Future Cinema - chmpr

Download Report

Transcript Future Cinema - chmpr

UCSD Center for Hybrid
Multicore Productivity Research
[email protected]
crca.ucsd.edu/sheldon
Next Generation Culture will develop out of
the capabilities that Multicore Computing
enables in relationships that we can create
between these areas:
Assets, Dynamics and Behavior
Computation for Virtual worlds and games
Multi-user, Extensible Virtual Worlds
Future Cinema – 3D, Ultra-high Resolution
Production
Initial proposed industry partners:
IBM, Intel, Sony, Emergent Game Technologies, Zaxel Inc.
Future Cinema –
•Ultra-high Resolution
•3D
•Networked Production Delivery
•Prototyped and authored in virtual worlds
•Real (vs. Reel) time cinema
•Multi-point Telecommunications -with all
of the above requirements.
Ultra high resolution
projection
technologies
4K – 3840 x 2160
pixel displays today
8K and beyond
tomorrow
Provide qualitative
advantages over film
projection, driving
adoption of digital
cinema in coming
years.
Digital Cinema Auditorium
ultra-high resolution cinema, 3-D sound;
tele-presence conferencing; stereo imaging,
Cine-Grid optical ultra-high bandwidth
High Resolution gives 3D the
needed fidelity.
Drives the development of
production techniques that
require greater computing
capacities at all levels.
Mutable Media Environments
Relationships between Virtual Worlds and Digital Cinema
Co-hosting virtual world at UCSD and Peking
University, distributing application
development and delivery between US and
China – using high speed global optical
network.
Z/Series Mainframe Computers – providing
large centralized processing for multi-user
virtual worlds, large memory footprint for
digital cinema development, extensive storage
and data management techniques for
distributed development
Cell Processor Compute Servers provide
accelerated parallel, vector computation
IBM Mainframe and Bladeservers at at both
UCSD and PKU.
San Diego Supercomputer Center, Calit2 and
CRCA collaborate on resource management.
Compute/Power Advantages.
Global lambda grid is the “new Hollywood” as cinema
becomes an end to end digital medium.
This is the platform to deploy multicore resources.
Calit2 and NSF Optiputer Project Allow for High Speed
Networks and Ultra-High Fidelity environments.
Calit2, UCSD
WAAG, Amsterdam
Audio Engineering Society Demo
First ever real-time IP
transmission of 4K
resolution digital motion
picture and 24 channel
digital audio
synchronized and mixed
from multiple remote
locations
Special event was part of
CineGrid demonstrations of
networked digital cinema
Demo took place during
the Audio Engineering
Society (AES) convention
at the Letterman Digital
Arts Center, Premier
Theater in the Presidio of
San Francisco,
October 5-8th, 2006
Streaming 4K video
and 24 channel audio
from multiple locations
Compensate for
fixed latency with
Max/MSP patch to
maintain
synchronization
Demonstrate remote
collaboration and
tele-matic production
• LucasFilm
– San Francisco
• UCSD
– San Diego
• USC
– Los Angeles
• Keio University
– Tokyo
AES Network Usage
• Vast majority of bandwidth is required for video
• 24 channels of uncompressed 24bit/96KHz is still
relatively small
• SMPTE control data is also very minor
Stereoscopic 4K for D-ILA
CineGrid 2008 Demo By
Todd Margolis, Andrew Prudhomme, Jurgen Schulze, Robert Twomey,
Rod Sterling, Stuart Levy, Robert Patterson, Greg Dawe, Sheldon
Brown
Current Modes of Stereo 4K
Pre-recorded Animations
Rendered on parallel, multi-core
systems.
Live Interactive Computer Graphics
GPGPU computation on NVIDIA
With Multi-threaded x86
Hardware Diagram
(left eye)
Zaxel
Servers
(right eye)
(left eye)
JVC
Projectors
(right eye)
(left eye)
Dell
Workstations
(right eye)
Producing CG animations
• 29 Mental Ray licenses running on 15 dual
Opterons
• Scalable City (clip length 4:04)
244 seconds = 5856 frames = ~500 hours render time
Prepare Source Files
• Transfer data to encoding systems
– 12,067 frame transfer from NCSA to Cinethumper at UCSD
• Using RBUDP over optiputer, the speed varied greatly by hour of day
• 120-400Mb/s = 30-170 minutes per/animation
– Copy files from Cinethumper to 4 zaxels = 20-40 minutes @ 800Mb/s (SFTP)
• Transfer video to display systems
