We explore a first proof-of-concept example of creatively using the Turing generation's hardware ray tracing cores to solve a problem other than classical ray tracing, specifically, point location in unstructured tetrahedral meshes. Starting with a CUDA reference method, we describe and evaluate three different approaches to reformulate this problem in a manner that allows it to be mapped to these new hardware units. Each variant replaces the simpler problem of point queries with the more complex one of ray queries; however, thanks to hardware acceleration, these approaches are actually faster than the reference method.
Compressed wide bounding volume hierarchies can significantly improve the performance of incoherent ray traversal, through a smaller working set of inner nodes and therefore a higher cache hit rate. While inner nodes in the hierarchy can be compressed, the size of the working set for a full traversal stack remains a significant overhead. In this paper we introduce an algorithm for wide bounding volume hierarchy (BVH) traversal that uses a short stack of just a few entries. This stack can be fully stored in scarce on-chip memory, which is especially important for GPUs and dedicated ray tracing hardware implementations. Our approach in particular generalizes the restart trail algorithm for binary BVHs to BVHs of arbitrary widths. Applying our algorithm to wide BVHs, we demonstrate that the number of traversal steps with just five stack entries is close to that of a full traversal stack. We also propose an extension to efficiently cull leaf nodes when a closer intersection has been found, which reduces ray primitive intersections by up to 14%.
Mesh colors provide an effective alternative to standard texture mapping. They significantly simplify the asset production pipeline by removing the need for defining a mapping and eliminate rendering artifacts due to seams. This paper addresses the problem that using mesh colors for real-time rendering has not been practical, due to the absence of hardware support. We show that it is possible to provide full hardware texture filtering support for mesh colors with minimal changes to existing GPUs by introducing a hardware-friendly representation for mesh colors that we call patch textures. We discuss the hardware modifications needed for storing and filtering patch textures.
Data movement, particularly access to the main memory, has been the bottleneck of most computing problems. Ray tracing is no exception. We propose an unconventional solution that combines a ray ordering scheme that minimizes access to the scene data with a large on-chip buffer acting as near-compute storage that is spread over multiple chips. We demonstrate the effectiveness of our approach by introducing Mach-RT (Many chip - Ray Tracing), a new hardware architecture for accelerating ray tracing. Extending the concept of dual streaming, we optimize the main memory accesses to a level that allows the same memory system to service multiple processor chips at the same time. While a multiple chip solution might seem to imply increased energy consumption as well, because of the reduced memory traffic we are able to demonstrate, performance increases while maintaining reasonable energy usage compared to academic and commercial architectures. This article extends our previous work E. Vasiou, K. Shkurko, E. Brunvand, and C. Yuksel, "Mach-RT: A many chip architecture for high-performance ray tracing," in Proc. High-Perform. Graph. Conf., 2019 with design space exploration of the L3 cache size, more detailed evaluation of energy and memory performance, a discussion of energy delay product, and a brief exploration of boards with 16 chips. We also introduce new treelet enqueueing logic for the predictive scheduler.
Although interactive ray tracing has been around since the late 1990s, real-time frame rates had so far only been feasible for low and mid-size screen resolutions. Recent developments in GPU hardware, that specifically accelerate ray tracing, make it possible for the first time to target head-mounted displays (HMDs), which require constant high frame rates as well as high resolution images for each eye. This allows for utilizing ray tracing algorithms in novel virtual reality scenarios, which are impractical to do with rasterization. In this short paper we present our experiences of applying real-time ray tracing to the problem of detecting cosmetic defects in sheet metal stamping simulations by creating a virtual light cage.
We present a technique for real-time ray tracing with the goal of reaching 240frames per second or more. The core idea is to trade spatial resolution for faster temporal updates in such a way that the display and human visual system aid in integrating high-quality images. We use a combination of frameless and interleaved rendering concepts together with ideas from temporal antialiasing algorithms and novel building blocks---the major one being adaptive selection of pixel orderings within tiles, which reduces spatiotemporal aliasing significantly. The image quality is verified with a user study. Our work can be used for esports or any kind of rendering where higher frame rates are needed.
Shadow volumes are a popular technique to compute pixel-accurate hard shadows in 3D scenes. Many variants exist that trade off accuracy and efficiency. In this work, we present an artifact-free, efficient, and easy-to-implement stencil shadow volume method. We compare our method to established stencil shadow volume techniques and show that it outperforms the alternatives.
