| Last Modified On : | February 2, 2009 3:10 PM PST |
Rate |
|
| Michael Abrash is a programmer at RAD Game Tools, working with Intel on the Larrabee project. He is co-author of the Pixomatic software renderer. He worked on Quake at id Software and on Windows NT and Xbox at Microsoft. Michael is the author of several books on graphics and performance programming, including Zen of Assembly Language and Michael Abrash's Graphics Programming Black Book, and has written columns on graphics and performance programming for Dr. Dobb's Journal and Programmer's Journal. |

English | 中文 | Русский | Français
Gina Bovara (Intel)
|
Eric Parker
1) Larrabee will be evaluated in the marketplace based on its performance relative to GPUs for graphics (e.g. Quake 3 arena timedemo). It must always tie or beat the GPUs - or die.
2) rasterization will take away cycles from shaders, physics or other Larrabee tasks.
Based on the overriding importance of 1), and the desire to realize exciting new capabilities for 2), I think serious consideration should be made to dedicating hardware to rasterization. Rasterization is going to require a lot of resources nearly all the time. Even if you can match the GPUs, the question is at what cost? If there are no resources left for shaders or other applications then what new value is Larrabee delivering?