I unplug my main desktop and notebook from outlets whenever I leave home. Can't risk having a surge/electric discharge blowing my stuff up, even though I use small-scale surge supressors.

snorg wrote:

So here I find myself contemplating buying a used sever, a 24 core beast that has redundant 1500w power supplies. One of my hobbies is 3D computer graphics and animation, so I figure this is a quick way to get a render farm in a box. My current workstation (which I built out of used server parts) probably pulls something like 400 watts at full load, I'm assuming this other system would be at least double that. So if I were running these 24/7/365 my fear is my power bill will jump a good $50-$60 per month, easily. I could probably run them at a 50% duty cycle and run my rendering jobs overnight when rates are cheaper, and just leave them in power saving mode most of the time. Would you do it? I don't strictly need it, but damn....24 cores in one box.

I don't do rendering so I can't give an informed opinion.

Do most of you run your systems, retro or otherwise, most of the time? Or only when active? I have a laptop I run pretty much all the time but it doesn't use that much power and is set to go to sleep when I'm not using it.

Laptops stay plugged in and on all the time. I set the screen to blank after 10 mins. I know some people say that's bad for the battery, but in my experience, it's good for the battery. I have a 6 year old laptop that stayed plugged in and on almost 24/7 and the battery is still at 100% capacity. No measurable cell degradation yet.

For desktops, it depends on its job. Some I configure just like the laptop, others I allow to sleep. I use battery backups on anything important, so I'm not as concerned about power outages/spikes.

My retro PCs stay OFF unless I am playing on them. They boot up in mere seconds, and I sometimes go days without being able to play on them, so it just makes sense for me to leave them off.

Retro PCs for me stay off, I only fool around with them once in a while.

This deal on this server is almost too good to pass up, but there will always be other decommissioned servers on Ebay.
Adding this to my network could mean the difference between a short film taking a weeks to render to taking days to render.
So I imagine it would only be running full blast for the time where I was actually finalizing something, any other time maybe no more than 6-8 hours a day full load, the rest of the time in sleep mode.

Here they're only on when I need them to be on (my main rig's on about 18 hours a day while most of the retro stuff only gets used sporadically), but it's got nothing to do with power consumption, which I'm not concerned about in the least (all power-saving features deliberately disabled everywhere I can find them for performance's sake)

Not having any TV's, radio's, mobile chargers or even laptops laying around, I think that kind of levels out the bills (I don't normally do any renderings, and heavy duty emulation no more than a couple hours a day tops)

How much is too much? if you're getting what you want out of it, then it's not. If you aren't, then it might very well be.

Seeing how performance per watt is the new measure that keeps improving, I think you are asking a good question! I certainly don't know the answer as I have, and likely never will, play around with such server gear.

Normally this isn't something I would seriously consider, either. However, the quoted system is only $200. Once you tack on the copy of Windows 2008 Server and the shipping (damn things weighs like 90 pounds) you are getting up around (or over) $400. I'm having a really hard time justifying that kind of expense, given I won't be able to run the thing constantly (the only good reason to own something like this is to run a massive website/forum, process SETI @home or do folding @ home jobs or other computationally intensive tasks).

Would be cool as hell but I think I'm going to have to pass.

I've run old servers like this before (thought not with quite so many cores) and I can tell you this:

• Your electricity bill will go up. You can do things like enable CPU throttling to reduce clock speed when they are idle, but servers weren't built to save power, they were built to work, so enabling all the power saving features rarely makes much of a difference.
• They are loud. They are designed to sit in a data centre where people rarely go, so noise isn't a concern. It's no surprise that some data centres require you to wear hearing protection when entering, for health and safety reasons.
• No, you can't turn the fans off. And no, you can't just cut the wires either, otherwise they start beeping extremely loudly to warn you that the fans have failed.
• Something will break. Something always breaks. And typically servers use custom parts so sourcing a replacement is a real pain. And don't think "oh I'll just stick a cheap hard drive on that IDE or SATA interface" because no, that's for an optical drive only, and the BIOS will complain loudly if you connect any other kind of device to it.
• Then you need special error correcting RAM, SCSI/SAS disks, a power supply goes and there's no way to get a replacement, etc, etc.
• On the plus side, they can be cheap to run in winter because you don't have to run your normal heating quite so much 😉

The only servers I've run have been ones I have gotten for free, and looking back now, I'd never pay for one (unless *maybe* if it only used bog standard components) - otherwise they are way more trouble than they're worth.

Snayperskaya wrote:

I unplug my main desktop and notebook from outlets whenever I leave home. Can't risk having a surge/electric discharge blowing my stuff up, even though I use small-scale surge supressors.

Same here, especially when one of my computers got ruined by a lightning strike --got myself paranoid ever since.

There's such thing as having a quality UPS if you can afford it. I once had a PSU draw 220v from a 110v outlet through the UPS (don't ask about that), then the whole thing went up into smoke and sparks while both the UPS itself and everything connected to it remained perfectly operative and didn't even shut down. Nevertheless, the very socket the PSU was plugged into, it couldn't even be found by reason of the cable that had forcibly become part of the unit as a mess of melted plastic. That gives you an idea how bad it was and how reliable these beasts can be.

