Aug 10–14 2014 / Vancouver http://s2014.siggraph.org/

Day 1: AUG 10 2014, SUNDAY

DAY 1 [1130..1230] nVidia see the big picture

largest flat wall screen
immersive displays
- MOSAIC with sync
- sync’s up to 4 gpus in workstations
- visbox, cave, quadro K5000 + 6000
UHD USING CLUSTERS
- Quadro Sync
- 40 quadro sync cards in a cluster
- control via NVAPI
- control + monitor using NVWMI
- BARCO/Elimit
not every surface is flat
- WARP + INTENSITY Adjustment API
- projection correction
- curved surfaces
- projection mapping
- projection mappping on to a one fifth scale physical car
quadro features for high resolution display walls
- MOSAIC
- GPU direct 4
- 4K at 60Hz
- Warp + Intensity API
- NVAPI/NVWMI – promatically control driver, remotely control remote gpu
From SD to 8K
- exponontial Pixel Growth
from HD to 4K & Beyond
- 1080p
- UHD 4K – 3840×2160
- DCI 4K – 4096×2160
- DILA 4K – 4096×2400
Driving ultra high res displays
- max single cable bandwithds
- DP
  - version 1.2 upto 4k@60Hz
  - version 1.1a upto 4K@30Hz
- HDMI
  - 2.0
- DVI
drive multi-displays
- independent desktops
with MOSAIC
- one large desktop: 4K 16 monitors
MOSAIC supported on NVS + quadro
- unified desktop up to 15 displayes
“mosaic with sync” features
- seamless tear-free displays
- sli bridge K6000, K5000
why is sync important?
multi-gpu sync
- framelock
- stereo lock
- swap lock
vertical sync
framelock/genlock
stereo lock
swap buffers
- taking longer than 1/60 sec
swap buffers in a cluster
- each nodes with different fps(60/30/15/10 fps)
- need a mechanism to ensure that each node will swap at the same time
nvidia opengl extension (via nvapi)
- swap group and swap barrier
- swap group -> provides synchronizaiton multiple gpus in a sigle host
- swap barrier -> provides syncronization of gups accross multiple nodes
- use rj45
what sync do
- sync multiple display
- coordinate buffer swaps
- align the scan out of multpile display, gpus systems
- maintain stereo alignment
- sync to an internal or external timing
G-SYNC game monitors
- different approach -> tells display when to update
- instead of wait until vsync happens
MOSAIC
bezel and overlap correction
NV-WARP – WARP + INTENSITY API

DAY 1 [0200,,..0300]

exchanging the material information
nvida mdl = material definition language
nvida ARC team
nvida iray
Iray Photoreal -> Iray interactive
nVidia UI composer
MDL usage example
- iray photoreal
nvidia iray
Iray 2014 Rendeing modes
- realtime – fast, 15pfs ~ 60 fps
- interactive – middle 20 fps ~ 2 fps
- photoreal – best, 10fps ~ min
traditional shading language part
- texturing => material definition => material implementation
new say
- MDL
  - proedural programming language
  - declarative material definition
- Renderer
  - rasterizer
  - raytracer
  - pathtracer
MDL is not a shading language
MDL
- materail
  - surface: bsdf(scattering), emission(edf, emission) intensity
  - volume: vdf scattering, scattering_coefficent, abosrption_coeffcient
  - geometry: displacement, cutout_opacity, normal,
  - backface
  - ior, thin_walled
MDL: elemental distribution functions
- BSDF(bidirectional scattering distribution functions)
- EDF(emissive distribution functions)
- VDF(Volumen distribution functions)
distribution function combiner
- normalized mix, clamped mix, weighted layer
- fresnel layer, custom curve layer, measured curve layer
measured materials
- fast scan for believealbe material
- quantitative measurements for predictive rendering
  - there is a device to measure the lights
AxF from X-Rite Spatially Varying (SV) BRDF
- Analytic material model
- measurement drives model parameters
  - paper: Practical SVBRDF capture in the frequency domain, siggraph 2013
measured isotropic BRDF
- 3d data
- can be used as any other SBDF in MDL
  - device to get light : radiant zemax imaging spheres device
Improving Dark Grazing Angles
procedural proramiming language
- pure, side-effect free functions with read-only render state
  - texture lookup
  - texture placement
  - proedural textures
- Iray 2013: Built-in functions provide a flexible instruction set for runtime execution
- Iray 2014: fully programmable environment – clouded sky video
- Iray 2015: full programmability – without losing GPU acceleration
C-like language
additionals benefits of MDL
- physically-based materials are an easy-to-use paradigm
- supports modern rendering algorithms
- allows simple compilers and early optimizations
- enables fast rendrers, especially on parallel architectures
- GPU firendly
- supports material catalogs
Practical layered material example
- customer asked a “shader” to render realizstic cloth materials
  - color changes and complex
  - reference for requirement
  - “a practical microcylinder appearance model for cloth rendering”, sig 2013, iman sadeghi
  - usually this endes up complex, rendering engine dependent shader code
- core ideas from the paper
  - if far, it’s not distinguishable
  - cloth has complex anisotropic behavior
  - threads are bent which creates secondary highlights
  - warp and weft might be of different olors, -> complex color variation.. 90 vertical, 0 horizontal
- make it in MDL
  - warp -> anisotropic glossy BSDF
  - weft -> anisotropic
  - warp + weft -> complex reflective behavior of cloth
  - transluenccy
  - the full material
just by using mdl, building blocks
- no more complex shader writing
- good building blocks
MDL in commercial products
- Iray for Maya
- mental ray
- Sketch up . bloom unit
- realistyServer BIM IQ
- UI composer
demo
- nvidia Iray viewer
  - download
at the end of 2014
- mentalray will support iray
GREAT SESSION
- IT’S
supporting MDL
- MDL specification is publicically available
  - MDL SDK => part of Iray SDK
  - contailns compiler and optimizer
- Lightworks
  - nvidia’s partenr in Iray and MDL on your application
www.nvidia-arc.com => Iray => MDL
- lightworks design booth #916

