I’m curious to see how many people care about this issue. I see a lot of PC components out there that aren’t very efficient and also that with a careful choice of components that it’s possible to build a system that is powerful enough for the majority and is still very frugal in power terms.
I’m about do some research in this area for my own enlightenment and I’m wondering what others on this forum think about this issue.

As an example, I find myself becoming less able to justify a performance gain of ~10% at most at full load when over clocking & over volting a Core 2 Duo from 3 GHz to 3.3 GHz, which requires disabling Speedstep and means that the increase in system power consumption is roughly 30% just whilst idling.
Please don’t take this as a criticism if you do this, I simply want to get an idea of how many people here consider this issue.

For me it's not so much power efficiency per se as its implications. Components that let me do more while putting out less heat are good. I'd rather not shell out $50 for a Ninja if I can help it, especially seeing as I like my CPUs around $70 anyway

For instance: for a long time, I ran my XP-M at 166*9 (stock is 133*14). I did not need more power, until I got a recording of a concert that skipped annoyingly, while maxing out the CPU. This prompted me to bump the CPU to 166*12 (woo... overclock). However, upon finding out that the CPU is still getting maxed and I'm still getting skipping, I'll likely bring it back down the next time I reboot. I don't need 2 GHz.

TDPs will definitely have an impact when I do finally upgrade, but stuff that I would consider worthwhile upgrading to (dual core, dual-DVI) hasn't gotten cheap enough yet. Maybe when the X2 3600+ becomes available.

For me, the importance of power efficient components increases with the amount of hours the PC is expected to stay turned on. My 24x7 server has more efficient components and aggressive underclocking than my workstation.

Lately I've been experimenting more with S3 standby and have been very impressed with the power savings and quick wake up times. I am able to have the machine drop to standby (consuming 3W) and wake up within seconds, usually quick enough to process a roaming profile or file share request before it times out. I think the use of standby is much cheaper than trying to spec efficient desktop components that bring idle consumption down below ~50w.

Very important. Power efficiency was actually a big factor in my most recent computer purchase, a 12" notebook.

I’ve found S3 standby to invaluable and use it instead of powering the system off in the daytime as it only consumes ~1W more. I work from home so I’m often manually putting the system into standby if I know I’m not using the PC for at least 10 minutes or so. It’s also configured to automatically enter standby after 20 minutes as well, with the monitor and disks turning off after 5 minutes.

I think it has been measuring how efficient my Core Duo laptop is that has focused my mind on how profligate my desktop is in comparison. Over the next few months I’m going to look into how much performance I can get using 80 to 120W laptop style power bricks in conjunction with a DC/DC converter boards. I’m going to use Yonah/Merom/Conroe CPUs with VGA from IGPs up to a 7900GS.

Because efficiency is so important for me, I built another PC seperate from my gaming PC to use on a daily basis. I now rarely use my gaming PC because I just don't have any time for it anymore. I cannot use standby mode because i use my spare cycles to fold. All of my 24/7 computers are undervolted while my server is also underclocked. What I would really like is a way to fold while using cool and quiet so when the server needs more power for transcoding it can pop out of cool and quiet, do the work, then return to CnQ and still continue to fold at 1Ghz and 1.1V.

Not many choices, are there? I do consider it, but I usually settle for the middle if anything. I think chasing down 35W AMD X2 is sort of stupid when conroe performs better even if it consumes more watts/hour. Same with videocard. If I was really energy concerned I would have lived with ATI X800/nv6600, but I have 6800gt (if I were to buy videocard now I would have went with 7900gt).

I guess my approach is, set performance requirements, say such CPU frequency, and so many FPS in HL2 for videocard, and go for the hardware that satisfies these performance requirements while consuming as little power/dissapating as little heat as possible.


BTW, to the OP, when you look at energy inefficient rigs, you have to consider, that not everybody jumps on latest and greatest. I still have P4c in rig, and I'm not going to change it for another year because it does everything I need to and changing it would take too much money. Upgraging right now for the sake of gaining extra 20-30 watt doesn't make sense.

Totally agree.

So nobody has yet said that they don’t consider power efficiency at all. Is this because people who feel that way have automatically ignored this topic, or that they don’t like to say?

It would be nice to be able to restrict a task to a certain percentage of CPU cycles. Some applications will allow you to limit the CPU cycles that they consume, does Folding support this? If you have a dual core CPU, you could limit the max CPU clock using RMClock and then assign Folding to one of the cores and set its Priority to low. This way the server could use one or both cores when needed, depending on if it’s multi-threaded and when it’s finished its work Folding would grab back a single core to do its thing. Not exactly what you’re after but a compromise of sorts.
With AMD’s quad cores offering individual clock speeds per core, it’s possible that CnQ for these CPUs could even allow what you suggest. Let’s hope the dual cores also support this individual core clocking, which seems likely!

Well I’m open to other ideas, what do you have in mind?

Generally speaking I’d agree with this. For those people prepared to buy and sell on eBay, the cost of upgrading doesn’t have to be that large; it’ll vary a lot depending on what you’re selling and what you’re buying of course.
I’m about to upgrade two machines for family members and when I looked at the cost difference between Yonah/Merom v Conroe v AM2 using my particular purchasing methodology, the Yonah/Merom route stood out as the obvious 1st choice to me. This decision also takes into account how the PCs are used as well of course.
The biggest stumbling block for me at the moment is finding an inexpensive highly efficient power supply in the PicoPSU mould. I’ve found quite a few options ranging from 60W up to over 200W that use a fanless power brick and DC/DC board. I need to determine what wattages I need for various systems and then bite the bullet and buy a number of them and see how they compare.
They mainly seem to be using 19v power bricks rather than the 12V of the PicoPSU; I’m wondering if the PicoPSU will be significantly more efficient because it doesn’t need to convert 19V to 12V. Any ideas?

You're asking a self-selecting group; the first thing SPCR teaches is that power efficiency is directly related to noise; someone who doesn't care about power efficiency basically hasn't absorbed anything SPCR has said over the past 4 years.



Well, given that the PicoPSU is line voltage->12V, it relies on the efficiency of the brick; whereas for 19V, it goes line voltage->19V->12V; from first principles, you would say the process with more steps would be less efficient; as to whether the difference is significant, it's hard to say.

Good point. There are still people on the forums using Prescott and Pentium D chips, so there are other factors involved in the decision making process. Maybe people who buy the inefficient Intel chips do so because they feel more comfortable with the Intel platform!

Well there’s only one way to find out, to actually do a direct comparison. Although, it might be hard to determine whether the efficiency is more down to the power brick or the DC/DC converter.

It would be so much easier if there was an ATX specification that covered this area which defined a standard connector for the power brick and a couple of voltage levels; maybe 12V and 19V! Then manufacturers could develop components that fit this standard which would lead to lower costs, higher availability and the option to choose the most efficient DC/DC boards and power bricks separately. A standard back-plate that allowed one of these power supplies to connect via the ATX power supply slot in an ATX case would be useful also. I know, I don’t ask for much.