Photo: Google
|
In June, Google and Intel announced a Climate
Savers Computing Initiative to raise the
energy efficiency of the world's computing equipment.
The goal is to reduce the amount of global carbon
dioxide emissions by 22 million metric tons per year by
cutting the amount of energy consumed by computing
equipment. According to the initiative's extrapolations,
this figure is equivalent to the annual emissions of 11
million cars.
The launch came one month shy of the new Energy
Star 4.0 standards in the United States,
which go into effect on 20 July. The new specifications
will raise the bar on which computers and related
devices can earn the Energy Star label, which denotes
the devices' superior efficiency. At the time it
launched, 40 companies had signed on with the
initiative, which sets deadlines by which companies are
required to meet certain energy consumption targets.
IEEE Spectrum spoke with Google's Bill Weihl about
the program and how participating organizations are
planning to meet its goals.
IEEE
Spectrum: What brought about the Climate
Savers Computing Initiative?
Bill Weihl:
We started talking to Intel about it a year ago, and
over time we conceived of it as an industry consortium
to try and drive higher efficiency for computers across
the industry. Not just in high-tech industry, but all
uses of computers. We've done a lot of work on our own
here on energy efficiency, especially on our own
servers. Our goal is to make that same efficiency
available to everybody, whether it's servers, laptops,
desktop workstations, and so on.
We launched in June with about 40 members who are
committed to, in the case of manufacturers, producing
high-efficiency systems and, in the case of everybody
else, to purchasing high-efficiency systems and making
extensive use of power management across their fleet of
desktops and other computers. We think that any company
that looks at the economics of it ought to conclude
fairly quickly that the measures we're talking about
will actually pay off in two to three years. For servers
it's probably closer to 12 months. So there's little to
no downside to commit to buying high-efficiency systems,
and my hope is that over the next six months virtually
every company signs up and commits to this.
IEEE
Spectrum: What are the program's targets?
Weihl: The
initiative starts in terms of efficiency targets with
the new Energy Star 4.0 standards for computers, which
require the power supply in a desktop to be at least 80
percent efficient.
To go beyond what Energy Star has done, we've set up
a staged yearly improvement in energy efficiency. In the
second year, we're then raising the bar on efficiency
for the desktop power supply to 85 percent, and then to
88 percent, and then 90 or 91 percent in each subsequent
year. So each year there will be a small and incremental
improvement in efficiency, so that by the end of the
fourth year of the initiative we'll be at 90 or 95
percent. And our intention is to set similar efficiency
targets for the motherboard for later years of the
program.
IEEE
Spectrum: How efficient are computers now?
Weihl: For a
typical desktop, the power supply is probably 65 to 70
percent efficient. Over the last 25 years, the computing
industry has been driven by price and
performance—whatever was cheapest and didn't cause the
box to overheat. There's been some attention paid to
energy efficiency—when things start to get too hot or
when the fan noise starts to get too loud.
The technology is there today to do better. The new
Energy Star standards limit the amount of power
computers can use in sleep, standby, and idle modes. So
when a computer is asleep or in standby mode, then it
can consume a single-digit number of watts. When it's
just idle, running, then it's about 65 watts, maximum.
IEEE
Spectrum: What's been Google's experience
with improving the efficiency of its own equipment so far?
Weihl: For at
least four years we've been using power supplies in our
servers that are over 90 percent efficient, and there
are a number of companies building servers with power
supplies today that are that efficient. When you want to
pack them in tightly, it becomes an issue of not just
energy consumption but also the amount of heat you
produce. We also expect to introduce targets for
motherboard efficiency, for dc-to-dc converters, and
that over the next four years they'll get to the 90 to
92 percent range. This will be a huge benefit to people
who run big server farms. Financial services, retail,
many industries today are remarkably IT intensive.
In a typical industry data center, for every watt the
server itself consumes, there's another 1 to 2 watts in
cooling and other overhead in the data center, which
means if you eliminate a watt from a server through
efficiency, you're eliminating 2 to 3 watts from a
facility. Our overhead is much, much less than that.
When you add together the components of a system, the
net efficiency you typically get from wall current to
actual components is about 50 percent. So you lose half
the energy coming out of the wall before you get close
to the computing components. And all that turns into
heat, which you then have to get rid of. So in an
air-conditioned environment, you have to spend more
energy to remove that heat. We're losing about 20
percent in the power conversion. In a typical desktop
it's more like 50 percent, and in a typical server it's
probably 30 to 40 percent.
IEEE
Spectrum: What are the price differences now
between typical and high-efficiency equipment?
Weihl: For a
desktop, we estimate it'll cost about $20 more to buy a
desktop system that is Energy Star 4.0-compliant.
That'll give you an energy savings of US $7 to $10 a
year. So it'll take two to three years to pay back,
which is about the lifespan of a system at a typical
organization. For a server it's probably about $30
additional for the increased efficiency, and it can be
paid off in about 12 months.
Many companies out there haven't specifically talked
to their vendors very much about the energy efficiency
of the servers they're buying. And in some cases we've
heard them say they'd love to buy more efficient
servers, but they just don't know where to get them. But
they're out there. We've already told our vendors that
in subsequent years we're going to raise the bar on how
efficient we want our systems.
IEEE
Spectrum: What are some of the hesitations
you've heard from other companies?
Weihl:
There's actually been remarkably little hesitation,
which is very gratifying. I've had some people I've
contacted say, “I don't pay the electricity bill,” and
“I don't even know who the guy is who pays the
electricity bill.” So those people have to meet, and the
organization has to make it relatively easy to move what
I think are small amounts of capital into the proper
budget to authorize the efficiency upgrades. Moving
small amounts of money from one part of an organization
to another is hard for some places.
There are also challenges sometimes to using power
management, depending on how a company's IT group
handles software updates. It might not be okay for the
client machines to go to sleep at night, because IT
needs to be able to log in or do something. So there are
challenges there. But I think everyone who's had a
chance to take a look at it and think it through has
said yes. And some of them just said they need a little
more time to think about this to figure out internal processes.
IEEE
Spectrum: What are some of the challenges in
making servers more efficient?
Weihl:
Servers tend to use more power than desktops. A desktop
uses 100 or 150 watts, while a server might use 150 to
300 watts. And it certainly depends on the server. There
are some that use more power than that. In addition,
servers tend to run all the time. There's less
opportunity to put them to sleep. There are some
organizations that are maxed out at their existing data
centers, not because of floor space but because of the
amount of power they can deliver or the amount of
cooling they can do. So if they can reduce the amount of
power consumed by the servers by 10, 20, 30 percent,
then they could put more servers in their facility and
possibly get more work done and possibly grow their
business faster because they don't have to build a whole
new data center. So the potential economic benefit to
many businesses in terms of higher efficiency is much,
much greater than what you would get just in terms of
consuming less energy.
But we also need people to do really simple things
that don't involve new hardware changes, like using the
power management settings built into their desktops. Put
your machine to sleep or turn it off when it's not in
use. Don't just use a screensaver, which is actually one
of the most energy-intensive things you can do on your
computer, because it's exercising the CPU, exercising
the graphics card, and it's keeping your display on. I
think most people wouldn't just leave lightbulbs burning
all the time when no one's using them, but that's what
we do with our computers, and it's kind of silly.
What we're doing with these systems today, it's a
little bit like buying a hybrid car that can shut off at
a stoplight, but we're somehow forcing the engine to
stay on. We've got the capability to have it shut itself
off when it's not doing anything, and we're not using
it. I'm sure the tools that are there could be easier to
use and so on and so on, but a lot of it is just lack of
awareness.
IEEE
Spectrum: How much do you think Google's
energy consumption will decrease?
Weihl: For
computers I would say at least 50 percent, and possibly
60 to 70 percent over, say, the next two years. People
talk a lot about huge server farms consuming all this
energy, and I think partly that's because for whatever
reason, they're sexy. When you think Google, you think
big server farms. But there are a lot more desktops and
laptops in the world than there are servers. Yearly
shipments of desktops are about 140 million. Yearly
shipments of servers are about 10 million. Now, a server
uses more energy, it's got a higher power draw, it's on
more hours of the day, and so on, so maybe desktops
overall consume only five times as much energy as
servers. Okay. Saving 60 to 70 percent of the energy
used by those several hundred million desktops today is
actually a really big deal. It's distributed across
people's homes, it's distributed around office
buildings, it's not in central facilities. But in
aggregate it's an enormous amount of energy.