DAY 1 [0300..0430]

BLENDER FOB, room 3 east hall
5 mil downloades, 1% active user => 50,000 active users?
open source
- ILM – openexr, alembic
- pixar – opensubdiv
- sony – open shading/color/image
- disney – ptex
blender foundation
- found in 2002 june
- hires 5 devs for support work
- handle donations and store incomes

Day 2: AUG 11 2014, MONDAY

DAY 2 [1000..1100]

BOF First Steps in Large-Scale Rendering
- [susan(pixar, white blonde), xx(), xx, xx, xx, david(pixar 19 years)] panel
- pixar david, 19
- ruca, from cube
- ?? from singapore
- **, toonbox
- **, google, softare in pixar and animallogic
- susan: pixar global rendering
- Mark Hills
- David Laur
- [email protected]
big farm problem
selling softare for the small render farms
- growing your farm
- some caution
  - hetro geneous workers
  - HPC (high performance computing)
very often it’s linux because of the
multiple levels of caches in pixar render farm
- go for memory upgrade
- it’ll be better on IO node
using farm for computing, processing, simiulation, general purposes
- 95% is rendering in pixar -> because of the GI
- 5% is simulation, using all IO
segement the pool
remote rendering
- route the low IO
- desktop is high end gpus
- linux =>
who’s doing GPU based the farm stuff
- use GPU based farm
one frame per core
4G per slot -> number of thread
bottle neck -> droppping
- finding the sweat shop
killing the machine
- keeping the tap on the production
multiple layers of controls
- most of the queueing
- multipile jobs coming in a smaller scope
8 years ago, hope, multithreaded
- the realizity -> it’s not scale
- 24 core machines into 6-8 blocks
different versions and renders
- 60% utilization=> cpu
- threading control
- 70-80%
always battle with cpu utilization with the vray settings
- renderman, arnold
licensing ->
- 1 license per machine -> tanking with jobs in one machine
rendering in the cloud
- local farm busy
- during the spike -> send to the remote farm
keep high io local
- pick the high io render
how to pick high io???
data transfer to the remote is an issue..
- make it fast to get it back and forth
- latency between the managed render service ??
- remote mounting from the
traffic is
- syncing based file system
- asset management is structured.
checking out in the cloud
- 10Mbit -> typical studio
- speed of the turn out
- 1Gbit -> at teast for studio
- i think we have 20G bit
pixar -> centralized in house *
small studio needs a turn key solution ???
- provision them remote
sync or reupte
upload mananged servcies -> recommended??
- upload the files through the web interface -> not good
- too slow
- managed service..don’t give you a lot of support, email that’s it
cost of the administration
configuration management software
multithreading on the virtual machines
provisioning the vm
- get the core.
  - gareenteed the core, io -> provision
- nfs over the network -> latency and slowness, cache on vps
is it a nfs, real solution??
segarate the traffic -> read from textrue, write to other file
traffic graph -> on the farm mount, know right away which file system is saturating the network
nfs
solution for the small company,
caching and latency
- capital expance vs operation cost -> tax credit, decision where to go
- financial advantage,, not only technical values
shortlist of priorities *
software tagging on the worker
scripts needed for rendring
- how to wrap around?
- send all render as an one big script?
wrapper script, lock
- artists setup and the farm
design and the pipeline
opensource scheuler ??
- 2 yeras ago -> liek 10
- opendrag,,
good reference
- best support of the vendors of the queuing software
amazon mka

