Ray tracing hardware

Lore & Background

The problem of rendering 3D graphics can be conceptually presented as finding all intersections between a set of primitives (typically triangles or polygons) and a set of rays (typically one or more per pixel). Up to the 2010s, mass market graphic acceleration boards, called GPUs, used rasterization algorithms. The ray tracing algorithm solves the rendering problem in a different way, finding all intersections of a ray with a set of relevant primitives. Rasterization can be performed using a stream computing model, one triangle at a time, but non-local effects such as reflections and shadows are difficult, and refractions nearly impossible to compute. Ray tracing can easily compute various kinds of physically correct effects, providing much more realistic impression than rasterization, but anything other than ray casting requires recursion and random access to the scene graph.

Various implementations of ray tracing hardware have been created, both experimental and commercial. In 1995, Advanced Rendering Technology (ART) was founded in Cambridge, UK, based on a 1994 PhD thesis, to produce dedicated ray tracing silicon (initially the AR250 chip). Products were first shipped to customers in 1998. In 1996, researchers at Princeton University proposed using DSPs to build a hardware unit named TigerSHARK. From 1999 to 2002, implementations of volume rendering using ray tracing algorithms on custom hardware were carried out by Hanspeter Pfister and researchers at Mitsubishi Electric Research Laboratories with the vg500/VolumePro ASIC, and with FPGAs by researchers at the University of Tübingen with VIZARD II. In 2002, the computer graphics laboratory at Saarland University produced prototype ray tracing hardware including the FPGA based SaarCOR chip and a more advanced programmable processor, the Ray Processing Unit (RPU).

In 2009–2010, Intel showcased their prototype Larrabee GPU and Knights Ferry MIC. In 2009, Caustic Graphics produced the CausticOne plug-in card that accelerated global illumination by organizing scattered rays into more coherent sets. In 2010-2011, Siliconarts developed RayCore, the world's first real-time ray tracing semiconductor IP. In 2013, Imagination Technologies, after acquiring Caustic Graphics, produced the Caustic Professional's R2500 and R2100 plug-in cards containing RT2 ray trace units. In 2014, Imagination Technologies announced the PowerVR Wizard family and its first GR6500 implementation for mobile and embedded uses. In 2018, Nvidia introduced their GeForce RTX and Quadro RTX GPUs with hardware-accelerated ray tracing using an RT core, featuring BVH traversal, compressed BVH node decompression, ray-AABB intersection testing, and ray-triangle intersection testing. In 2020, AMD announced RDNA 2 micro-architecture support for real-time hardware accelerated ray tracing. In 2021, Imagination Technologies announced the IMG CXT GPU. In 2022, Samsung, Arm, Qualcomm, and Intel each announced hardware-accelerated ray tracing in their respective products. In 2023, Apple announced the A17 and M3 chip family with hardware-accelerated ray tracing.

Reader's Guide

The significance of ray tracing hardware lies in its ability to accelerate a fundamentally different approach to 3D rendering compared to traditional rasterization. As the article describes, rasterization struggles with non-local effects like reflections, shadows, and refractions, while ray tracing can compute physically correct effects for a much more realistic impression. The article notes that ray tracing is inherently suitable for parallelization, but requires recursion and random access to the scene graph for anything beyond ray casting. The development timeline shows a progression from experimental and off-line systems (ART's RenderDrive in 1998) to consumer real-time hardware (Nvidia's GeForce RTX 2080 in 2018). The article also highlights that ray tracing hardware has been implemented in various forms: dedicated chips (AR250), FPGA prototypes (SaarCOR), plug-in cards (CausticOne, Caustic Professional R2500), integrated GPU blocks (PowerVR Wizard, Nvidia RT core, AMD RDNA 2), and mobile SoCs (Samsung Exynos 2200, Apple A17). The legacy of this hardware is that it enables real-time ray tracing in consumer graphics cards, mobile devices, and embedded systems, moving the technology from offline rendering to interactive applications. The article also notes that a neural processing unit (NPU) is sometimes used to further accelerate the output from real-time ray tracing hardware.

Did You Know?

• The AR250 chip was described as 'the first time ray-tracing has been reduced to a single-chip design', achieving performance at '15 times the speed of a 266-MHz Pentium II processor'.
• The GeForce RTX 2080 and 2080 Ti became the first consumer-oriented brand of graphics card that can perform ray tracing in real time.
• RayCore, announced in 2011, is described as the world's first real-time ray tracing semiconductor IP.
• The complexity of a well implemented ray tracing algorithm scales logarithmically due to objects being placed into BSP trees or similar structures.