Milestone 4.2 Enhancements Stereo/multicam rendering: The output statements have been extended to accept an optional camera name as the first parameter in an outputs string. Custom exporters or host application plugins can use this to setup multi-camera renders. The new additional syntax is: camera AOV_name AOV_type filter driver # for a regular driver camera driver # for a raw driver Multi-camera renders of the same scene can share the reading in and processing of files, acceleration structures, tessellation, displacement and texture cache by having the different cameras render sequentially. Interleaved bucket rendering for different cameras or sending the results from multiple cameras to the same multi-layer driver is not yet supported. (#2404) Improved motion blur sampling quality: Motion blur sampling has been improved, resulting in less noise given the same AA sample and time budget. The improvement is most prominent in scenes where motion blur is the primary source of noise in the image. There are now two possible values for the AA_motionblur_pattern option: the new default pattern, and the legacy jittered pattern which is provided as a backup until the new sampler is sufficiently battle-tested in production. (#2131, #4037, #4044, #4099) Improved glossy sampling quality: Some scenes will now render with slightly less noise given the same amount of samples and time. In particular, microfacet-based BRDFs viewed at grazing angles will show significant improvement. (#3571, #3888, #3900) Improved CPU utilization: The thread scheduler has been redesigned achieving much better CPU utilization in scenes that get stuck in a particularly slow bucket near the end of the frame. It is no longer necessary to use very small buckets to prevent these slowdowns. (#3873) Improved memory performance and usage in Linux: Linux-specific optimizations have been added which reduce memory usage and improve the threading scalability in certain situations, usually during scene preprocessing and shutdown. These require matched AiMalloc() / AiFree() allocations. If you allocate memory in your shader/procedural with AiMalloc() and then free it with free(), it is possible Arnold could crash. (#3779) Faster ray accel build: Acceleration structure construction is now up to 1.3x faster while using the same memory. (#3898) Faster subdivision: Catmull-Clark mesh subdivision is now up to 1.5x faster and uses about 5% less peak memory. (#3903, #3913, #3949) Faster displacement: Polymesh vertex displacement has been sped up, particularly for lower thread counts. On some highly subdivided models weve seen around 2x faster displacement even with 8 threads. In addition, vector displacement artifacts on UV seams have been eliminated. (#4008, #4017) Faster polymesh processing: Polymesh sanity checks are now about 3x faster, so we have removed the seldomly used options.mesh_sanity setting and now always perform these checks. Mesh processing/optimization time is now reported in the stats and this step is about 1.5x faster than before. (#3912, #3917, #3927, #3952, #3986) Faster rendering of many-light scenes: Scenes with many spaced out lights will now render faster thanks to improved culling of low contribution lights. Weve seen up to 2x faster renders in some challenging production scenes. (#4100) Faster SSS: Raytraced diffusion-based SSS should now show higher performance on objects that have not been assigned to be a part of an SSS set via a sss_setname user data string, specially on scenes with spatially overlapping objects, like hair over skin. (#3808) Faster image shader: Texture map lookups coming from the image shader are now significantly faster. The noise shader has been made slightly faster too. (#3963, #4114) Faster AiNodeGet* and motion vectors: AiNodeGet*() is now substantially faster on multi-threaded machines. Before, excessively calling these functions would cause Arnold to not scale to multiple threads. Now Arnold is able to scale when these functions are called during rendering. There is still a substantial overhead for calling these functions and so AiNodeGet*() usage is still discouraged from being called during shader evaluate. Usage of AiWorldToScreenMatrix(), AiShaderGlobalsGetPixelMotionVector(), and AiDriverGetMatrices() are all substantially faster now and will scale to multiple threads. One benefit of this is that computing motion vectors is now substantially faster. (#3096, #3836, #3849) Faster string processing: We now scale better when handling strings. For instance, if many threads are frequently making calls that pass in strings, such as AiNodeDeclare(mynode, some_string, uniform RGB) we will now scale to all these threads and execute this call more quickly. This matters mostly in deferred loading of geometry from DSO/.ass procedural nodes. (#3921, #3957) Faster AiNode(): Performance scaling has been improved when many threads are rapidly calling AiNode(), as it can happen with deferred loading of geometry from DSO/.ass procedural nodes. Under normal situations, this bottleneck is not hit and so no speedup will be seen. (#3954) Faster deep output using less memory: Renders that output deep images are now up to 15% faster and use less than half the memory. (#4081, #4102) Faster AOV composition with no memory overhead: The memory overhead incurred when usingoptions.enable_aov_composition has been eliminated. This can save gigabytes in highly-threaded, high AA, multi-AOV renders. Complex renders with lots of transparency and AOVs could be up to 20% faster. (#4038, #4065) Faster image output: Taking advantage of the non-locking process_bucket API call introduced in 4.1, we have increased concurrency and improved the speed at which AOVs are written to disk, with a 15% to 30% speedup in built-in drivers like driver_exr, driver_tiff, etc. Custom drivers written by third-party developers should be updated accordingly for maximum performance. (#3857) Faster backlighting: The translucency, or backlighting computation in the standard shader, controlled with the Kb parameter, will use on average 50% fewer shadow rays. (#3890) Skip skydome importance table if color is not linked: