Power consumption ratings of the MediaSmart Server

by Alex Kuretz on March 13, 2009 · 13 comments

in Guides

In my initial review of the EX485 and EX487 MediaSmart Servers, I noticed that the published power consumption statistics did not match the measurements I observed. For my testing process I was using a P3 Kill-A-Watt to measure the power consumption, and fully loading the system using the free utility Prime95 to stress the electrical system.

Here are the Power Consumption measurements reported in the EX48x Product Information, and the EX47x Product Information pages on the HP.com website.

Power Consumption for the EX485 and EX487

Power Consumption for the EX485 and EX487

Power Consumption for the EX470 and EX475

Power Consumption for the EX470 and EX475

For a refresher, here is a recap of the observations from my review.

HP also claims to have improved the power consumption of the server while it is on, with the base EX487 consuming 60 watts under full load with two drives in the system versus the EX475 which consumed 73 watts under full load with two drives in the system. Unfortunately my testing with the Kill-A-Watt P3 and using Prime95 to put the system under a high CPU load showed the EX487 consuming 72 watts, so I’ll be following up with HP on this to try to get an explanation for the discrepancy I observed.

I did follow up with HP, and received the following explanation for the different measurements I observed. HP informed me that they used a different tool for loading the system that instead of fully maxing out the system (which Prime95 does), the HP tool “puts a typical load on it of what they would expect users to do”.

I understand and appreciate that perspective, my only concern is that when comparing various devices it is important to be consistent in the way the measurements are taken. So for a final power consumption comparison, I have performed the same tests on the EX487 server as I did on the EX475 (a note: during my time at HP I created the measurements published for the EX47x servers using the same testing methodology and tools that I use here). Here are the combined results of my testing and the HP published results.

Idle Power Consumption
Number of Drives EX47x Idle HP EX48x Idle MSS.net EX48x Idle
1 52 44 42
2 60 55 52
3 68 65 62
4 76 76 71
Active Power Consumption
Number of Drives EX47x Active HP EX48x Active MSS.net EX48x Active
1 60 50 63
2 73 60 72
3 86 70 81
4 99 81 89

These results show us that using the same testing methodology for both servers produces closer results than would be expected based upon the published product specifications. However the EX48x servers still show a significant reduction in power consumption over the previous generation, and once you consider the new power saving feature that allows you to consume only 1 watt by putting the EX48x to sleep, you will definitely save a little bit of money in energy costs by purchasing the new generation of MediaSmart Server.

Thank you to HP for providing insight into the testing process used on the EX48x servers.





Article by

I'm Alex Kuretz, and I'm the founder of MediaSmartServer.net. I was the Lead Test and Integration Engineer at HP for the MediaSmart Server until April 2008 when I moved on to other opportunities outside HP. I've kept active in the Windows Home Server community, creating several add-ins and helping users make the most of their Home Servers.


{ 13 comments }

Lars Hoffmann March 13, 2009 at 9:03 am

Hi Alex,

Thanks for following up on the power consumption. However, I have to politely partially disagree with you. Your say that you have conducted your tests using the same software, but you have forgotten that you did not complete the test under the same thermal prerequisites (which are extremely important when calculating Power consumption since molecule movement is strongly correlated to temperature). This is clearly confirmed by looking at your tables below. You claimed in an earlier post that you have not modified the thermal setup of your EX47x but you did it for the EX48x. A post on another blog revealed that especially cleaning the Northbridge leads to a significantly improved thermal footprint (temperature drop >10 degrees C). I believe that explains why your consumption is lower overall idle and under load. It would be interesting to see how the EX47x numbers would change once you clean the CPU and Northbrige there and install a 2GB RAM module. If thermal physics rules apply (and I do not see why they should not), you should see an improvement in your numbers here as well.
Now I do agree that it looks to me that adding a drive to the EX47x consumes more power than adding the same to the EX48x. But looking at idle versus active, the EX48x fares actually worse from your tables, which I again believe (an have stated in some unfortunaltely unanswered posts) is related to the higher thermal footprint of the Intel chipset verus the AMD chipset.
In conclusion, I believe once you repeat the test for a thermally modified EX47x, the savings will be minimal. And if you use one of the available power saving add-ins (Power Switch or Lights Out)you have the same opportunity of putting the EX47x to sleep and it would be interesting to know what the power consumption is then.
Cheers,
Lars

Alex Kuretz March 13, 2009 at 5:59 pm

Hi Lars,

Thanks for your input, and you do make some valid points. I believe the test is fair and accurate, for a few reasons.

- You are correct, I did apply new thermal paste to both CPU and Northbridge on my EX487, however the measurement for the two-drive active load (72 watts) measured the same both in my initial review in December (prior to the thermal alteration of the server) and in the testing I just performed.

- I have doubt that improving the thermal performance of the system will change the power consumption by more than a watt or two. I also did not see a 10c thermal improvement, more like 1-2c improvement from the application of thermal paste.

- I intentionally compared the two servers with their shipping configuration (except for the already discussed minor thermal change of adding new Arctic Silver thermal compound), to most closely represent what the average user will be buying.