DAY 2 [0530..0700]

threading
- threading => intel tbb: use it all of them
- openmp
- c++ thread
- boost threadin
problems
- standarize tbb
pixar’s system => presto
the reality is
- use tbb
multithread vfx
multhreading and vfx .com
- get the contact information
using tbb ->
anyone using tbb
- in gaming company, are using tbb on multiple gaming project
- platform specific apis
- internal framework wrapper
- os differences
- how do you finetuning betwen the different flatforms
  - follow the least constraints
  - what’s your targets? -> 4 cores?
  - job based, use core until the jobs done
  - until the stop is locked
  - game
  - do you know anything that will scale to?
  - 12 or 16 cores?
why not tbb?
- openmp is easier
tbb, not good for length
- microsft tpl -> we use opensource
c++ standard will have something similar
- similar to c++ 0x
jame from intel
- a few thoughts from intel’s perspective
- true industry implementations
  - open cl, openmp port io,
- standard is slow
  - tbb or current collections?
  - 2007 -> tbb, filling the gaps
- lambda in c++11
gpu computing
- cuda vs opencl
- sebastian from nvida
- 4 categories of reasons to use gpu computing
  - hardware: unified virtual memory. get the variables, fully debuggerable hardware.
  - tools do debug memory,
  - profiling
  - collection of libs
  - threast?, has cpu fall back
  - optics lib, raytracing on the gpu
- what’s holiding up?
  - industry standards? -> monopoly
  - how serious
  - sweat spot?
- highly unpredictable
- why not use gpu?
- unified memory -> mem and gpu ram
- like to debug.
  - gpu = 100x of cpu
  - honest pair cpu vs gpu: 5 to 10x
rasment in disney animation studio
- gpu 2x4x speed up
  - real optimzation on cpu
  - got 20 to 30x in the cpu
- real problem not doing much of gpu
  - volitability of the environment, hardware and display driver
- was painful 2-3 years on linux
- design algorith that can scale
  - 10 of 1000s of threads
- why not use cuda?
  - code maint cost
  - design first on cpu
- weta digital
- just optimizing and profiles
really goes down to standard platform
- standard c++, opencl, tbb
cuda is getting mature
larger data set -> 28G ram
- need a bigger memory
- trying
10G memory footprint on window.
gpu with 24G ram
why do we need gpu?
care about the energy efficiency?
- data center cooling problmes
- battery hot, cooling
performace and cooling?
power distribution in the data center
DOE, exso scale comuting
1000 cores -> parallel programming
what goes inside of the cpu
energy efficient -> time efficient
- avoid moving the data around
- optimizing the data movement
- main memory vs cpu or gpu
- data movment within the chip, l1, l2 and registers
triple nested for loop to pass the asignment ->
- intel xxx using what/
- assignment from the code
intel vtune amp -> cpu and memory bandwith per socket
mpc : layer upon layer *
autodesk maya animation
- 95%
- using tbb
- dreamworks session -> two way to improve
  - graph level parallism =>
  - node level parallism =>
- motion builder -> make it past?
  - threading at the beginning
  - internal scheduler
  - thread classes
  - split and runnign as they want
- story tool
  - not happy with the speed
  - optimize the speed
- autodesk to dreamworks
  - dreamworks prima
  - built from scratch
  - 20 years old
  - thread safe
  - keeping it thread safe
- whole lot of unit test
  - threading bug in the code
  - you can fix it
  - used tbb a lot -> hitting some limitations
- intel new thing -> sbb(vecotrization)
  - check it out after siggraph

DAY 3: AUG 12 2014, TUESDAY

TUESDAY 0945..1045

amd
mantle
- new low level programming interface for windows pc
- designed in collaboration with top game developers
- lightweight driver that allows direct access to GPU
- good for big game: battlefield 4
mantle benefits
- higher graphics performance
- empowering lower spec system
- predictable performance behaviour
- sharing optimizations between PCs and game consoles
- new rendering techniques
  - more flexible, can be used on other gpu like intel and nvidia
why a new api?
- success of the GCN architcture
  - uniform d
- many collaboration with gaming industry and AMD
- not a replacement for industry-standard API like DirectX or OpenGL
the small batch problem
- draw calls are generated by the CPU to create work for the GPU
- too many draw calls can make a system CPU-limited, leaving GPU under-utilized
- batchig many objects into a single draw call adds code complexity and constrains artists
  - 3-5K draw calls
  - 10K draw calls
  - draw calls possible with heavy optimizations
  - 100K+ -> draw call target with mantle, 60fps
multi-threading
- BEFORE
  - multi-threading gmae engines using directx and opengl
  - cpu0 -> game
  - cpu1 -> render
  - cpu2 -> dirver render
  - doesn’t scale beyond 2-3 cores
  - latency limits performance
  - driver thread often boecomes a bottleneck
- AFTER
  - with mantle
  - cpu0 -> game
  - cpu1 -> render
  - cpu2 -> render
  - .. cpu5 -> render
  - scale linearly with number of cores
  - minimal latency
  - no driver thread overhead
MANTLE core features
- app-controlled execution model
- generalized resources
- monolithic pipelines
- fiexlible memory management
- page talbe remapping
- hybrid resource binding model
- bindless -> huge in next
MANDTLE extended features
- advanced flow control
- advanced anti-aliasing functions
- multi-device support
- asynchronous compute
prog graphics application potential to use mantle
- graphics api overhead
  - large/complex cad & DCC models consist of many individual assemblies, not easy or in many cases
  - AEC models of complex
  - Mantle reduces API overhead, avoiding inefficient state changes between those
- multi-threading
  - CAD
- memory management
  - cloud/vdi solutions may require more efficient use of graphics memory, partitioning, allocation,etc
  - mantle can provide
- fast syncronization between compute eigne and graphic engines
- drive optimizations
  - avoid driver bottlenects
  - dumb driver
- data transfer control over PCIE
- optimzations
  - optimize shader code to gain performance
  - use highly optimized code for specific
- vertical market segments where mantle can help
  - cad, medical
- to get the evaulation copy of firepro mantle driver
  - [email protected]
  - firepro developer sdk
  - http://developer.amd.com/tools-and-sdks/graphics-development/firepro-sdk
- intel is approaching to them
ACTION ITEM
- needs a slide
- it’s a replacement of opengl or directx
- will it be good? not sure