Monte Carlo ray tracing offers the capability of rendering scenes with large numbers of area light sources---lights can be sampled stochastically and shadowing can be accounted for by tracing rays, rather than using shadow maps or other rasterization-based techniques that do not scale to many lights or work well with area lights. Current GPUs only afford the capability of tracing a few rays per pixel at real-time frame rates, making it necessary to focus sampling on important light sources. While state-of-the-art algorithms for offline rendering build hierarchical data structures over the light sources that enable sampling them according to their importance, they lack efficient support for dynamic scenes. We present a new algorithm for maintaining hierarchical light sampling data structures targeting real-time rendering. Our approach is based on a two-level BVH hierarchy that reduces the cost of partial hierarchy updates. Performance is further improved by updating lower-level BVHs via refitting, maintaining their original topology We show that this approach can give error within 6% of recreating the entire hierarchy from scratch at each frame, while being two orders of magnitude faster, requiring less than 1 ms per frame for hierarchy updates for a scene with thousands of moving light sources on a modern GPU. Further, we show that with spatiotemporal filtering, our approach allows complex scenes with thousands of lights to be rendered with ray-traced shadows in 16.1 ms per frame.
We introduce stochastic lightcuts by combining the lighting approximation of lightcuts with stochastic sampling for efficiently rendering scenes with a large number of light sources. Our stochastic lightcuts method entirely eliminates the sampling correlation of lightcuts and replaces it with noise. To minimize this noise, we present a robust hierarchical sampling strategy, combining the benefits of importance sampling, adaptive sampling, and stratified sampling. Our approach also provides temporally stable results and lifts any restrictions on the light types that can be approximated with lightcuts. We present examples of using stochastic lightcuts with path tracing as well as indirect illumination with virtual lights, achieving more than an order of magnitude faster render times than lightcuts by effectively approximating direct illumination using a small number of light samples, in addition to providing temporal stability. Our comparisons to other stochastic sampling techniques demonstrate that we provide superior sampling quality that matches and improves the excellent convergence rates of the lightcuts approach.
Nowadays computing is heavily-based on accelerators, however, the cost of the hardware equipment prevents equal access to heterogeneous programming. In this work we present Brook GLES Pi, a port of the accelerator programming language Brook. Our solution, primarily focused on the educational platform Raspberry Pi, allows to teach, experiment and take advantage of heterogeneous programming on any low-cost embedded device featuring an OpenGL ES 2 GPU, democratising access to accelerator programming.
In this paper we describe and evaluate an implementation of CPU-style SIMD ray traversal on the GPU. We show how spreading moderately wide BVHs (up to a branching factor of eight) across multiple threads in a warp can improve performance while not requiring expensive pre-processing. The presented ray-traversal method exhibits improved traversal performance especially for increasingly incoherent rays.
We propose and evaluate what we call Compressed-Leaf Bounding Volume Hierarchies (CLBVH), which strike a balance between compressed and non-compressed BVH layouts. Our CLBVH layout introduces dedicated compressed multi-leaf nodes where most effective at reducing memory use, and uses regular BVH nodes for inner nodes and small, isolated leaves. We show that when implemented within the Embree ray tracing framework, this approach achieves roughly the same memory savings as Embree's compressed BVH layout, while maintaining almost the full performance of its fastest non-compressed BVH.
In this short paper we present a machine learning approach to detect visual artifacts in rendered image sequences. Specifically, we train a deep neural network using example aliased and antialiased image sequences exported from a real-time renderer. The trained network learns to identify and locate aliasing artifacts in an input sequence, without comparing it against a ground truth. Thus, it is useful as a fully automated tool for evaluating image quality. We demonstrate the effectiveness of our approach in detecting aliasing in several rendered sequences. The trained network correctly predicts aliasing in 64 x 64 x 4 animated sequences with more than 90% accuracy for images it hasn't seen before. The output of our network is a single scalar between 0 and 1, which is usable as a quality metric for aliasing. It follows the same trend as (1-SSIM) for images with increasing sample counts.