I don't have use for anything as powerful as what's being discussed here, but I do run an old socket-940 Opteron server. It holds the bulk of my data files because I decided it made sense to centralize them, both for convenient access and more consistent/reliable backing up.
The server board is a Tyan S2882-D, which is a dual socket board, but I only have 1 socket populated because it's all I need. The CPU is a dual core. Out of 8 memory sockets, only 2 are populated because it's ample for current usage.
The system has plenty of headroom above the performance that I actually need out of it, so much of it's potential is untapped.

The K8 is friendly to throttling and undervolting. This was a major reason I chose it over P4 Xeon boards.
A great deal of idle power was saved with software undervolting. I don't remember the factory defaults, but I think it was something like 1.1v at minimum. I lowered it to I think 0.875v at 1GHz. It's set 0.05v higher than the minimum that passed all the stress tests I ran on it.
This lowered the CPU power usage so much that the fan doesn't even run most of the time. It's configured to start ramping up the fan at 45C, and it tends to switch on and off around that threshold. It barely starts to rotate and then stops because it's already cooled a few degrees. In cooler room temps it just stays off completely.
The worrying scenario is when you hit it with a sudden load. The CPU temperature can rise 20C in moments, while the fan is still accelerating to catch up. It's kind of scary to watch this happen. I would like to figure out some way to limit the CPU to a moderate power state until the fan has reached a minimum speed, then unlock the CPU to continue up to it's highest performance state. That way the fan vs CPUtemp drag race wouldn't happen.

The hard drives are set to spin down after 20 minutes. There are multiple data drives that are usually asleep, but the OS drive spins constantly.
For a while I used a Hitachi 2.5" laptop drive for the OS, but I failed to realize that drive was autoparking it's heads every few seconds, only to have them unparked moments later by the OS. This quickly ruined the drive. It's the same problem that WD Green drives became famous for.
There is a DVD drive but it's unplugged because they draw power all the time. I'll just plug it in if I need it.
It uses onboard video. I did add a PCIX SATA card, it increased the draw by a few watts.

I unplugged some case fans that I decided weren't needed. As mentioned above, the CPU fan is thermally controlled and hardly runs at all when the CPU is in it's lowest performance state (the mode in which I have undervolted it).

I think the system initially used something like 100W at idle, but after all the tweaking the idle power ended up at 65W. I've added more drives since then so it might be closer to 70W now (even though they spin down, they use a little power just by being plugged in).
There are still 2 120mm fans blowing constantly over 2 cages of hard drives (usually needlessly), and an 80mm exhaust fan separate from the PSU. I would like to improve these to be controlled based on detected temperatures, but I haven't done it. I made an attempt with one of the hard drive fans, but the fan I tried won't turn completely off.

The motherboard doesn't support S3 standby. This seems to be a typical limitation of server boards of the era that have multiple PCI buses. Maybe it's a chipset limitation. Not sure whether that limitation still applies on newer boards.
The only boards I saw with S3 support were entry level boards that had desktop chipsets.
It will support hibernation (S4), but the boot delay would be annoying. Also, wake on LAN is kind of a pain to deal with. So I've left mine running fully awake while idle.

==
As far as cost - unless you actually need something proprietary, then server parts are generally cheap. Lots of server components get liquidated and few people are buying them. Socket 940 CPUs and Registered ECC RAM were both much cheaper than the equivalent desktop parts would have been. The PCIX SATA card was cheap also, probably about equal to the cost of a common 32bit PCI card.
The PSU I use is an AcBel API4FS06 550W, which I think are still cheaply available. Thousands of them got liquidated as NOS a few years ago, and I grabbed a few. The one I use on my modern desktop/gaming machine seems a bit stressed, but on the server it's perfectly happy.

My board and case are standard EATX. A proprietary case and motherboard combo might create some headaches, depending how weird it is.
Standard ATX cases are easier/cheaper to find than EATX. I splurged on the EATX because I just wanted it.

Right now my server (at about 65-70W idle) uses less power than my primary desktop that I leave running for at least 18hrs a day (in the low 100s). So the desktop is the biggest power hog that concerns me, but that's another topic.

Everything in my office/games room is on remote-control powered sockets, only my main PC/server runs 24/7, and it's on a UPS.

Anything I want to use I power up with the remote control (speakers, monitor, games machines, retro computers etc), and then the relay in the power strip physically turns it off when I'm finished - it doesn't sit in standby. The power draw from all those wall wart power supplies, monitors/speakers in standby etc would be a decent chunk of electricity I'd bet.