TUESDAY 1100..1200

maya
- what’s new in maya 2015 ext 1
- could import with dropbox a 123?
- openubdiv surface
- alembic
- biforost -> prt
- xgen
- rendering
  - vp2.0
  - color management
  - maya hardware can be saved as exr
retopology options
- auto and guided retopology – mudbox
- cloud integration: reality capture data <- maya 2015 ext 1
- gpu caching
- data exchanging
- maya quad draw -> manual retopology
MAYA 2015 new modeling toolkit
- Mesh Editing Tools:
  - Target Weld
new boolean
- will give you cut and uv will be there too
- gemoetry + texture
new reduce tool
- looks good, poly reduction
- make nice clean geometry
- some triangles, ngons quad -> click quad draw tool in modeling tools
  - hit magnet tool
  - and use quad draw
how to clean the dense scan model data
- file -> send to unity, send to unreal
- file -> game exporter
- fbx review tool: free download
- file -> cloud import/export => a360, 123D, dropbox
- pipleline cache -> gpu ahe export -> exported to alembic
- grab mesh
- use magnet tool
- quad draw tool
- dragging and draw the quads
- remeshing the surface
- relax tool
- soft seletion
- sticking to the original surface
- transfer map option
- look at uv first
new uv editor
- plannar mapping
- grab edge of the head
- cut and open up the head
- fenecing
- unfold3dunfold
- unfold3d optimize
supports ptex file
ACTION ITEM
- get the slide
- trynew modeling tool
- ptex
- try export to alembic

TUESDAY 1230..0130

OSL open shading language from sony image works
more renderers and apps
released OSL 1.5
- reparameter (enables lookdev tools)
- shadergroup – specific attributes
  - per-group renderer output
  - re
featured change: group pickling
- http://opensource.imageworks.com/?p=osl
send the param as a string
serilized a shader group
archive a shaer group, inluding .oso file
- debugging
- benchmarking/ optimization target
- can be run with testshader or other tools – no renderer needed!
on deck
- make it faster for SIMD-oriented renderers:
- output cuda kernels somehow
- optimized new usage cases
- ecosystem

pixar

new version of renderman
renderman RIS & OSL wayne wooten
native ray trace renderer
whats’ RIS?:
- integrators,
- VCM bi-directional path tracer
- BXDFs
  - glass, skin, disney
- patterns
  - osl, SeExpr, Texture, Voronoise
SSS
Why RIS?
visual effects, 2014
- simplfy workflow
- fast feedback
- easy to use
- heavy gemoetry
- volumes & particles
- raytrace everything
typical day at work
- check last night’s renders
- dailies
- breaks
- meetings/ coffee
- make it pretty(4h)
- launch nightly build
- 4-6 hours to get things done of your time
OSL
*** image from a wooden clock from carbon frame
- oak feeding Disney BXDF
  - from Michel J. Anders
- carbon fiber feeding Blend BXDF
  - from v0ray examples
- backdrop 1 line simpletex.osl
- OSL api is easy to use
OSL UDIM Texture
- used in Mari
  - stands for U-dimension
  - UDIM(u,v) = 1000 + (10*u) + v + 1
OSL UDIM Texture
- written in c++
OSL UDIM Texture
- original naive implementation was very slow
  - vtune to the rescue
  - threading took a hit( multi-thread usting)
  - string manipulation was shlow
- what to do?
  - bottleneck was OSL codebase
  - discussed with Larry Gritz
  - just under a week it ran fast
OSL UDIM Texture
- open source really works
Evaluation Results
- OSL pros
  - feels ‘homey’ to RSL users
  - targeted language and features
  - avoids plugin versionitis
- OSL cons
  - disngle point, lack of SIMD/coherence
  - JIT time
  - LLVM targets libm.so,

OSL Gaffer

John Haddon, imageengine
What’s gaffer?
- applicationf framework for VFX
- node based
- open source (BSD)
  - cortex,
two birds
- gaffer -> lots of node, to manipulate scene hirerace
- none for deforming each objs
- need to allow
- play with OSL
  - no renderer suppported it
loading shaders
- search for shaders on env[“OSL_SHADER_PATHS”]
  - build menu
node graph
executing shaders
- conver node network to OSL::shadingAttribState
- implement renderservices::get_attributes() to provide access to prim vars and image channels
- call shading system::execute() with fake shaderglobals
- convert colsurecolor result ack to prim vars and image channels
- abuse debug() closure
noise
show some demo
further work
- UI implenetation
- optimisation
- using osl as a shading language
- image convolution/ edge access
- nei
http://github.com/imageengine/gaffer

general heckenberg? animal logic
physically based renderman pipeline => with renderman
very successful in 4-5 studios
“lego movie”
plastic -> lighting directions, specular lights
retraining the renderer translator
shader translation