– Copy AVI movies from external drive = 4 x 10 minutes
Encode TIFF files into AVI movies
• 1 minute of movie takes ~20 minutes to encode
• Progressive or Interlace(psf)?
25 – 60 minutes for all 4 videos to encode
Zaxel AVI Encoding (lossless)
FILE New Media Art Festival
Sao Paulo Brazil, Yokohama Japan, San Diego US
1st ever 4K feature film to premiere streamed
to 3 continents at once.
Program had HD teleconference between 3
sites projected in 4K quadrants.
4K 70 minute film with 5.1 audio streamed along with HD teleconference to:
UC San Diego ---- FILE Sao Paulo --- Keio University - Tokyo
Jim Chen’s VLAN proposal using
Multicast in Chicago
T-LEX
GOLE
C-wave
vlan 2712
vlans
2711/2
U
Mackenzie
vlan 2711
Ampath
GOLE
USP
Final Configuration: unicast VLAN assignments
T-LEX
GOLE
C-wave
vlans
2712/3
vlans
2711/2
U
Mackenzie
vlan 2711
Ampath
GOLE
USP
4K Production Times
• 4K frame renders took over 24 hours
• Encoding to AVI movies took another 12 hours
• File transfers onto & off external drives took another hour
• Entire film transfer with ftp would take about 30 days ~5Mbps
• Using RBUDP entire film at 800 Mbps took about 40 minutes
HD transmission
Sao Paulo – SD – Keio
3 plans
Special Thanks to Hector Bracho for rendering these drawings
Out: component
In: component
Converte
r
Out: HD-SDI
In: Hd-SDI
HDV cam
In: analog
Audio AD
In: AES Out: AES/EBU
Ultragrid
Send to Keio
NETWORK
Ultragrid
Receive from
Keio
Out: HD-SDI Out: AES/EBU
Sao Paulo – Keio
PLAN A
1. Sao Paulo - Keio = Ultragrid bi-directional with AJA cards
a) send - ultragrid
b) receive - ultragrid
O1V
mixer
Sony SXRD 4K
In: HD-SDI
Out: component
HDV cam
In: component
In: analog
iHDTV
Send to SD
Out: analog
NETWORK
Sao Paulo – San Diego
iHDTV
Receive from SD
Out: component
In: component
Converte
r
Out: DVI
PLAN A
2. Sao Paulo - UCSD = iHDTVw/ Blackmagic
a) send - iHDTV
b) receive - iHDTV
Sony SXRD 4K
In: DVI
O1V
mixer
Out: component
HDV cam
In: component
In: analog
VLC
Send to SD
Out: analog
NETWORK
Sao Paulo – Keio
VLC
Receive from SD
Out: component
In: component
Converte
r
Out: DVI
PLAN B
1.Sao Paulo - Keio
a) send - VLC or iHDTV
b) receive - VLC or iHDTV
Sony SXRD 4K
In: DVI
O1V
mixer
Out: component
HDV cam
In: component
In: analog
VLC
Send to SD
Out: analog
NETWORK
Sao Paulo – San Diego
Qvidium
Receive from SD
Out: component
In: component
Converte
r
Out: DVI
PLAN B
2. Sao Paulo - UCSD
a) send - VLC
b) receive - Qvidium
Sony SXRD 4K
In: DVI
O1V
mixer
NETWORK
HDV cam
HD VTC
Out: analog
H.323
Out: component
Sao Paulo – SD – Keio
In: component
Converte
r
Out: DVI
PLAN C
1.Sao Paulo – SD - Keio
a) send – H.323
b) receive – H.323
Sony SXRD 4K
In: DVI
O1V
mixer
2 cameras, 2 HD systems & 2 4K Playback systems
Install Equipment
Zaxel 4K Servers
4K Streamers
UltraGrid HD
10Gb Switch
2 cities with very different configurations
Configuring networks and equipment for this level of
global service is currently non-trivial
Dozens of people reconfiguring switches on many
private networks
Many dedicated technical staff at each site to configure
content, computers, AV infrastructure, and networking
Some of these people don’t get to sleep for a few
weeks.
High Resolution Cinema development with same asset
pipeline as game environment. Mining virtual world for
behavior data and use as cinema prototyping platform for
rapid development of cinematics.
Virtual World Generated Movie
Trailer
http://www.sheldon-brown.net/downloads/sc_trailer_stereo_720_12142007.wmv
Future Cinema –
Milestones Year 1 and 2
•Utilize Virtual World as Cinematic Production
Environment
•Create “Virtual World Director” application
•Explore hybrid schemes of network delivery
methods utilizing global Lambda Grid.
•Test implementations of “Machinima” computed and
delivered across global Lambda Grid.
Future Cinema –
Deliverables Year 1 and 2
•Create real time 3D 4K projection environment
•Create 3D 4K movie production pipeline utilizing virtual
world as production environment
•Devise methodologies for rendering algorithms to
automatically generate 3D 4K frames
• Create database schemes that track changes in assets
from virtual worlds to cinematic uses of same assets.
[email protected]
crca.ucsd.edu/sheldon