New Micron Product Will Improve Server Performance, Memory Capacity

Micron is gearing up its LRDIMM technology, which officials say will give data center servers a 57 percent increase in system bandwidth and three times the memory capacity of current servers using RDIMMs. Driving this need for more memory capacity is the growing use of multicore processors and such technologies as virtualization, and will become even more important as cloud computing adoption grows. Micron is planning to start mass production of its 16GB LRDIMMs in 2010.

Micron Technology is developing technology designed to increase server memory capacity and improve performance.

Micron officials July 30 announced the company is producing DDR3 LRDIMMs (load-reduced, dual-inline memory modules) that they say will give servers using the technology up to 57 percent better bandwidth and three times the memory capacity of current servers.

And that will grow, they said, as mainstream DDR3 production migrates from 2-gigabit components to 4Gb.

All that is important as enterprises, through their use of multicore processors and server virtualization technologies, drive up the demand for memory, according to Robert Feurle, vice president of DRAM marketing at Micron.

"With the rise in virtualization, our new ... modules allow customers to easily expand their memory capacity," Feurle said in a statement. "While traditions RDIMMs [registered DIMMs] limit the amount of memory that can be accommodated due to their loading profile, LRDIMMs eliminate that problem by reducing the module load."

Micron in the fall will begin sampling the 16GB LRDIMMs, which are built using the company's 1.35-volt, 2Gb 50-nanometer DDR3 memory chips. The high density and small die size enable Micron to increase server module capacity, officials said.

They already are sampling 8GB LRDIMMs.

The 2Gb 50nm DDR3 chip is in the qualification phase with customers and is ramping toward high-volume production, they said.

A key to the LRDIMMs is that they use Inphi's iMB (isolation memory buffer) chip rather than a register, which reduces the bus load when transferring data between the processor and memory. The LRDIMMs reduce the bus load by half for dual-rank modules and 75 percent for quad-rank modules when compared with current RDIMMs.

Reducing the load bus means that servers can handle higher frequencies of data, which improves performance, and support more modules, which increases memory capacity, officials said.

"If you're in the server industry, you know that load-reduced DIMMs are going to deliver some much-needed performance and bandwidth boosts for next-generation servers," Micron officials said in a July 30 blog post.

Using RDIMMs, a typical server currently can accommodate up to three quad-rank 16GB RDIMMs per processor, officials said. Using LRDIMMs, that same system can support up to nine quad-rank 16GB LRDIMMs, which increases memory capacity from 48GB to 144GB.

"Adoption of this approach to memory technology will further enable server virtualization and cloud computing," Paul Washkewicz, vice president of marketing at Inphi, said in a statement. "This technology delivers the much-needed higher bandwidth and memory capacity demanded by data center servers."

Mass production of 16GB LRDIMMs is scheduled to begin next year, according to Micron.