OTOY Presentation - 2016 NVIDIA GPU Technology Conference - April 5 2016
Click here to load reader
-
Upload
otoyinc -
Category
Technology
-
view
59.432 -
download
3
Transcript of OTOY Presentation - 2016 NVIDIA GPU Technology Conference - April 5 2016
PowerPoint Presentation
The Future of Rendering #GTC16 #OTOY #OctaneRenderOTOY Inc.April 2016
Rendering in 2020: Looking ahead 5 yearsUnbiased spectral path tracing - 16K/eye (90 deg.) HDR 240+ fpsLive and cinematic IP, games, WYSIWYG social sandbox all convergeHolographic VR, AR, MR, light fields -> mobile / volumetric / FV displaysCloud Rendering - hybrid adaptive live streams (web-like) vs. app/binaryGPU FTW* better and better rays/Watt on client + server (*until RT ASIC?)
2015 Recap
OTOYs Mission:Practical digital holographic* content creation and publishing for everyone*(Digital Hologram: 8D light field volume + depth + reflectance)
OCTANERENDER AND SPACE STATION 76
3
OCTANERENDER AND SPACE STATION 76
OCTANERENDER AND SPACE STATION 76
OCTANERENDER AND SPACE STATION 76
OCTANERENDER AND SPACE STATION 76
OCTANERENDER AND SPACE STATION 76
OCTANERENDER AND SPACE STATION 76
OCTANERENDER AND SPACE STATION 76
OCTANERENDER AND SPACE STATION 76
3.X
OCTANERENDER AND SPACE STATION 76
3.0May 15th
3.0 OXR Octane Film Buffer
OXR a file which stores a snapshot of the Octane film buffer (its a subset of ORBX)
Links and/or embeds to assets required to resume rendering are included (not needed w/ ORC)
All imager/post processing nodes work on film buffer, not baked in OXR file
ORC import or export OXR film buffers. Key requirement for commercial launch
3.0 Deep Motion buffers (OXR)
Deep Motion = multi-light for time/space
Render @ 24 fps but output 240 frames per sec.
Huge savings just in frame and scene load times...
Essential for VR renders where 90 120 fps is baseline BTW Carmack says 300 fps!
Can also be used to speed up stereo rendering
3.0 ORC OctaneRender Cloud
ORC is deeply integrated in Octane 3.0 and all 3.0 plug-ins going forward
Cloud render is one a click away within Octane. Each ORC render can pop open a web page to view status/result.
Delta synching only changed assets are sent to ORC. Huge savings in bandwidth vs. earlier workflows.
ORBX/OCS versioning ORC can roll back to a previous snapshot sent to ORC.
3.0 ORC Octane Live App Stream (experimental)
Users can run Octane and ORC entirely from a browser on any device even on a plane w/ LTE hotspot:
3.0 ORC subscriptions
Monthly subscription covers storage, I/O and other overhead - $9.99/month (ORC Test Drive)
Render credits based on GPU minutes (~40 OctaneBench).
ORC jobs mapped into 20 GPU workloads (~800 OctaneBench) for best price/perf ratio in $9.99 buckets
Load times+ export times minimized across nodes ORC is best for multi-frame or tiled rendering
3.0 ORC 2016 roadmap
Aiming for 50% more power per GPU minute once service scales close to last year RTM numbers
CPU jobs like OKX compression can run for free while GPU is rendering next frame
Adding more regions and many more providers over the coming months
AWS, Azure, SoftLayer and OTOY LA (mini ORC region)
3.0 May launch overview
May 15th Release of Octane 3.0 and ORC
OR 2.x -> 3.0 = $49 upgrade ($399 for new licenses)
ORC Test Drive = $9.99 (1000+ GPU render minutes)
3.0 Plug-in updates in line with 2.x path
Whats next?
OCTANERENDER AND SPACE STATION 76
3.1
Alpha 3.1 build as of today
3.1 CUDA cross compiler
CUDA is an amazing GPGPU toolchain.
Were sticking with CUDA across all OTOY projects for now
We plan to target new platforms such as Android, OSX and consoles through this unified codebase
3.1 CUDA cross compiler for non-GPUs
CPU will be first backend we ship in 3.1 alpha wont be fast (20 core Xeon < GK104 )
thinking about outlier hardware options beyond GPU and CPU
Each new GPU family/device we support requires 2-8 weeks of optimization work including Pascal (next)
3.1 Octane Shader Compiler
Part of larger compiler work being done at OTOY
OSL is subset of Octanes shader VM
AOV shaders are in Octane but extensions to OSL are needed to expose them fully
With 3.1 we are introducing API for any shader language - with OSL. as default
3.1 Octane Shader Compiler
We are working with NVIDIA on an MDL front end for the Octane Shader Compiler
Also exploring GLSL (would love to see shader toy textures)
On CPU we support both JIT and VM modes for shaders to cover all platforms
3.1 ORBX native plug-ins
Designed to keep Octane 3.1 features growing with new 1st and 3rd party extensions for long time to come
Limited only by what Octane C API exposes but that is a lot!
ORBX plug-ins may be pulled in to Octane node system like LiveDB materials must be whitelisted.
