Patent shows possible future direction for ATI
Dave at Beyond3D has written an intriguing commentary
on an ATI graphics patent. We've talked in the past about how future GPUs may merge or share their computational resources, erasing the distinction between pixel shaders and vertex shaders. ATI seems to be more keen on this approach than NVIDIA, at least for the near term, and this patent would seem to explain how ATI might feed such a graphics processor with a workloadusing a combination of multithreading and command queuing. These concepts should be familiar from other parts of the computing world; this just applies them to a graphics processing unit with both math and texturing capabilities.
The patent describe one embodiment in which there are two command thread queue "Reservation Stations", one for pixel commands and another of vertex commands, these are both linked to an arbiter device that can distribute the instructions from either reservation station to either an Arithmetic Logic Unit (ALU) or a texture unit
Dave says such talk seems to jibe with some of the rumors about Microsoft's Xbox 2 "Xenon" architecture, which reportedly will have as many as 48 ALUs. Large numbers of math units like these capable of processing either pixel or vertex data could alleviate bottlenecks when one type of processing, either vertex or pixel, is at a premium. If this queuing mechanism can efficiently balance both types of math with texturing work, such a chip could be very powerful indeed.