Reliability of a standard PC power supply in a server envirornment

[ncovert]

I'm building a cheap web server and was wondering about how reliable would a standard PC power supply be when being turned on 24/7. The power supply I am using is a 200W Newton unit (I'm cheap, it's gonna be a P3 based server, 200W will be plenty.) If I remember right, it had japansese caps and none were bloated.

[Rulycat]

Re: Reliability of a standard PC power supply in a server envirornment

Japanese caps would probably make it okay, as long as the guts are half decent. Make sure it's a relatively efficient unit though and not sub 70%, or you'll be wasting $$$ on electricity costs that you tried to save with the PSU.

[ncovert]

Re: Reliability of a standard PC power supply in a server envirornment

Japanese caps would probably make it okay, as long as the guts are half decent. Make sure it's a relatively efficient unit though and not sub 70%, or you'll be wasting $$$ on electricity costs that you tried to save with the PSU.

I would take your advice on efficiency, but I don't worry to much since I don't pay the electric bill :p And the guts are decent; it's a Newton (basically Delta with japanese caps)

[c_hegge]

Re: Reliability of a standard PC power supply in a server envirornment

Believe it or not, I've never seen a newton with Japanse caps before. All the ones I've seen had LTEC and CapXon.

[ncovert]

Re: Reliability of a standard PC power supply in a server envirornment

Believe it or not, I've never seen a newton with Japanse caps before. All the ones I've seen had LTEC and CapXon.

Interesting. I've seen them with Rubycon, Nichicon, and UCC, with a few teeny tiny Ltec's sparingly used

[tom66]

Re: Reliability of a standard PC power supply in a server envirornment

What's the rated MTBF and is the supply new? Even good japcaps will eventually fail. It just depends on how long you need it to run. And can you tolerate downtime?

[momaka]

Re: Reliability of a standard PC power supply in a server envirornment

Those older Delta, Newton, HiPro, LiteOn, and Astec PSUs are indestructable (especially the ones with Japanese caps). I'd trust it to run 24/7 for quite a few years, provided the location where you'll have that computer isn't too dusty.
I still often see old computers like that in 24/7 operation used for print server duty and whatnot. They just *won't* die.

[Wester547]

Re: Reliability of a standard PC power supply in a server envirornment

I still often see old computers like that in 24/7 operation used for print server duty and whatnot. They just *won't* die.

Not throwing a lack of concurrence at you, but if I've read correctly, don't power supplies lose their ability to deliver their rated wattage and specs over time (even if it takes a while and/or just ultimately lowers efficiency)? And as for Hipro, I'm surprised those older units last so long given that you often find brands like Asiacon on the secondaries (that will cause eventual failure, make no mistake). Maybe the thick heatsinks help, or maybe they just have good cooling. Or perhaps quality primaries like Panasonic extend the life of those Asiacons/Arcons/G-Luxons/OST/whatever.

Other than that, so long as you have good capacitors, a good fan, and an otherwise well-built power supply, the above brands would be prudent choices. With those exceptions, I think Antec, Delta, Enhance Electronics, Seasonic, and anything along that nature would serve you (the OP) well. I would think that a power supply could last quite a while being ran 24/7, and if I'm not mistaken would die sooner bearing away many on/off cycles over a long period of time rather than just being kept on 24/7. In my experience, Newton Power (the older ones I've used) are very good and usually can last a very long time without needing a recap or anything else in the case of elder units (I noticed the worst caps on them to be Taicon - and those may never be a problem despite not necessarily materials having supplied from Nichicon).

[momaka]

Re: Reliability of a standard PC power supply in a server envirornment

Not throwing a lack of concurrence at you, but if I've read correctly, don't power supplies lose their ability to deliver their rated wattage and specs over time (even if it takes a while and/or just ultimately lowers efficiency)?

Not exactly.

The only thing that really ages "fast enough" to be noticeable are the caps and the only result you'll get from that is slightly higher ripple as the PSU "ages". Of course, if you use the PSU so that the caps don't sit without a charge and if they are running well below their maximum rated temperature (as is the case with well-designed PSUs), the caps will "age" so slowly that the PSU will become obsolete way before any of that matters.

