A new startup is looking to take the bite out of the current memory crunch.
MetaRAM, which started up about two years ago and officially announced itself to the world Feb. 25, is releasing a new product, MetaSDRAM, which looks to add memory capacity to workstations and servers without significantly adding to the overall cost of a system.
The San Jose, Calif., company is the brainchild of Fred Weber, the former chief technology officer of Advanced Micro Devices who played a key role in developing that company’s Opteron processor technology, and Suresh Rajan, a former Nvidia executive.
What the company has developed is a new type of chip set that server vendors can use in conjunction with a system’s DIMMs (dual in-line memory modules) that will allow them to double or quadruple the capacity of a standard system without radically changing the makeup of the infrastructure or system motherboard.
For example, a pair of two-socket servers based on quad-core microprocessors for Advanced Micro Devices can now offer about 64GB of memory each. With the chip sets that MetaRAM has developed, a vendor can manufacture a single system with 128GB capacity, while increasing the utilization of the CPUs.
On a much larger scale, the startup’s executives believe that their technology can scale the memory capacity of an eight-socket system anywhere from 512GB up to 1TB.
The MetaRAM technology also looks to close the gap between processor compute power, which according to Moore’s Law doubles about every 18 months, and DRAM (dynamic RAM) capacity, which doubles about every 36 months. This gap requires vendors to add expensive DRAM technology to each DIMM, which drives up the price of the system.
Although Weber worked at AMD, the MetRAM technology will work with both AMD and Intel processors, and Intel Capital has made an investment in the company, as has several other venture capital firms.
The Problem with Memory Systems
“What we found was that the memory system in both workstations and servers were becoming a big bottleneck in terms of system performance,” Rajan told eWEEK.
“CPUs had continued to scale over the last 10 years, but what we both saw was that the memory system had not kept pace with the CPU. Therefore, all the improvements that AMD and Intel and Sun [Microsystems] and IBM had done to the CPU were not being fully utilized by the end-user because the memory subsystem had not kept pace.”
These type of multicore processor advances, Rajan said, are important as workloads on servers increase. However, to keep pace, vendors are looking for ways to increase the amount of memory without adding huge costs to the base price of the system. He said that as the use of virtualization and “in-memory” databases increase, the burden to add more memory into each system will grow and prices will skyrocket.
That is where the MetaSDRAM products come in.
The technology behind the MetaSDRAM chip set works by making multiple DRAM memory chips look like one large DRAM chip to the system’s memory controller. Instead of the memory chips communicating directly to the memory controller, that communication is now routed through the chip set This setup eliminates some of the limitations associated with system memory.
Instead of four separate 1GB DRAMs, the system believes it is communicating with a single 4GB DRAM.
Right now, a four-way system with 256GB of memory costs about $500,000, Rajan said. With the MetaSDRAM chip set eliminating the need for additional DIMMs with more expensive DRAM chips, that same system will cost about $50,000.
Although none of the top-tier OEMs are offering systems with the MetaRAM technology, several smaller vendors such as Appro, Verari Systems, Rackable Systems and Colfax International, will use the new technology.