TUESDAY 0200..0300

BOF, alembic
no major releases
moved to github/alembic/alembic
a little more collaboration
focus on stability, optimization
a few improvement, namespace support,
slow dev now.
will add hooks on importer and exporter
doulbe precision issue
game use abc becuase it contains more info than fbx
ACTION ITEM
- check out github repo

TUESDAY 0300..0400

vray: sig: vray day1: Digital Domain | Rendering Heavy Assets: X-Men’s RFK Stadium
merged light cache
only 1 gi compuattion for the whole sequence
tree
- 400 individual trees
- billons of polygons
- 42 species,
- speed tree -> speedbridge => sent to DD pipeline
- lookdev done in maya and 7vray
- vray proxy -> full gemoetry -> vrayproxy
  - super fast scene handling
  - fast export and renders

Tehe RFK Stadium

lidar scan to sue as modeling reference
stadium size adjusted to fit over the whilehouse and match the live action
40k vs 60k seats
main strcutre: concnreat
small details: lights, air con boxes, pipies and seats
split into 7 sections
shading and textures applied to what’s in camera only
- keep the geo -> for gi calc
- blew shading and texture
textures are from
lookdev done in maya & vray

destroying everything, stadium

stadium lifted by its metallic structure
use of our deferred render pipieline to render the seat
create a vrscene for a single seat -> one vrscene contains geo, mtl, tex
massive open file format
- CDL rigging description
- APF aniamtion info
render
- load vrscene ->
landed stadium
- destoyed stadium modelling from FX drop simulation
- new faces created for the inside parts of the cube
- no uvs on the new faces
- first step ->
  - rgb mask color set in the vrayproxy accessed via vrayvertexcolor node
- world space noise not “sticking” to the geometry
- adding ‘rest’ attribute as another color set
- noise mapped on the inside faces using rest color set as coordinate space
deep images
- smoke coming out from houdini
- all renders done with vray
- breaking up was all done in houdini
vgo
- tree -> proxy, geo proxy
render at 2K or 3K
- using the external render time
- amazon ec2?
- rendering everything in house
- render with vrimage and convert to exr
ACTION ITEM
- should apply on the farm
- one gi computation
- vrayproxy
- CDL, APF, vrscene

TUESDAY 0345..0445

Gaffer: vfx
kartick will go, i won’t

TUESDAY 0400..0500

renderfarming
file size -> use isilon
cloud -> use managed service and set up the bandwidth

DAY 4: AUG 13 2014, WEDNESDAY

WEDNESDAY 0900..1000

user case SVA AT NYC
open pipeline
priority
- render speed
- job frame count
- thesis prgoress
- submission time
- limit by the users
improvement
- using shutgon
- monitoring student progress
crunch time
- tracke
pipelinefx -> cube
- they are using shotgun and cube
future impromvemnt
- utilizae asset managmentment
  - seamless progcessing
- better integration with shotgun
- work and render over chould services
- limit concurrent jobs per student
- assign individual render days for seniors
- show eta all the time
- how many threads per job
- hyperthreading
render management software
discussion points
- automation vs learning the manual way
- universal pipeline management system for education
windows
- use AD to control user directory
- qube -> tag the job with your name
- qubeproxy user -> service
how to define # of threads
ACTION ITEM
- way to scale the job
- check out “open pipeline”
- check out the thread per job and hyperthreading

WEDNESDAY 1000..1100

openscenegraph
state of play : robert osfield
all the stuff will be on the OSG website
osg-3.2.1 stable release
JMVP joint medical visualization project
digital learning foundation -> go check the website
volume rendering
scripting
osg ui
- using qt5
improvements
- script support
- create custom widgets in lua
- create behaviours/callbacks in lua
- modifying scene graphi within lua
- complete user interface in lua
osg-3.4 release plan

WEDNESDAY 1030..1130

study on the cloud rendering
guy from a montreal
- chris -> director of technology digital domain, 20 yeras,
- coal peterson -> business render
- -> studio cloud
- dave psoar -> softimage, autodesk, mentalray, ILM Rnd in light and rendering. in montreal
- daren gran -> consultant, CTO DD, CTO dreamworks animation
dashboard and monitoring
- render farm -> squeze every dollar out
- becomes like a noise function
- visualization of the realtime data is really important
IO, cpu, swap, thread
loving islon
next gen animaiton session ->
creative iteration -> and fit them in budget
human engineering
cloud
- manager in the cloud
- all thing in the cloud?
- SAS ->
- overflow render workers
- amazon DWS
- performance issue?
- configuration nightmare
- studio cloud
- join AWS submit
  - C-section
  - new industry
user interface of the future render farm
- web based is way to go
- open up
- concerned about the security
  - they can’t see the name of the jobs from the other showss
- RESTapi
- control apis, browsers
- next gen ui customization -> web based
commercial product
- are you happy customer?
- reporting, artists friendlyness,
rendering 2D
- error reporting
- look up the log or python traceback
- different ui for different departments
scheduling interesting in animal logic
- bulk of the jobs are rendering but processing
singular of inhouse schedular
optimization
- utilization? throwput?
compute scheduling, render scheduling
segarate all of your data
- connect the file
gamifying the submission
- web view ->
- public shaming
sprung and elastic search
pixar -> john, susan fong,
digital domain ->
pipeline -> event driven
- rabbit mq
- using event backend and render management
hooking up the scheduling based
properl data coming in
- stats dat
- graphis
gpu computing
- really great at gi computing -> beauty pass
- depends on the the projects
- evolve so fast
- problem with the driver
nvidia open c??
- dedicate the
very diverse
- dollar intensive
- low cost commotiy
ACTION ITEM
- get the info about next gen animation sessionx
- join AWS submitt ->
- add security in the crazyq.. per projects, login based

