AMD has announced that it plans to release a line of 12-core server processors within two years.

The company unveiled its server and workstation roadmap at a media event on 7 May in San Francisco, which also included plans for a line of 45nm chips by the end of the year.

The 45nm Shanghai chips will use immersion lithography techniques to produce a line of quad-core processors which the company claims will be up to 20 per cent more power efficient than its current 65nm Barcelona chips.

Shanghai will be completely socket compatible with Barcelona, allowing manufacturers to drop the new chips into existing platform designs.

AMD plans to expand its 45nm line-up to a six-core model dubbed Istanbul by the second half of 2009. Both chips will feature 6MB of level-3 cache and use Nvidia nForce chipsets.

Randy Allen, corporate vice president and server and workstation general manager at AMD, said that the next major upgrade will come in 2010 when the company hopes to launch a line of server chips running on its new Maranello platform.

The two 45nm chips, codenamed Sao Paulo and Magny-Cours, will sport six and 12 processing cores respectively.

The Sao Paulo chips will use 6MB of level-3 cache on a single die, while Magny-Cours will join two six-core dies together to create a 12-core chip.

Both chips will sport AMD chipsets and will support hyper-transport 3.0 connections as well as DDR-3 memory.

Allen told reporters that he expects the six-core chips to become a mainstay for many users. Whereas companies have been more than ready for dual and quad-core chips, Allen sees six cores as a barrier of sorts.

"The vast majority of the market is scaleable to at least six-core [chips]," he said. "We believe that there are going to be more workloads that stop at six cores."

Allen still predicts an eager market for the 12-core Magny-Cours chip. "Some of our largest markets are people running installations, and they are asking when they are going to get to 24 cores," he said. "Their workloads are embarrassingly parallel."

One possible solution for these highly parallel workloads has been the use of graphics chips for general processing, a concept known as GPGPU.

Allen explained that, while the company has looked into "leveraging" graphics processors, it is not something AMD is currently looking to address with its server processors.

"If you look at the initiative of accelerated computing, the concept is so simple and everyone is enamoured with it," he told reporters.

"But for different problems there are different solutions, and for some problems there are better solutions than an x86 processor."