Memories of green

In the world of computer hardware, no component garners more attention than the CPU. Hardware enthusiasts wait with bated breath for the next processor to shatter the existing speed barriers. That’s hardly a surprise, either. The CPU boasts all sorts of sexy adjectives: Fast. Powerful. And let’s not forget hot. After all, it consumes the highest percentage of power among all the components.

Of course, those other parts, while perhaps not as wow-inducing, play critical roles, and they, too, add a few digits to your monthly power bill. Consider memory. According to Burton Group, while the CPU consumes 35 percent of the energy used by a server, memory burns 15 percent. That figure is high enough to warrant some attention, argues Kevin Kilbuck, senior marketing manager at Micron, given that every watt is becoming increasingly precious in the datacenter. “Server customers, and to a higher degree, datacenter operators have no idea how much memory is a contributor [to energy usage], and that is something they can control,” he says.

By “control,” he means that memory is far easier to add or swap in than, say, a more energy-efficient power supply or processor. Moreover, it’s well documented that more memory yields better application performance. In tests by VeriTest, Web server performance, both data throughput and number of HTTP requests per second, increased by 73.27 percent when upgrading the amount of RAM from 512MB to 2GB and increased more than ninefold when upgrading from 512MB to 4GB.

In an effort to sweeten the deal, as it were, for companies to invest in more memory, memory vendors are stepping up with more efficient wares. Among them is Micron, which earlier this year announced its Aspen line of power-efficient memory components. They include a 1.5V DDR2 FBDIMM, as well as 1Gb-based, DDR2 RCC (reduced chip count) modules.

The RCC modules, according to Micron, deliver the same level of performance and capacity as those built with legacy 512Mb devices, but with half the number of components. This adjustment, Micron says, reduces the memory’s overall energy consumption by 40 percent.

The other half of the Aspen equation is the 1.5V DDR2 FBDIMM, which uses a lower voltage than the 1.8V DDR2 technology commonly found in servers. That adjustment results in a 20 percent energy-consumption improvement. The move from 1.8V to 1.5V also has the happy benefit of reducing the heat that servers produce, which means lower cooling requirements.

By Micron’s calculations, its Aspen line yields total energy savings of 60 percent, which at first blush might not seem all too impressive, considering memory represents, again, around 15 percent of the power drain in a server. But therein lays the magic of economies of scale. A datacenter with 1,680 servers, each running 8GB of memory, could save $56,000 in a year. One running oh, say a million servers (the rumored number of servers in Google’s total server farm) would see savings of $33.5 million per year. The company has a calculator on its Web site for estimating how much it costs to power 1.8V modules compared to 1.5V.

DDR 2.5?

If you’re a true memory enthusiast, the 1.5V figure might look familiar: It’s the voltage of the next-generation DDR3 SDRAM. In fact, we’re already seeing DDR3 memory components for desktops, so one might wonder why Micron would bother with creating a lower-voltage DDR2 model instead of focusing on its younger brother-to-be. The reasoning, Kilbuck says, is that we won’t see server platforms supporting DDR3 — which is not backward-compatible with DDR2 — for at least a couple more years. Thus in the interim, server vendors can offer lower-power datacenter hardware from which customers can enjoy the reduced strain on their monetary and power budgets.

If there’s a drawback to Micron’s Aspen line, it’s that customers can’t necessarily swap out a 1.8V DDR2 module, pop in the 1.5V version, and enjoy the savings. The server’s voltage levels need to be adjusted to take advantage of the 0.3V difference, and many existing machines out there aren’t capable of that.

Even if voltage compatibility weren’t an issue, there’s the difficulty of IT staff going from machine to machine to swap in new memory. “There are some that can and some that can’t. Even the ones that can, there’s the question of ‘Does it make sense to throw away what I have?'” says Kilbuck. “You do get payback in terms of the electric savings. It’s like, you’re not going to get rid of your car and get a hybrid, but you might say, ‘Oh yeah, gas mileage,’ and get one the next time you buy a car.”

In other words, keep an eye open for servers to come out in the not-too-distant future running Aspen. Micron says it has server vendors lined up but, for now, won’t specify which ones.

Thanks for the memory

Yet even if your existing servers can’t take advantage of lower-voltage memory today, it’s still possible to reap performance benefits by adding more 1.8V memory to your machines. “You’re probably going to have bigger impact on performance by adding memory than any upgrade you could do,” he says. “There’s open sockets that just aren’t being used. Just plug the memory module.”

Datacenter operators certainly are searching for ways to wring more work out of their racks through techniques such as virtualization. There, memory becomes all the more important.

There’s another potential boon for Micron and any other memory manufacturers working to boost energy efficiency of their memory wares: forthcoming Energy Star standards for servers. It’s tough to predict when we’ll ever see those, or what they’ll look like, given just how diverse the world of servers is. Still, when it comes to building an Energy Star-compliant system, every watt counts.