Personally, I wouldn't go the route of using old servers; I run loads of kit like that in my day job and you don't want it in your house. You're better off investing in newer, lower power kit. For example, a good few years ago I kept a dual Athlon MP workstation in service as my main server in addition to my desktop. It was quite happy chugging along serving my website, running MySQL etc. Both it and my main PC were plugged in to the UPS which has a load meter on the front; it was always around the 50-60% level. I eventually decided to upgrade it to (again a few years ago now) a low power Athlon II X4 (40w TDP) and replaced an array of 8x 500GB SATA drives with 5x 1.5TB drives. The UPS load indicator dropped to around 10-15% with the new system. That made a noticeable difference to my electricity bills!

megatron-uk wrote:

Everything in my office/games room is on remote-control powered sockets, only my main PC/server runs 24/7, and it's on a UPS.

Anything I want to use I power up with the remote control (speakers, monitor, games machines, retro computers etc), and then the relay in the power strip physically turns it off when I'm finished - it doesn't sit in standby. The power draw from all those wall wart power supplies, monitors/speakers in standby etc would be a decent chunk of electricity I'd bet.

Personally, I wouldn't go the route of using old servers; I run loads of kit like that in my day job and you don't want it in your house. You're better off investing in newer, lower power kit. For example, a good few years ago I kept a dual Athlon MP workstation in service as my main server in addition to my desktop. It was quite happy chugging along serving my website, running MySQL etc. Both it and my main PC were plugged in to the UPS which has a load meter on the front; it was always around the 50-60% level. I eventually decided to upgrade it to (again a few years ago now) a low power Athlon II X4 (40w TDP) and replaced an array of 8x 500GB SATA drives with 5x 1.5TB drives. The UPS load indicator dropped to around 10-15% with the new system. That made a noticeable difference to my electricity bills!

You can also have some nice savings by investing in 80PLUS Gold/Platinum/Titanium PSUs for 24/7 systems. If correctly dimensioned they will always work at their peak output efficiency (around 60% max load).

I have sometimes used low end Coppermine Celerons for some long running tasks, including very long downloads (going on for weeks to months), and for VoIP. Those CPUs use little power, so I like them for jobs that they aren't overburdened to handle. Desktop Coppermines don't have any throttling capability but they don't have much need of it. The lower end models are especially thrifty. These machines can run in the 20-30W range (measured at the wall).

Certainly the P4/Athlon era is a lot worse about power usage. The CPUs and the motherboards are both more power hungry, so I don't find them as practical for long run times as a P3. However, many of them do have working S3 standby support, so that can work out well for a sporadically used desktop. That feature won't help if the computer needs to stay awake for a long term task though.

I like the mobile-style voltage/clock throttling features that started to be supported on K8 and later systems. This was the major reason I used a K8 for my file server instead of a P4 Xeon. You can set them to undervolt and underclock while they still stay awake and functional.

===
I leave my main desktop on during most of the day.
My old desktop PC was an ABit AN7 nForce2 board, and even though it's power consumption was fairly high, at least it had working S3 standby. I liked that feature and made plenty of use of it. I could leave my computer running and it would just drop into S3 when I wasn't using it, then I could come back and my desktop would reappear within seconds. It was totally reliable and very nice.

My current Phenom2 PC has been a serious step in the wrong direction for power usage. It uses about 50-100% more power than my nForce2 system did. I think a big part of that is the graphics card, and part is the bigger PSU that goes along with it. The motherboard itself is also probably more hungry, as is the CPU to at least some extent.
Something that really irks me about this machine is that S3 standby is broken. It's supposed to have the feature, but like so many other motherboards, if you try to use it then the computer will wake up dead. End result is that it runs at full power all the time. In the summer I might install a CPU throttling utility, but last time I tried that, I didn't like the laggy responsiveness that it caused.

I've thought about maybe retiring my main PC from daily use and just using it for games instead. I could get some equally capable but more efficient machine for web browsing/etc. I already have some of the components so it might not cost much to put together. This creates confusion about how to allocate the monitors though.
With some adjustments, I could even have the file server (which runs 24/7 anyway) double as a basic desktop, but I'm not sure I want to clutter up it's software environment by doing that.

Phenom II's (especially the X6 models) pack way more punch than a 462. They are respectable CPUs even nowadays, even thought power consumption of AMD high-end CPUs don't fare well against Intel.

If you run Windows you can always set a power plan to limit the idle/low-load maximum CPU clock. Don't know how Linux handle power schemes.

shamino wrote:

... My current Phenom2 PC has been a serious step in the wrong direction for power usage. It uses about 50-100% more power than my nForce2 system did. I think a big part of that is the graphics card, and part is the bigger PSU that goes along with it. The motherboard itself is also probably more hungry, as is the CPU to at least some extent.

Something that really irks me about this machine is that S3 standby is broken. It's supposed to have the feature, but like so many other motherboards, if you try to use it then the computer will wake up dead. End result is that it runs at full power all the time. In the summer I might install a CPU throttling utility, but last time I tried that, I didn't like the laggy responsiveness that it caused ...

I recommend K10stat to control Phenom IIs, wonderful little tool for both under and over clocking/volting. 😎