And as for Hipro, I'm surprised those older units last so long given that you often find brands like Asiacon on the secondaries (that will cause eventual failure, make no mistake). Maybe the thick heatsinks help, or maybe they just have good cooling. Or perhaps quality primaries like Panasonic extend the life of those Asiacons/Arcons/G-Luxons/OST/whatever.

Thick heat sinks do help, but that's not the only reason. The primary reason why they last so long is because the whole filter circuit on the output of the PSU is very well-designed. This off-loads quite a bit off stress from those inferior caps, which in turn makes them last much longer.

Also, HiPro and Delta regularly use much larger rectifiers than necessary. This does 2 things: there's less heat coming from the rectifiers which in turn improves the efficiency. Now add those already cooler-running rectifiers to a very thick heat sink, and you get almost no heat out.

I would think that a power supply could last quite a while being ran 24/7, and if I'm not mistaken would die sooner bearing away many on/off cycles over a long period of time rather than just being kept on 24/7.

Theoretically correct, but in the real world there's more important things to consider. For example, if you run it 24/7 and never turn it off to clean the dust on fans, then it will fail much quicker in 24/7 operation than if you cycle it multiple times a day. Also, 24/7 operation puts much more strain on the fans and they are more likely to fail that way (and of course when a fan fails, so will the caps and you know the rest).

And even if you do keep the fans clean, it still doesn't matter that much. PSUs can take an awful lot of power cycle counts before they fail.

Just for some random numerical data: my family's computer (Dell Dimension from 2006) has a Newton PSU. It's had over 8000 power cycles so far (as did the HDD). I opened it to clean the dust about 4 years ago. Had I not done that, the fan would have failed. I have several other computers that are in the multiple thousands of cycles as well. Haven't had a problem with them either.

[Wester547]

Re: Reliability of a standard PC power supply in a server envirornment

Thick heat sinks do help, but that's not the only reason. The primary reason why they last so long is because the whole filter circuit on the output of the PSU is very well-designed. This off-loads quite a bit off stress from those inferior caps, which in turn makes them last much longer.

Also, HiPro and Delta regularly use much larger rectifiers than necessary. This does 2 things: there's less heat coming from the rectifiers which in turn improves the efficiency. Now add those already cooler-running rectifiers to a very thick heat sink, and you get almost no heat out.

I'm guessing FSP never did the same with their PSUs, as while I have a well designed 300-Watt from them (a FSP300-60BTV), after maybe about ~7,000 power cycles and ~12,000 hours of use, it has a bloated Fuhjyyu on the secondary and has boot problems. Despite having great heatsinks and a high quality, high RPM/CFM ball bearing fan from Pyrotechnic (and nice vents), the bad caps on that power supply didn't last. My guess is that FSP don't quite overbuild their PSUs to have overspec'd capacitors or rectifiers and filter circuits (though that particular unit has two Teapos rated at 85C/680uf/200v on the primaries). But that definitely explains why the Hipros last, thanks. It seems to me, though, that Newton Power's choice of caps in their older units were far more astute than those of Hipro's.

Theoretically correct, but in the real world there's more important things to consider. For example, if you run it 24/7 and never turn it off to clean the dust on fans, then it will fail much quicker in 24/7 operation than if you cycle it multiple times a day. Also, 24/7 operation puts much more strain on the fans and they are more likely to fail that way (and of course when a fan fails, so will the caps and you know the rest).

And even if you do keep the fans clean, it still doesn't matter that much. PSUs can take an awful lot of power cycle counts before they fail.

Well, I'm still sure it puts more strain on a PSU to have, say, 5,000 power cycles and 20,000 power on hours than 500 power cycles and 20,000 power on hours. Aren't heat-cold cycles very pernicious to the soldiered connections and don't, say, motors need more power when first they start (thus pulling more power from the PSU upon boot)? Of course, because these are older units that we are speaking of, perhaps the leaded solder can take it.

