Evaluation of high performance multicache parallel texture mapping
ICS '98 Proceedings of the 12th international conference on Supercomputing
ZR: a 3D API transparent technology for chunk rendering
Proceedings of the 34th annual ACM/IEEE international symposium on Microarchitecture
| Hi-index | 0.00 |
A revolutionary new approach from an alliance of major industry partners is developing the most dramatic advance in multimedia performance the PC has ever seen. Code-named "Talisman," the new architecture, being jointly developed by Microsoft, Silicon Engineering, Samsung Semiconductor, Philips Microelectronics, Cirrus Logic, and Fujitsu Microelectronics, combines advanced DSP technology with a new approach to 3D graphics to provide vast improvements in performance and image quality. The hardware architecture is designed to dove-tail with Microsoft's DirectX/Direct3D API, making possible sophisticated techniques to accelerate multimedia performance to new levels while maintaining compatibility with this industry-standard API. The first implementation of this new hardware architecture is known as "Escalante," and was described at HotChips8 as "Touchstone". This article focuses on the 3D aspects of the Talisman architecture, specifically discussing the Escalante implementation. The Talisman architecture achieves performance levels significantly higher than can be attained using traditional approaches by exploiting the innate spatial and temporal coherence characteristics of animated 3D graphics. Rather than completely re-rendering each scene every frame, image layers--surfaces or elements of the total scene--are rendered and then transformed and composited during display to form the final image. Surfaces are re-used over several frames before needing to be re-rendered. The resultant offload from the rendering processor allows it to produce images of vastly higher quality, using shadowing and multipass operations in real-time. An image compositor assembles these elements at video rate, performing scaling, translations, rotation and skew to each element independently, allowing reuse of the elements over several frames, and resulting in full frame-rate (75Hz) full-screen (1024 脳 768 or higher) animation at full true-color pixel depth (24-bits). Extensive use is made of compression and decompression to reduce memory usage and bandwidth, allowing such a system to be built with only 4 Mbytes of memory. The concept of "chunking"--dividing the geometries into small areas--provides many benefits, including on-chip Z-buffering, cost-effective anti-aliasing, and block-oriented compression. Advanced rendering features such as anisotropic MIP-mapped textures, shadows, transparency, motion blur, lens flare, and more, are implemented in hardware.