AMD to double memory capacity in server applications

Posted by Wolfgang Gruener

Sunnyvale (CA) – AMD has come up with a new technology that will allow the company to install 16 memory modules per CPU socket – instead of only 8 that are used with Opteron CPUs today.

The new technology, called G3MX, has been designed to address increasing memory needs especially in high-performance, multi-socket servers that are running memory intensive applications such as databases and virtualization software. AMD representatives told TG Daily that the company expects G3MX to appear in 4P+ systems, but they “could imagine” the technology to be used in 2P servers and workstations as well.

G3MX will not be immediately available, as it is developed to work with DDR3 memory and the processor generation following Barcelona. This next-next generation processor is not expected to be introduced before 2009. The technology to extend the server memory footprint is currently being developed in collaboration with IDT and Inphi, who are planning to sell G3MX components, AMD said.

Intel currently has a clear advantage when it comes to pure memory capacity in server systems: While AMD is limited to 8 modules per CPU, Intel will support up to 32 modules with its upcoming Tigerton Xeon 7300 processor – however, Intel uses FB-DIMMs, which use more power than DDR3 devices and Intel still has a latency disadvantage when compared to the Opteron processor.

We will have to wait until 2009 to find out whether G3MX will translate into a competitive advantage for AMD or not.