Skydome lights will no longer build their importance tables when the color is not linked to anything. This can save a few seconds of preprocessing time when using high resolution tables. Also, IPR renders will now automatically respond to changing the skydome color. (#3020) autobump in SSS: Arnold can now optionally take into account the effect that autobump has on the ray traced BSSRDFs result via the sss_use_autobump render option. This helps capture the high frequency details of the surface more accurately, at the expense of slightly longer render times. For backwards compatibility and performance reasons, this option defaults to false. (#3872, #3894) Smoother bump near shadow terminator: We have fixed long-standing faceting issues near the shadow terminator boundary when using strong bump mapping. The improvement is most noticeable with detailed, high-frequency maps or when seen from far away. (#3891) Reduced edge darkening in microfacet BSDFs: Thanks to a new less aggressive shadowing/masking term, the built-in Cook-Torrance BRDF and microfacet BTDF now conserve a bit more energy and exhibit less edge darkening, particularly at high roughness settings. The impact is expected to be very subtle in most scenes. (#4043) Anisotropic Cook-Torrance BRDF: The Cook-Torrance BRDF functions in the shading API as well as the standard shader now actually make use of the anisotropic shading parameters. This BRDF has an anisotropic shading effect that is very similar to the anisotropy in the Ward-Duer BRDF in behavior, but with a greater amount of energy conservation and less edge darkening. (#4042) Improved ambient and reflection occlusion: Ambient occlusion artifacts around edges of low-poly objects have been cleaned up. In addition, the object visibility attributes (visibility, self_shadows) or the minimum occlusion distance (mint > 0 in the AiOcclusion call itself) can now be used to remove artifacts near silhouettes for reflection occlusion, or to disable occlusion from bump-mapped normals. (#3972) Ambocc distance in utility shader: Added a new ao_distance parameter in the utility shader to control the length of ambocc rays for scenes where the default of 100 units is too short. (#3861) Refraction opacity AOV: The standard shader now can output refraction opacity as an AOV. It is advised to write the complete RGBA set of channels, since the alpha component will constitute a mask that can be used to mix it in with the regular opacity, multiplying in refraction opacity to regular opacity based on the mask. (#2880) Motion-blurred rolling shutter: Cameras now have a rolling_shutter_duration parameter with which it is possible to control the duration of exposure of the scanlines in a rolling shutter camera. Valid values for this parameter are in the 0 to 1 range, where a value of 0 gives you an instantaneous exposure of each scanline (the default value and the rolling shutters previous behavior), and a value of 1 exposes every scanline for the entirety of the cameras shutter interval (the same result that a camera without rolling shutter would give). (#4009) Support for custom shutter shapes: Added an additional curve value to the shutter_type enum in cameras. This, in conjunction with a new camera parameter shutter_curve, allows arbitrary shutter shapes, linearly interpolating shutter open/closed values between points. You can define as many points as needed. Coordinates are increasing in X from 0 (corresponding to shutter_start) to 1 (corresponding to shutter_end), and values in Y must be non-negative. (#3934) # example shutter_curve usage persp_camera { ... # double-trapezoid shutter with outer edges being more gentle, # and the first trapezoid emphasized much more shutter_type curve shutter_curve 8 1 POINT2 0.0 0.0 0.2 3.0 0.3 3.0 0.4 0.0 0.6 0.0 0.7 1.0 0.8 1.0 1.0 0.0 } Volume plugins: A new node volume has been added which allows volume plugin DSOs (specified via the dso parameter of a volume node). Volume plugins allow a plugin writer to wrap any volume format or generator for efficient rendering. Volume plugins provide volume creation, destruction, channel sampling, and gathering of ray extents, as well as an automatic bounding box and step size for ray marching (although these can be overridden by the user in the volume node). Ray extent computations inside the plugin will tightly bound volume data along the ray so that the number of volume shader evaluations are drastically reduced in sparse volume datasets. For a short tutorial on how to implement your own volume plugin DSO, see this article on the arnoldpedia. (#3466, #3978, #3996, #4045, #4046) Density shader: A new built-in shader density is available which uses AiVolumeSampleChannel() to sample from the new volume plugins. This allows volume sampling to work regardless of what source the volume data comes in (as long as there is a volume plugin for the source). Currently it has parameters for controlling RGB scattering, absorption, emission, forward vs backward scattering strength, interpolation quality, and offsetting the sampling position (for example with a noise shader to create the appearance of a higher resolution volume than what came from the source data). (#3466, #3994) Static OIIO linking in Windows: Just like we do in Linux and OSX, we now statically link the OpenImageIO library in Windows too; we dont distribute the file OpenImageIO.dll anymore. (#3969) Scene update on free mode: When using the free render mode, you can now force a scene update using AiRender(AI_RENDER_MODE_FREE), which will take into account parameter changes and nodes created or destroyed. (#3862) Multithreading in procedurals: The enable_threaded_procedurals option no longer affects the loading of the text-based .ass, .obj and .ply formats, which are now always loaded in parallel. This option will still apply to binary procedurals (.so/.dll/.dylib), forcing sequential processing when disabled. (#4096) IES files in texture_searchpath: If not found in the default path, photometric IES files are now searched for in texture_searchpath too. (#4036)
Posted on: Wed, 04 Jun 2014 17:15:02 +0000
Trending Topics
Recently Viewed Topics
© 2015