Half-precision Floating-point Ray Traversal
نویسندگان
چکیده
Fraction (10 bits) Sign (1 bit) Exponent (5 bits) 16-bit floating-point format defined in IEEE 754-2008 standard Storage support on most of the modern CPUs and GPUs Native computation support especially on mobile platforms (Up coming nVidia Pascal desktop GPUs are announced to have native computation support) Pros: Smaller cache footprint (compared to "regular" 32-bit floats) More energy efficient Often doubles throughput Cons: Less accurate Already 65520 is rounded up to infinity
منابع مشابه
Using Half-Precision Floating-Point Numbers for Storing Bounding Volume Hierarchies
Bounding volume hierarchies, which are commonly used to speed up ray tracing, are heavily accessed during ray traversal. Reducing the memory footprint for the bounding volume hierarchies leads to a better cache hit ratio and, therefore, faster ray tracing. This paper examines an approach of using 16-bit halfprecision floating-point numbers to store bounding volume hierarchy inner nodes. Compare...
متن کاملWatertight Ray/Triangle Intersection
We propose a novel algorithm for ray/triangle intersection tests that, unlike most other such algorithms, is watertight at both edges and vertices for adjoining triangles, while also maintaining the same performance as simpler algorithms that are not watertight. Our algorithm is straightforward to implement, and is, in particular, robust for all triangle configurations including extreme cases, ...
متن کاملArbitrary 3D Resolution Discrete Ray Tracing of Implicit Surfaces
A new approach to ray tracing implicit surfaces based on recursive space subdivision is presented in this paper. Interval arithmetic, already used to calculate intersections in ray tracing and ray casting (numerically or subdividing 1D or 2D spaces), is now used here to implement a ray tracing based on reliable rays traversals into a potentially infinite octree-like subdivided space, eliminatin...
متن کاملAccelerating X-Ray CT Reconstruction using SIMD and Half Precision Floating-Point on Intel Xeon Processor
Our group worked on accelerating the X-ray Computed Tomography (CT) reconstruction with statistical image reconstruction algorithm, using single-instruction multiple data (SIMD) instructions to accelerate the regularizer part. Also, we used half-width floating point data format to mitigate the memory bandwidth problem. Our results show that we could achieve 2.5x speedup by combining SIMD instru...
متن کاملMixed Precision Training
Increasing the size of a neural network typically improves accuracy but also increases the memory and compute requirements for training the model. We introduce methodology for training deep neural networks using half-precision floating point numbers, without losing model accuracy or having to modify hyperparameters. This nearly halves memory requirements and, on recent GPUs, speeds up arithmeti...
متن کاملذخیره در منابع من
با ذخیره ی این منبع در منابع من، دسترسی به آن را برای استفاده های بعدی آسان تر کنید
عنوان ژورنال:
دوره شماره
صفحات -
تاریخ انتشار 2016