Keeping the fans clean is of course vital, though I've heard you can potentially damage the fans by doing it too often. However, I've had a (probably a sleeve bearing since it did it almost silently) fan kind of seize on a 250-Watt PSU from Hipro (a HP-P2507F3P; you posted that PSU in the power supply build thread if I remember) in an old Dell Dimension 8200 I had. It didn't kill the PSU, to my fortune, though note that this was extremely cold weather (37F, about, and the room temperature was probably 14C or something) so it only felt a bit hot at the PSU exhaust intake at the most, though I still imagine it can't be great at all for those Asiacons. Needless to say, that PSU survived ~16,000 hours/6,000 power cycles since December of 2001 so you are definitely correct about it being well made. Note that the fan isn't entirely dead as it will still spin without stop if I take a Q-Tip when it's powered off and "spin" the fan manually, though I noticed that it stops quickly (much quicker than the NMB CPU fan below it) when the system is powered off so I have a feeling the fan is going, more than the capacitors. It slightly "twitches" when it doesn't work and powers on, whatever that means. That, or it works, but with an unusual sound coming from it every other second, kind of like a car engine revving up.

Being somewhat overbuilt, it feels more like a 275-300W PSU to me since I was able to run a Pentium 4 1.7 GHz Willamette (66W TDP) on it, along with 512MB of PC800 RDRAM (two DIMMs/four slots filled up), a Radeon 9500 Pro, a Sound Blaster Audigy LS, two 7200RPM HDDs, a USB 2.0 card, a dial-up card, and a LAN card in it, along with two DVD-RWs at full load without any issue from Spring of 2003 til' Summer of 2009.

Just for some random numerical data: my family's computer (Dell Dimension from 2006) has a Newton PSU. It's had over 8000 power cycles so far (as did the HDD). I opened it to clean the dust about 4 years ago. Had I not done that, the fan would have failed. I have several other computers that are in the multiple thousands of cycles as well. Haven't had a problem with them either.

If I may ask, how many power on hours does the Newton have?

Nevertheless, I would think that, again, to be more on-topic, so long as the server in question is powered down at least sometimes to clean the fans that a 200-Watt Newton Power should do fine as a 24/7 server PSU so long as it's enough to power the system as a whole.

[lti]

Re: Reliability of a standard PC power supply in a server envirornment

Keeping the fans clean is of course vital, though I've heard you can potentially damage the fans by doing it too often. However, I've had a (probably a sleeve bearing since it did it almost silently) fan kind of seize on a 250-Watt PSU from Hipro (a HP-P2507F3P; you posted that PSU in the power supply build thread if I remember) in an old Dell Dimension 8200 I had. It didn't kill the PSU, to my fortune, though note that this was extremely cold weather (37F, about, and the room temperature was probably 14C or something) so it only felt a bit hot at the PSU exhaust intake at the most, though I still imagine it can't be great at all for those Asiacons. Needless to say, that PSU survived ~16,000 hours/6,000 power cycles since December of 2001 so you are definitely correct about it being well made. Note that the fan isn't entirely dead as it will still spin without stop if I take a Q-Tip when it's powered off and "spin" the fan manually, though I noticed that it stops quickly (much quicker than the NMB CPU fan below it) when the system is powered off so I have a feeling the fan is going, more than the capacitors. It slightly "twitches" when it doesn't work and powers on, whatever that means. That, or it works, but with an unusual sound coming from it every other second, kind of like a car engine revving up.

I saw one of those power supplies with the same problem. Its fan was still working properly, but the thermistor in the fan control circuit had fallen out of the clip that holds it against the heatsink.

[999999999]

Re: Reliability of a standard PC power supply in a server envirornment

How many hours of use and years of age does the PSU have? That is the primary issue when it has good quality caps, and whether it's a dual ball bearing fan.

If the PSU is as old as the rest of the P3 platform and you need to rely on this server, I would buy a new power supply or at least replace ALL the capacitors in it, and the fan if not dual ball bearing.