WEDNESDAY 1130..0130

exhibitions

WEDNESDAY 0130..0230

cg in usa + canada
full sail university, winter park, florida

WEDNESDAY 0100..0200

vray: method stuido
- transforer movie
fx
- on volumentric elements simulated in houdini and rendered in vray

WEDNESDAY 0200..0300

vray talk
- ILM, Jan, captain america
using vray for katana
vray – 3 years, gi-joe
using render man for a while
started doing alembic
ACTION ITEM
- check out open image io
- https://sites.google.com/site/openimageio/home

WEDNESDAY 0300..0400

vray talk
- scannline, 300
- 25 years in cg, 5 yrs in
max and maya pipeline
vray is a core, used since v1
flowline => major scanline tool -> tightly integrated with vray
scannline
goes to custom vray shader
8K texrtures
- ??? exr
change some of the animations
vray shader out of box works well
flowline -> watersolver, fluid solver,
- fire, smoke and water
- anything takes less than 12 hours in the farm
questions
- reusing fx sim -> can’t do
- incamera depth of field ->
- vray-RT
ACTION ITEM
- check out flowline
- ask hubbard about this?
- ask hubbard about the best texture format for vray

WEDNESDAY 0400..0500

vray talk
- kevin marko, chaos group
- kevin, sup blur 12 years
blur reel 2013 lorez
- league of legends
- bioshock infinity
- load of the rings
- batman and justice league
- starwars
- halo
- starwars
batman archam night
- game trailer
- WB -> rocksteady was a client
- they have game models in progress
- hair -> vray hair system
- hair shader
- love architectures -> can save the geo
- looped the building 800,000 lights
- more shader test
- all animation done by the mocap
thor 2 prologue
- thrusday morning
- concept design
  - helmet, shield
- got digital doubles from vfx
- upraise them
- used vray skin texture to create skin variation and differnt morph targets
- making hugo
- creating a cloud
  - card changing to the different camera view
- showing different angle of the image sequences based on the camera
- battlefield look dev
- nuke and mari pipeline
construct
- nvidia
- entirly gpu rendering project
- vray -rt gpu
- 500 sec -> 45 sec renders
GPU rendering
- small team
- massive computational power
- small footprint
- make it more responsive
- vray rt -> only on max?
rendering final
- renderd in 5 days
  - 3x 3dboxx 4920 ->
  - 3x nvidia k6000’s
  - 6x nvidia k40s
  - 3dboxx 4920 gpu edition
- construct teaser at 4K, stereo, 60fps
- used nvidia vca cluster
  - reduced a 10 year workstations cpu -> 5 days
  - 14 min on single vca
  - took 9.5 hours on vray cpu
- VCA ??
- vray for motion builder
  - virtual camera capture
  - ability to capture a virtual camera location
  - compose in real time to photorealistic scenes
- realtime rendering is happening in motion builder
- http://www.boxxtech.com/products/workstations/single-socket/3dboxx-4920-gpu
- roughly around $20,000
how did you get a fund?
- sponsers -> boxx, nvidia, chaos group, autodesk motion builder,

WEDNESDAY 0500..0700

opengl es 3.1
11 years of opengles
- 2002 working group
- 2003 1.0
- 2004 1.1
- 2007 2.0
- 2012 3.0
- 2013 3.1
what’s new since sig 2013
- release 3.1 at GDC
es 3.1 goals
- bringing ogl 4.x to mobile
  - game-changing functionality: compute shaders, indirect drawing
  - other advanced featured
- enable very rapid adoptation
  - expose hidden capabilities of current devicds
  - backward compatibility with es 2.0/3.0
  - run on any es 3.0 HW
- improve application portability
  - tighter specs means less undefined behaviour
  - ever stricter conformance testing