A primary advantage of deferred shading is eliminating wasted shading operations due to overdraw. We present a new algorithm that we call Deferred Adaptive Compute Shading, for providing further reduction in shading computations. Our method hierarchically shades the image while reducing the number of required shading operations to below one shading computation per pixel on average. We determine whether to shade a pixel or approximate it using previously shaded pixels around it, based on an estimate of the image variance at the pixel location. The algorithm is designed to dynamically reconfigure itself to achieve optimal warp coherence and measurable performance gain. We extensively evaluate our algorithm, demonstrating that it produces high-quality results and is robust and highly scalable while providing significant performance improvements in complex scenes.
We introduce moment transparency, a new solution to real-time order-independent transparency. It expands upon existing approximate transmittance function techniques by using moments to capture and reconstruct the transmittance function. Because the moment-based transmittance function can be processed analytically using standard hardware blend operations, it is efficient and overcomes limitations of previous techniques.
Geometric aliasing is a persistent challenge for real-time rendering. Hardware multisampling remains limited to 8x, analytic coverage fails to capture correlated visibility samples, and spatial and temporal postfiltering primarily target edges of superpixel primitives. We describe a novel semi-analytic representation of coverage designed to make progress on geometric antialiasing for subpixel primitives and pixels containing many edges while handling correlated subpixel coverage. Although not yet fast enough to deploy, it crosses three critical thresholds: image quality comparable to 256x MSAA, faster than 64x MSAA, and constant space per pixel.
We introduce a pragmatic algorithm for real-time adaptive super-sampling in games. It extends temporal antialiasing of rasterized images with adaptive ray tracing, and conforms to the constraints of a commercial game engine and today's GPU ray tracing APIs. The algorithm removes blurring and ghosting artifacts associated with standard temporal antialiasing and achieves quality approaching 8x supersampling of geometry, shading, and materials while staying within the 33ms frame budget required of most games.
Rendering 3D-360° VR video from a camera rig is computation-intensive and typically performed offline. In this paper, we target the most time-consuming step of the VR video creation process, high-quality flow estimation with the bilateral solver. We propose a new algorithm, the hardware-friendly bilateral solver, that enables faster runtimes than existing algorithms of similar quality. Our algorithm is easily parallelized, achieving a 4× speedup on CPU and 32× speedup on GPU over a baseline CPU implementation. We also design an FPGA-based hardware accelerator that utilizes reduced-precision computation and the parallelism inherent in our algorithm to achieve further speedups over our CPU and GPU implementations while consuming an order of magnitude less power. The FPGA design's power efficiency enables practical real-time VR video processing at the camera rig or in the cloud.
It is generally accepted that Poisson disk sampling provides great properties in various applications in computer graphics. We present KD-tree based randomized tiling (KDRT), an efficient method to generate maximal Poisson-disk samples by replicating and conquering tiles clipped from a pattern of very small size. Our method is a two-step process: first, randomly clipping tiles from an MPS(Maximal Poisson-disk Sample) pattern, and second, conquering these tiles together to form the whole sample plane. The results showed that this method can efficiently generate maximal Poisson-disk samples with very small trade-off in bias error. There are two main contributions of this paper: First, a fast and robust Poisson-disk sample generation method is presented; Second, this method can be used to combine several groups of independently generated sample patterns to form a larger one, thus can be applied as a general parallelization scheme of any MPS methods.
This paper presents MoonRay, a high performance production rendering architecture using Monte Carlo path tracing developed at DreamWorks Animation. MoonRay is the first production path tracer, to our knowledge, designed to fully leverage Single Instruction/Multiple Data (SIMD) vector units throughout. To achieve high SIMD efficiency, we employ Embree for tracing rays and vectorize the remaining compute intensive components of the renderer: the integrator, the shading system and shaders, and the texturing engine. Queuing is used to help keep all vector lanes full and improve data coherency. We use the ISPC programming language [Intel 2011; Pharr and Mark 2012] to achieve improved performance across SSE, AVX/AVX2 and AVX512 instruction sets. Our system includes two functionally equivalent uni-directional CPU path tracing implementations: a C++ scalar depth-first version and an ISPC vectorized breadth-first wavefront version. Using side by side performance comparisons on complex production scenes and assets we show our vectorized architecture, running on AVX2, delivers between a 1.3× to 2.3× speed-up in overall render time, and up to 3×, 6×, and 4×, speed-ups within the integration, shading, and texturing components, respectively.
