one OS
Using all GPUs
Rendering different point
of views from n different
users
Windows
GPU GPU GPU GPU
Game
process
N Networked Clients
Dx
Context
Dx
Context
Dx
Context
Dx
Context
Dx
Context
N Dx Contexts
Using NvIFR
![Page 41: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/41.jpg)
Higher CCUs
Having one complete VM per game is too heavy to reach 80
CCUs per servers
Extreme CCUs can run baremetal OS with DX shimming to
run loads of casual small games
Looking at some very specific and very limited applications
![Page 42: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/42.jpg)
Video workloads
XBMC Multimedia Center
Mostly HW DXVA decoding
Compositing, menus, PIP
Reencoding with NvIFR
Injected with Detours
Dx9Ex Present path
hooked
NvIFR running in side
encoding thread
![Page 43: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/43.jpg)
Virtual Set Top Box (vSTB)
Windows
GPU GPU GPU GPU
64 transcoded
composited
streams
Storage Content Server 64 streams
MultiCast
![Page 44: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/44.jpg)
48 XBMC sessions
![Page 45: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/45.jpg)
Demos
GRID server (3 x K340)
Streaming n sessions of XBMC
7 Mbps per stream
HW decoding, encoding
![Page 46: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/46.jpg)
GRID Support
For partners
SDK with collection of samples, tools, tips and techniques
Very responsive and knowledgeable team
Solution engineers
Release planning
![Page 47: Accelerated Cloud Graphicson-demand.gputechconf.com/gtc/2013/presentations/S3543-Accelerating... · NvFBC / NvIFR Frame Buffer Capture (NvFBC) Simple onboarding, low CCU A display](https://reader030.vdocuments.mx/reader030/viewer/2022040902/5e72e26f1cbb782710174a1d/html5/thumbnails/47.jpg)
Thanks
Phil Eisler, Jon Barad, Thomas Ruge