ogl es 3.1 overview

bill qualcom
what’s new
- compute shaders
  - tighly coupled graphics and GPU computing
- draw-indirect
  - draw command parameters from memory instead of from args
- SSO (separate)
compte shader basics
- opencl, opengl
- use gpu for general purpose computing
  - new single-stage pipeline, separate from the graphics pipeline
- opg es
  - create a thread per vertex and a thread per fragment
- glDispatch Compute()
- SSBOs shader storage buffer objects
- images
  - like texture, but writable
  - 4 image binidng points
  - image atomics avail as ext
- serialziation and synchronization primitives
  - atomic counters
  - shader memory barriers
  - api memory barriers
draw-indirect
- new way to dispatch work on gpu
  - draw call parameters stored in gpu memory
  - glDrawArraysindirect()
  - glDrawElementsindirect()
  - glDispatchComputeIndirect()
  - useful for draw parameters set by GPU
  - for example, in a compute shader
separate shader objects
- mix and match vertex shaders and fragment shaders
Texture features
- multisampled textures
- stencil textures
- textrue gather
shader language features
- array of arrys
  - int a[2][3[4]
- explict uniform location
  - layout(location=0) in vec4 position; // vertex shader
  - layout(location=2) out vec4 diffuse; //
- shader bitfield ops
  - extract, insert, reverse, pack/unpack, count, lsb, msb
  - extended precision support
  - frexp / ldexp
- shader helper invocation
  - bool gl_helperinvocation

ogles 3.0 ecosystem rollout

implementaions from every major mobile GPU vendor
- tools, docs, sdks, etc
platform:
- ios 7, iphone 5s >
- android 4.3
unity mobile survey: 8% -> 21% adoptation
rollout es 3.1
- doc, sdk, emulator, tools
- imageination, nvida, qualcomm
platform support
- adroid L
- android extension pack, ASTC textures, etc
conformance test improvements
- updated es 3.0 conformance relased in jan 2014
- es 3.1 con is live 6 june 2014
- 13 submissions from
  - arm, imagination, intel, nvidia, qualcomm, vivante
  - desktop: intel, nvidia
reference compiler
- now supports virtually all of GLSL ES 3.1
- compute shader
  - (minus arrays of arrays)
- images
- atomic counters
- SSBOs
- texture gather
- texture LOD
- SSO
- .. and mores

#version 310es
layout(localsize z= 1000) in;
void main()
{
 barrier();
 }

glslangValidator: test.comp

see -> khronos.org/ refernce compiler

basemark os I opengl es 3.1

rightware 2009 finland, www.rightware.com
biz area
- ui creation tool, kanzi solutions
- systems and chip performance evaluation software
ogles 3.1 test overview -> avil Q4 2014
demo
- water using compute shader
- particles 10000
- post -processing
  - bloom, dof, anitializsing,e vignette
- SSAD
- deferred lighting G-buffer

google android extension pack (AEP)

google android extension pack (AEP)
- android L mandates ogl es 3.1
aep es3.1 extensions
aep
- tessellation
- geometry shaders
- ASTC texture compression + more
AAA grpahics effects
- global illumination
- physically-based shading
- deferred rendering
- HDR tone mapping
- smoke and particle effects

GFXBench Car Chase Preview – OpenGL ES 3.1 + AEP

hdr deferred rendering pipeline
phycially-based shading
DOF and motion blur
4k and astc support

GLES compute shader vs OpenCL

next gen opengl initiative

will talk a lot about why???
opengl is a huge success
problems with ogl today
- gl programming model deosn’t match with modern gpu architeture
- arcane and archanic syntax
  - legacy
  - needless complexity
- cpu intensive
  - state validation, dependency tracking, error checking
  - hard to predict where cpu load will occur – varies with implementation
- not multicore-cpu-friendly
  - primitive threading model, inconsisitenlty implemented
  - it’s broken right now
- implementation variability
  - spec looseness
incremental change is not enough
goal on next gen
- clean, modern architecture
- multi-thread /multicore friendly
- reduce cpu overhaead
- architurecture-neutral
- predictable performance
- improved reliability and consistency
key design principle: explicit control
- apllicaiton is telling the driver what it wants
status
- Tom Olson -> Chair
- Bill Licea-Kane(Qualcomm) -> Language chair
- Graham Sellers(AMD) and Jeff Bolz(nvidia) -> API editor
who’s on board
- sony, pixar, oculs, blizzard, epic games, ea, apple, unity, google, EA, RTT, amd, qualcomm
- samsung, nvidia, imagination, , broadcom, arm, amd, hicorp, vivante, valve

OpenGL ecosystem support in Unity

democratizing games
demo game: ios “monument valley”
feature unity 5.0
- deploy to webgl
unity multiplatform deployment

[ios, android, windows, ps, web, xbox, wii, browsers, linux]
platform not using opengl backend

=> playstation, xbox,
driver support
- www.g-truc.net
drivers and supports
- nvidia 340.52
- amd 14.7 rc1
- intel 10.18.10.3652
- mesa 10.2.2
- macosx 10.9.4
opengl 4.2 is on most driver
reevaluation the opengl implementations
- now it’s good
opengl in unity
- extensions
  - gl_arb_es2_compatibiliyt
  - gl_arb_es3_compatibility
  - gl_arb_es3_1_compatibility
- gles 4.1/ gl 4.3 or 4.2 + compute shader stuff
- player base level-> es2.0
- editor base level-> gl3.2
next ecosystem
- 600K monthly ative users
- 750K registered
- 3 million registered developers
engine, support, learn, cloud, community, asset store
[glnext, ogl, dx, libgnm, mantle, metal]