Note that the EX47x numbers were generated in my lab at HP over 1.5 years ago, and I don’t have direct visibility into the EX48x numbers other than what I shared in the article above.

I recall your questions about thermal footprint, and such discussions are in part what prompted me to follow up with HP to get more info. I also should have access to a new EX487 server in the very near future and so will test that system in the same fashion to see if there is any discrepancy in power consumption measurements as compared to my current slightly modified server.

Thanks for your comments!
Alex

Stephen Foskett March 14, 2009 at 10:02 am

Alex,

I’m surprised that they use that much power. Although it’s way down from the killer servers of a few years ago, not to mention the average tower desktop, the MediaSmarts are tiny! What’s using all of that juice? They don’t have a 3D video card or killer CPU, and the drives are fairly efficient these days… I’m using a Mac Mini as a home server myself and it’s less than 20 watts at idle!

Stephen

JohnBick March 14, 2009 at 2:30 pm

Interesting that the INCREASE in power consumption as drives are added is less for the EX48x than for the EX47x. I would have expected it to be the same if the drives are the same. Any thoughts as to what’s going on?

Chris P. March 15, 2009 at 2:50 pm

I’m going to politely disagree with Lars here…

Improving the thermal efficiency of the heatsinks attached to the CPU or bridge should not effect power consumption at all unless the fan activity in the server is significantly different. The only effect of lower thermal resistance from chip to ambient air is that the power consumed by these devices and wasted as heat due will be more efficiently transferred away from the chip.

Alex Kuretz March 16, 2009 at 10:31 am

Hi Stephen,

It was great meeting you at the summit, thanks for stopping by! A problem with WHS is that it does a pass every hour to check if any files need to be migrated or duplicated, this means that the drives are spinning most of the time and so HP does not implement a disk drive spin-down. The CPU does throttle for load using the Cool-n-Quiet driver for the AMD platform and a similar utility for the Intel platform which saves some power. However they are still 35watt parts for the CPU, plus chipset, memory, network, etc.

I think the biggest savings gain would be made from a change to the WHS Drive Extender functionality to allow it to spin down the drives for long periods of time.

I hope that makes sense,
Alex

Alex Kuretz March 16, 2009 at 10:37 am

John, good point. I did use different drives in my tests of the EX48x than I did in the test lab at HP, plus the Kill-A-Watt tends to fluctuate by a couple watts for most measurements, so I would attribute the difference of a watt or two in either direction as just variation in the measurements. It is interesting that we see 8 watts per drive for the EX47x server and closer to 10 watt per drive for the EX48x server.

Chris, thanks for chiming in, I’m no electrical engineer but as I mentioned before I have a hard time imagining that some thermal paste on a northbridge heatsink would change the power consumption by meaningful amount.

Alex

Ed Tittel March 17, 2009 at 1:17 am

Dear Alex:

I measured power consumption during my testing for the recent Tom’s article but neglected to include that information. I think that the incremental increase in power consumed will vary to some degree according to what kind/size of drives you add to the units. I observed the following measurements:

1. 6 W per 500 GB Seagate 7200.11 drive
2. 7 W per 750 GB Seagate 7200.11 drive
3. 7 W per 1 TB Samsung SpinPoint drive

I also observed lower overall power consumption overall for the lower-powered Sempron 3400+ and the LE-1640, which went up by 5-10 W at idle for the LE-1660 and the BE-2350 (despite nominal TDP being identical for the various LE and BE parts).

Never having had my hot little hands on the EX48* models, I can’t add to that part of the discussion, but your numbers are very much in line with what I observed using a Seasonic Power Angel on my EX475.

Thanks for sharing this info with us. FWIW, I agree that an “idle-spindown-after-extended-timeout” would make a very nice enhancement to all MediaSmart models.

–Ed–

Colin April 4, 2009 at 10:33 pm

What surprises me is that the idle power consumption is little different than the active. You’d hope it would be much lower (Mac Mini-esque, even).

Alex Kuretz April 5, 2009 at 12:38 pm

This is primarily due to the hard drives not spinning down. The power difference between active and idle is mostly due to the processor stepping down to a lower speed and voltage when not under load.

Colin April 5, 2009 at 5:26 pm

Interesting.

Out of curiosity, have you tried a similar test with WD GreenPower drives to see how much of a difference they make?

Alex Kuretz April 5, 2009 at 7:55 pm

I have not tried a similar test, as I only recently acquired a GreenPower drive and it is in use in my primary home system. However, I would only expect the difference to be a couple of watts per drive at best.

Colin April 5, 2009 at 8:41 pm

I’d actually expect up to 8 watts idle, 10 watts at load per drive according to the benchmarks I’ve seen, though I guess it would depend on the drive.
http://www.tomshardware.com/reviews/wd-caviar-gp,1703-8.html

While we’re on the topic, I assume the stock PSU on the MediaSmart isn’t 80+ certified, is it? It’s just a 1U form factor, right?

My gears are turning now. I’d love to see how a GP/80+ PSU would affect the numbers.

Heck with the processor and extra RAM. Those are the mods I’M interested in!

Comments are closed, visit the forums to continue the discussion.

Previous post:

Next post: