Intel’s Dual-Core Xeon First Look
by Jason Clark & Ross Whitehead on December 16, 2005 12:05 AM EST- Posted in
- IT Computing
How does power consumption affect the bottom line?
We've illustrated each system's power consumption levels at different load levels. For the technical folks, those numbers are important. But, for the platform decision maker, the CTO, Director of Technology and any person in a role that looks at the bigger picture, it's important to illustrate how it affects the bottom line. Obviously, if you use more power, it costs you more money. To illustrate this, we used a current rate of 14 cents per kWh, which was taken from a power bill of a resident in Connecticut, New York. We then used the data to extrapolate the cost of each platform at the various load levels.
At 40-60% load, it costs $42 a month to run a Bensley system while an Opteron system would cost $25. Factor that over a year, and a Bensley system would cost $504 to run and an Opteron would cost $300. Now, if that doesn't pique your interest, let's say that you have 40 servers at a datacenter, with the same power characteristics. Over a year, it would cost you $8,160 more to run the Bensley system.
So, you've read through the article and are waiting for us to tell you what platform is better. That entirely depends on what matters to you: performance, power, or both. If all you care about is performance, then Bensley is that platform. If you care about how much power you consume, then Opteron is that platform. Now, if you want the best performance per Watt, then Opteron is that platform. At least that is what the database test results that we've shown here dictate. Obviously, given Intel's roadmap, they are developing platforms to address performance per Watt. Woodcrest will be the first product that is focused on performance per Watt, which we will see in the second half of 2006. Until then, the choice is yours.
We've illustrated each system's power consumption levels at different load levels. For the technical folks, those numbers are important. But, for the platform decision maker, the CTO, Director of Technology and any person in a role that looks at the bigger picture, it's important to illustrate how it affects the bottom line. Obviously, if you use more power, it costs you more money. To illustrate this, we used a current rate of 14 cents per kWh, which was taken from a power bill of a resident in Connecticut, New York. We then used the data to extrapolate the cost of each platform at the various load levels.
At 40-60% load, it costs $42 a month to run a Bensley system while an Opteron system would cost $25. Factor that over a year, and a Bensley system would cost $504 to run and an Opteron would cost $300. Now, if that doesn't pique your interest, let's say that you have 40 servers at a datacenter, with the same power characteristics. Over a year, it would cost you $8,160 more to run the Bensley system.
So, you've read through the article and are waiting for us to tell you what platform is better. That entirely depends on what matters to you: performance, power, or both. If all you care about is performance, then Bensley is that platform. If you care about how much power you consume, then Opteron is that platform. Now, if you want the best performance per Watt, then Opteron is that platform. At least that is what the database test results that we've shown here dictate. Obviously, given Intel's roadmap, they are developing platforms to address performance per Watt. Woodcrest will be the first product that is focused on performance per Watt, which we will see in the second half of 2006. Until then, the choice is yours.
Facebook
Twitter
RSS
67 Comments
View All Comments
coldpower27 - Friday, December 16, 2005 - link
And I am interested how did you get a difference of $8140 to begin with.Furen - Friday, December 16, 2005 - link
I took a lot of shortcuts just to get a rough approximation but here it goes:406W - 260W = 166W
166W * 24H/day = 3984WH/day
3984WH/day * 365 days = 1,454,160WH/year = 1,454kWH/year
1,454kWH/year * $.14/kWH (which is overpriced, by the way, since consumers normally pay more than businesses) = $203.58/year
$203.58/year * 40 systems = $8143.30/year for 40 systems.
Viditor - Friday, December 16, 2005 - link
Good point...forgot about the power conversion and power supply loss.Furen - Friday, December 16, 2005 - link
I didn't mean PSU power loss, but rather that many data centers convert the AC input to DC at the distribution centers and the convert it to AC again just before sending it to the server, since servers are not built to operate on DC.PSU loss is already reflected in anadtech's measurements, since power consumption is measured at the plug.
Furen - Friday, December 16, 2005 - link
man, I would kill for an edit function on comments...Viditor - Friday, December 16, 2005 - link
You're forgetting the cost of cooling (which is much higher than just the CPU...)
coldpower27 - Friday, December 16, 2005 - link
They are measuring total power draw of the 2 systems which includes the energy used by the cooling system. I am not forgeting anything. I am only interested in cost of electricity used by the 2 systems.
Anandtech isn't incorporating cost of cooling into it's numbers either.
Jason Clark - Friday, December 16, 2005 - link
If we had complete systems from both vendors, this would have been possible. Unfortunately, we had a pre-production validation platform and a motherboard and two cpus from amd :)... So, what we did was make the bensley system as close to the open air opteron system as we could. I agree, we need to get some complete systems with their cooling mechanisms in place, and we'll work on the vendors next year for that.Viditor - Friday, December 16, 2005 - link
The cooling systems I refer to are the airconditioning, not the HSF or the case fans...By doubling the heat output, you are also doubling the air con requirements.
coldpower27 - Friday, December 16, 2005 - link
Which would be offset by the heating provided in the Winter time.