3 speakers

Johan Andersson, TD, Frostbite – EA

Valve

John McDonald, Valve
gl is important
valve
- valve did
  - steam for linux
  - steam runtime
  - steamos
  - VOGL -> debugger
- valve contributed
  - glslang
  - LunarGLASS/ LunarGOO -> help opengl debugging
portal on android
- uses ogles 2.x
grpahics today
- direct3d is the primary on desktop
- only on windows
- tied to os ded10/11 are vista_
  - dx11 -> 87%
  - dx10.1 -> 15%
  - dx10 -> 13%
  - dx9 -> 4%
- gl distribution
  - ogl 4.4 -> 67%
  - gl 3.3 -> 29%
  - gl 3.1 -> 4%x
glnext is shaping up
- new and modern api

WEDNESDAY 1030..1200

global vr meetup
SVVR: http://svvr.com/

WEDNESDAY 1200..1300

Still Handling Very Large Data Sets – to Mars and Beyond, nasa, db&w finest plugins
http://www.jpl.nasa.gov/
Managing very large datasets, with an emphasis on planetary visualization, at the media lab at the Jet Propulsion Laboratory – a journey from the past to future.
We’ll be talking about the use of very large datasets at the JPL as well as large datasets in general. We’d love to see you joining us there.
bmw, JPL

XX

topics
- history of tiled megatextures
- current state of tiled / megatextures
- tecnical challenges
- future challenges
- other datasets
basic principle of tiled-megatextures
- mipmaps
- different requirements
  - offline/ prouction rendering
  - realtime
offline / production rendering
- pixar (surprise) textrue on demand
- peachey 1990 PRMan 3.0 / tin toy et. al
achivement unlocked: tour de france 2000
starwars ep2
- pod racing sequence (1999)
ideas spread a lot quicker
realtime
- clipmapping -> ** term for mipmaps
- terravision
- keyhole(wow!) / google earth
- quick diversion: legal minefields
- games: IdTech 5 -> jon carmac used this
Idtech5/rage
- one huge texture -> changed the way level artists are working
state of the art
- common feature in production renderers
  - offten due to using the open source opeinimage io lib
  - some home-made solutions as well
- mainstream in games
  - idtech5 and rage as a tech showcase
  - directly supported in state of the art apis
  - improved tools for texture artists
technical challenges
- focus on production rendering
  - multiple cores
  - cache management
  - io management
  - io bandwidth
  - use in rendering (ie displacement)
  - suitability of image file formats -> exr, tif are fine..
future challenges
- io and cpu power grows and rhe requirements
- gpu rendering/computing
other large datasets
- meshes
  - huge single meshees
  - repeated meshes
  - procedureal meshes
  - terrain, displacement
- particles
  - KRAKATOA
- large scale rendering – Manuka/Arnold
  - two renders has mesh compression of meshes
- meshes
  - huge single meshes
  - in memory compression
  - load on demand, segment to mesh parts
  - spatting (for viewing of point clouds, qsplat)
  - volumetric solutions
- repeated meshes
  - instancing
- procedural meshes
  - “cheap” to recreate during the render or may be traced directly:
  - terrain
  - displacements
meshes: terrains
- tesslate, optimize
meshes/point clouds: splatting
volumetric rendering
[email protected]
ACTION ITEM
- research huge texture technology used in rage 5 from idtech

Siggraph

Siggraph 2014

Day 1: AUG 10 2014, SUNDAY

DAY 1 [1130..1230] nVidia see the big picture

DAY 1 [0200,,..0300]

DAY 1 [0300..0430]

Day 2: AUG 11 2014, MONDAY

DAY 2 [1000..1100]

DAY 2 [0530..0700]

DAY 3: AUG 12 2014, TUESDAY

TUESDAY 0945..1045

TUESDAY 1100..1200

TUESDAY 1230..0130

pixar

OSL Gaffer

TUESDAY 0200..0300

TUESDAY 0300..0400

Tehe RFK Stadium

destroying everything, stadium

TUESDAY 0345..0445

TUESDAY 0400..0500

DAY 4: AUG 13 2014, WEDNESDAY

WEDNESDAY 0900..1000

WEDNESDAY 1000..1100

WEDNESDAY 1030..1130

WEDNESDAY 1130..0130

WEDNESDAY 0130..0230

cg in usa + canada

WEDNESDAY 0100..0200

WEDNESDAY 0200..0300

WEDNESDAY 0300..0400

WEDNESDAY 0400..0500

WEDNESDAY 0500..0700

ogl es 3.1 overview

ogles 3.0 ecosystem rollout

basemark os I opengl es 3.1

google android extension pack (AEP)

GFXBench Car Chase Preview – OpenGL ES 3.1 + AEP

GLES compute shader vs OpenCL

next gen opengl initiative

OpenGL ecosystem support in Unity

3 speakers

Valve

WEDNESDAY 1030..1200

WEDNESDAY 1200..1300

XX

Popular

Latest

Category

Tag Cloud