AMD Revamps Graphics Road Map

Advanced Micro Devices is looking to revamp its graphics road map with a line of low-cost, low-watt GPUs that will have ramifications for both gamers and the enterprise market.

During the next two months, AMD will introduce a series of ATI Radeon GPUs (graphics processing units) that will cost between $200 and $300, offer up to 1 teraflop—1 trillion calculations per second—of performance and work at between 110 and 150 watts.

While AMD is looking to make an immediate impact in the gaming and PC enthusiast markets with these new graphics processors, the company also has an eye on mainstream computing as well as HPC (high-performance computing). The updated Radeon road map also includes the company's new FireStream 9250, which is designed for HPC and offers a bump in performance from the previous 500 gigaflops to 1 teraflop.

Rick Bergman, senior vice president and general manager of AMD's Graphics division, said the company wanted to get away from increasing the die size of its graphics processors and increasing the clock speed, which ultimately did not benefit users.

Instead, AMD decided to focus on a smaller silicon die. In this case the company is shifting to a 55-nanometer manufacturing process, which means the die measures only 16 millimeters square. This die uses less power, gives a performance boost and allows the company to squeeze more processors onto a single wafer, which helps reduce costs. Most other GPUs are made on a 65-nm process.

The strategy will also affect how AMD creates graphics for the enterprise, whether it's GPUs for workstations or more mainstream graphics that can exploit all the capabilities of Microsoft's Windows Vista OS.

"There isn't an enterprise customer that has a huge, desk-side box anymore," Bergman said. "It's important for those enterprise customers, even the performance users, to have a solution that fits into their needs. In the workstation market, they are looking for [a GPU] that works at 150 watts or below, which lets them then use a standard power supply and standard PCI [Express] slot."

AMD plans to release the first of these GPUs, the ATI Radeon HD 4850, on June 25, followed by the Radeon HD 4870 on July 8. After that, the company is planning to release a third product, right now code-named R700, which will offer two GPUs on the same board. These chips will be built on 55-nm manufacturing.

In addition to working within 110 to 150 watts, these processors will support Microsoft's DX10.1 graphics API and support GDDR5 (graphics double data rate 5). AMD is also embracing CL, an open programming language that will allow developers and ISVs to program a GPU like a CPU in order to develop applications.

While AMD's historical rival has been Intel on the CPU side, Nvidia has become an increasingly tough competitor on the graphics side. Not only does the rivalry between AMD and Nvidia involve PCs, but HPC as well.

On June 16, Nvidia announced its Tesla 10 series GPU, which offers 240 graphics processing cores and will compete with the FireStream 9250. Nvidia is also developing its own programming language called CUDA (Compute Unified Device Architecture), which rivals OpenCL.