Imagination Technologies IP Path to Ray Tracing
Imagination Technologies has introduced its latest ray tracing IP, the IMG CXT for its flagship B-series GPU IP. The announcement marked the debut of Imagination’s PowerVR Photon ray tracing architecture.
Photon, said Imagination, is the industry’s most advanced ray tracing architecture, bringing desktop-quality visuals to mobile and embedded applications. The biggest news is that it has already been licensed for multiple markets.
The Photon architecture represents a decade of development by Imagination in making ray tracing viable in low-power-enabled devices. The company intends to deliver a major advance in the visual possibilities for smartphones, tablets, laptops, and automotive solutions.1
Want More Tech News? Subscribe to ComputingEdge Newsletter Today!
Imagination Technologies has been firing rays in computers since 1994 when they first introduced deferred rendering. To get it right you must do a visibility test and that’s the same as a ray tracing (but without bounces and material considerations). In terms of ray tracing as it is generally understood (physically accurate renderings), Imagination has been in the business since 2010. So, the company knows a little bit about ray tracing. The company has defined ray tracing into six levels and uses this definition to chart its path of development.
The Photon architecture inside IMG CXT sits at Level 4 of the RTLS, making it the most advanced architecture currently available It provides enhanced ray tracing performance and efficiency to deliver a desktop-quality experience for mobile gamers and developers.1
The IMG CXT-48-1536 RT3 features the Ray Acceleration Cluster (RAC), a new low-power, dedicated hardware GPU block which accelerates and offloads more of the ray-tracing computations from the shader cores compared to less-efficient Level 2 RTLS architectures. 1 Imagination says the IMG CXT RT3 offers up to 1.3 GRay/s. That, says the company, delivers photorealistic ray traced shadows, reflections, global illumination, and ambient occlusion, with high frame rates, in a mobile power budget.1
The RAC consists of the Ray Store, Ray Task Scheduler, and Coherency Gatherer and is closely coupled to two 128-wide unified shading clusters (USCs) that Imagination says, have high-speed dedicated data paths for efficient and low-power ray-traced deployment.1
The box tester unit performs the search for ray intersecting with objects in 3D space. It tests rays against axis-aligned boxes from the scene hierarchy. The RAC has a dedicated box per triangle testing hardware (dual triangle tester units). Those blocks get the RAC to a Level 2 RT solution.
Ray traversal is in hardware and uses a dedicated task scheduling and USC interface. These blocks use dedicated ray store and dedicated ray state tracking units.
The box predictive scheduler handles ray traversal, tracking, and monitoring in hardware with tight integration with the CXT USCs. Ray Store keeps ray data structures on-chip during processing which provides high-bandwidth read and write access to all units in the RAC. That, claims the company, avoids slowdowns or power increases from storing or reading ray data to DRAM.1
The Ray Task Scheduler offloads the shader clusters, deploying and tracking ray workloads with dedicated hardware, keeping ray throughput high and power consumption low.2
The packet coherency gatherer unit analyses all rays in flight and bundles rays from across the scene into coherent groups enabling them to be processed efficiently. Imagination says they have patented that technology, and it gets the RAC up to Level 4 RTLS.2
Imagination says a RAC can deliver up to 433MRay/s ray throughput per RAC, do up to 16GBoxTesVs, and all that scales with increasing RAC units, and scales with IMG multi-core. It is compliant with VulkanRT ray query and ray pipeline and represents Level 4 RTS in hardware.
IMG CXT, says Imagination, is a significant step forward for rasterized graphics performance, with 50% more compute, texturing, and geometry performance than Imagination’s previous-generation GPU IP.1 The company claims its low-power superscalar architecture delivers high-performance at low clock frequencies for exceptional FPS/W efficiency, while Imagination Image Compression (IMGIC) greatly reduces bandwidth requirements.1
Imagination says the RTL-based IP Photon architecture can be scaled to the cloud, data center, and PC markets through Imagination’s multi-core technology.2 That, claims the company, can generate up to 9 TFLOPS of FP32 rasterized performance and over 7.8 GRay/s of ray tracing performance while offering up to 2.5x greater power efficiency than today’s Level 2 or 3 RTLS ray tracing solutions.1
Bounding Volume Hierarchy (BVH) is a popular ray tracing acceleration technique that uses a tree-based “acceleration structure” that contains multiple hierarchically-arranged bounding boxes (bounding volumes) that encompass or surround different amounts of scene geometry or primitives.
Coherent ray tracing solves the problem of the integer BVH node decompression overhead by spreading the de-compression cost over many rays.
Imagination has a demonstration video that features global illumination (GI), per-pixel ray tracing, denoising, lighting, tone mapping, and TAA. RT GI adds ambient lighting grounding the objects in the scene and giving the most realistic lighting. The demo ran at 1080p between 30 and 60 fps in a mobile power budget.2
Imagination is also offering3 a ray-tracing software tool, PVRTune. It allows developers to see low-level ray tracing counters such as rays per second, box tester load, cache hit rate, and the number of recursive transverse rays per second.
The PVRRay tracing simulation Vulkan layer in the Imagination SDK. It can be used to simulate the capabilities and behavior of the CXT block.
As Imagination’s first Level 4 RTLS IP, CXT offers developers a hardware block dedicated to accelerating ray tracing that promises more power and area efficiency than other solutions. Consumers have come to appreciate the realism ray tracing provides and now will demand it in all their devices in the near future.1
The first dedicated ray tracing chip was the AR350 ASIC created by Advanced Rendering Technology (ART; Cambridge, UK) in 1997. The next dedicated hardware effort was by Caustic in 2006, but it was only implemented in FPGAs. And in 2010, Imagination bought Caustic.
Imagination Technologies pioneered hardware-accelerated ray tracing 10 years ago and has scaled it for the mobile market with its new PowerVR Photon architecture.1 Ray tracing has always been a someday technology. In 2017 based on a demo Nvidia made using four Quadro AIBs, I predicted (by extrapolating Moore’s law) that we could see real-time ray tracing by 2021 or 2022. Then in late 2018 Nvidia lit up the world and showed RTRT running off of a single AIB. That was a defining moment for the industry and combined with Nvidia’s powerful marketing consumers learned about ray tracing. The next challenge was to get RTRT to work on a mobile device. Adshir in Israel showed RTRT on a mobile device the same time Nvidia showed theirs on a PC. Adshir’s solution was all software and although elegant and highly effective, Apple, Samsung, Qualcomm, and others in the mobile market didn’t employ it—still a mystery to me. In the meantime, Adshir has been acquired. Nonetheless, consumers knew there was such a thing as ray tracing and by golly they wanted it.
Three companies in the mobile space are in a race to provide RTRT: AMD, Imagination Technologies, and SiliconArts. AMD will show up in Samsung phones. Imagination will be showing up in phones next year and one can speculate about what round-house company will announce it. SiliconArts is waiting for permission from its customer to announce its use of the technology. So by the end of 2022, we will have tangible examples of RTRT in a mobile device. The some-day tech is finally here.
1 Power Photon, Imagination Technologies, https://www.imaginationtech.com/graphics-processors/architecture/powervr-photon-architecture/
2 Bounding volume hierarchy, Wikipedia, https://en.wikipedia.org/wiki/Bounding_volume_hierarchy
3 Imagination white paper, https://www.imaginationtech.com/whitepapers/ray-tracing-levels-system/