AMD rolls out a next-generation graphics architecture that is ahead its competitors. AMD's HD 2900 uses more advanced process technology, has unique hardware capabilities and in one configuration supports a faster memory bus than the Nvidia part.

Since last year, Nvidia has enjoyed an unusually long time-to-market advantage over its archrival ATI Technologies, now the graphics division of AMD. Analysts are holding off calling a winner in the latest generation until they can benchmark both chips, but they say it's clear AMD is at least standing toe-to-toe with its competitor again.

The AMD HD 2900 will come in versions made in 80nm and 65nm CMOS and offer a pool of as many as 320 arithmetic logic units. By comparison, the Nvidia devices are made in a 90nm process and offer an array of 128 ALUs.

In addition, a high end version of the new AMD architecture will sport a 512-bit wide memory interface to GDDR4 graphics memory. In its latest generation, Nvidia is only linking to GDDR3 memory with interfaces no wider than 400 bits.

Internally, AMD distinguishes its memory interface with a 512-bit memory controller laid out as a ring bus around the periphery of the chip. AMD says the distributed controller will be highly scalable because it requires no central hub. It supports up to eight 64-bit memory channels in the new generation.

The HD 2900 also includes a specific hardware acceleration pipeline for tessellation. The feature enhances the chip's ability to support finely detailed features and shadows.

By: DocMemory
Copyright © 2023 CST, Inc. All Rights Reserved