3.1 ORBX native plug-ins
ORBX plug-ins can be offered, sold or licensed through otoy.com alongside OR and DCC plug-ins
ORBX packager in 3.1 adds crypto, digital signing and various lossless compression options for all ORBX node types
ORBX plug-ins can also be used in ORBX Media Player
3.1 ORBX plug-in ecosystem
Scripting: looking at V8 plug-in for JS/LLVM script nodes
Textures: Substance Designer plug-in being developed
Physics plug-ins for rigid, soft and fluid dynamics (e.g. TFD)
3.1 ORBX Denoiser case study
Atlus ORBX native WIP plug-in
Hooked into OR film buffer system
Needs its own GPU (3 GB+ VRAM)
But results are really nice even at 16 spp!
3.1
3.1
3.1
3.1
3.1 ORBX native plug-ins:
Shader languages: OSL, MDL, GLSL et al
Media/3D file handlers: glTF ( in place of FBX), CT scans for medical data as volumes
Widgets and UX extensions
UDIM/PTEX plug-ins (implemented in some apps, should be supported widely)
3.1 Native skinned meshes and bones
Optimized character animation inside Octane for dynamic characters
Basis for new animated mesh types (other than .abc)
Exploring glTF instead of FBX as it is more compact and can also be supported in ORBX media player
3.1 Refinement of 3.0 features:
New Displacement and Rounded Edges codebase
Adaptive Sampling
UX scene tools (proxies for audio, light field volumes, and scene editing)
Faster Mesh loading/compiling
Whats next?
OCTANERENDER AND SPACE STATION 76
engine
OctaneEngine Major expansion of Octane ecosystem
OctaneEngine = Octane API refactored as ORBX module + running as web/socket server
Foundation for major features beyond 3.1
Legacy app/plug-in model is starting to hold us back..
OE will be subscription option in tandem w/ 3.1 box licenses
OctaneEngine Zero GPU mode
Octane plug-ins will no longer require a GPU on local machine connect to OE server over LAN (or WAN on high speed connection).
Max, Maya and Blender 3.0 plug-ins pushed out until now so we could include OE API support
Speed is looking on par w/ local GPU over 1 Gbps
OSX - MacPro (aging D700s), MacBook (no AMD/NV GPU) et al.
OctaneEngine New rental price points for artists, studios and pros
$19.99/month same features as Octane standalone/network slave
$9.99/month - Octane on up to 2x GPUs. Ideal for students, smaller studios, or remote second license
OE Plug-ins (using new OE API) will be $9.99/month
OE bundles/suites (e.g. 2, 3 or more plug-ins) priced at $19.99/month
OctaneEngine New Enterprise Plans (monthly pricing TBD)
Network rendering - up to 200 GPUs via certified GPU appliances w/ 10GE+ networking
ORC in a box create a virtual private ORC region using your GPU appliances. All the features of the ORC public cloud, but provisioned in your own DC.
NVIDIA and other partners working on HW certification path.
OctaneEngine Asynchronous Render Graph API
New API allows Octane to build node trees with asynchronous I/O and queues similar to ORC but internalized in Octane node system in future OE release.
Simple use case: baking camera -> texture input of light mapped mesh -> real time OR preview
API for light field photon cache could be used for biased rendering shaders and plug-ins.
OctaneEngine Collaborative Scene Graph API
Allows multiple clients to connect to a single OctaneEngine node and receive a unique viewport for the current render target
Changes in the scene graph can be handled by any connected application through new OE API (under development)
Powerful framework for merging app workflows e.g. Motion Builder + Max +Nuke + Unreal Engine all controlling and editing the same scene across LAN
OctaneEngine WebSocket API
OctaneEngine exposed as web service via WS/TCP and UDP. This is coming to ORC down the line
Web desktop recreating Octane Standalone UI in HTML5 combined with OE web service, means no need for standalone app.
Services like Vizor.io and clara.io can tap into Octane as a web service for live viewports or offline jobs powered by ORC credits.
OctaneImager
OctaneImagerCompositing and post processing framework -> OE
Independent filter and imaging SDK
2D and deep pixel/deep motion effects input for info channels, mesh data as well as multiple beauty passes layers and rendertarget groups
Can be used for 2D media passes (e.g. OXR/EXR deep images) independently of OctaneEngine
Filters and NPR effects via Octane Shader Compiler
OctaneImager
OctaneImager
OctaneImager
Brigade - Towards Real Time Path Tracing
Dynamic noise free rendering in Brigade ( ~3200 OctaneBench )
vs. single 980 (~1/32 the GPU power @ 100 OctaneBench )
Brigade - Towards Real Time path tracing at 120+ fps
Octane Towards Real Time path tracing at highest quality
OCTANERENDER AND SPACE STATION 76
1, 2, 3 2012-2016
OCTANERENDER AND SPACE STATION 76
42017
OCTANERENDER AND SPACE STATION 76
4+ Brigade
4
OctaneRenders cinematic final frame quality
Brigades speed
Interactivity >= OpenGL / DirectX legacy pipeline
4
Octane 4 integrates all of Brigades features to date
Supported in all OE plug-ins including final releases of Unity/UE4 and Motion builder plug-ins
Focus on next gen HW for much better speed
Deeper APIs for game engines especially UE4!
4
Filtering + smart sampling to get noise free PT in real time
Mobile + Console backbends e.g. Android/Vulkan, PSVR and iOS (key to VR/AER)
Speed ups will be as impactful for offline rendering
Much faster IES lights, point lights. Etc.
HAIR / FURThank you!