As for dust issues, seal up every place air can get in except areas you put filter panels over. My ~ 10 year old 24/7 fileserver in the basement has never needed to have dust cleaned out, though it may help that there's no carpet down there and people don't spend much time down there so there's few skin flakes or other materials to turn into dust.

[Wester547]

Re: Reliability of a standard PC power supply in a server envirornment

I saw one of those power supplies with the same problem. Its fan was still working properly, but the thermistor in the fan control circuit had fallen out of the clip that holds it against the heatsink.

Wow... that might be what happened. But the fan did make something like an intermittently dead/struggling bearing sound if it didn't twitch and worked, which makes me wonder if it would be that same problem. That might explain why seemingly dead Hipro PSU fans is a commonplace problem. But if someting like the above happens, doesn't that mean it's no longer a temperature controlled fan? Also, I meant to say the P4 1.7 GHz is ~64W TDP, not 66/67. ^^; Not that it deviates from what I was trying to say. And I agree with 999999999 as anything newer than a P3 would need more power on the 12v rail (Newtons rated at 200W usually only have something like 6A at the 12v rail from what I know).

[Behemot]

Re: Reliability of a standard PC power supply in a server envirornment

I'm guessing FSP never did the same with their PSUs, as while I have a well designed 300-Watt from them (a FSP300-60BTV), after maybe about ~7,000 power cycles and ~12,000 hours of use, it has a bloated Fuhjyyu on the secondary and has boot problems. Despite having great heatsinks and a high quality, high RPM/CFM ball bearing fan from Pyrotechnic (and nice vents), the bad caps on that power supply didn't last. My guess is that FSP don't quite overbuild their PSUs to have overspec'd capacitors or rectifiers and filter circuits (though that particular unit has two Teapos rated at 85C/680uf/200v on the primaries). But that definitely explains why the Hipros last, thanks. It seems to me, though, that Newton Power's choice of caps in their older units were far more astute than those of Hipro's.

This. Many Fortrons (usually full of dust TBH) I boguht were already bloated. Not that I won't change the caps anyway just because of their unreliability which I won't cover with warranty this way.

Well, I'm still sure it puts more strain on a PSU to have, say, 5,000 power cycles and 20,000 power on hours than 500 power cycles and 20,000 power on hours. Aren't heat-cold cycles very pernicious to the soldiered connections and don't, say, motors need more power when first they start (thus pulling more power from the PSU upon boot)? Of course, because these are older units that we are speaking of, perhaps the leaded solder can take it.

Well if there is thermistor (and there should) on the input, it does not actually like spikes. I have one A7N8X-E Deluxe with damaged turning on/off circuitry, basically it requires manual break of the +5 V SB rail coming into board to turn off. I was tuning on/off whole PSU before and succesfully destroyed one thermistor. No problem since the change (and than I stole it from parent's computer into my, swaped for Fortron with switch on green wire).

[lti]

Re: Reliability of a standard PC power supply in a server envirornment

Wow... that might be what happened. But the fan did make something like an intermittently dead/struggling bearing sound if it didn't twitch and worked, which makes me wonder if it would be that same problem. That might explain why seemingly dead Hipro PSU fans is a commonplace problem. But if someting like the above happens, doesn't that mean it's no longer a temperature controlled fan? Also, I meant to say the P4 1.7 GHz is ~64W TDP, not 66/67. ^^; Not that it deviates from what I was trying to say. And I agree with 999999999 as anything newer than a P3 would need more power on the 12v rail (Newtons rated at 200W usually only have something like 6A at the 12v rail from what I know).

The fan wouldn't start up because the temperature of the air inside the power supply is not warm enough to make the fan spin. Look at these pictures. The thermistor should be inside the clip, not sitting above it. Check that and the fan. The sleeve bearing Adda fans that Hipro uses are known to seize up. The power supply in that post seems to have the oldest working Adda sleeve bearing fan in the world. It still spun freely and the bearing still had plenty of oil, and that power supply is now nine years old and ran 24/7 for its entire life.

I have an older Hipro power supply from a Compaq that had a ball bearing fan and the +12V rail is only rated for 2.1A.

[999999999]

Re: Reliability of a standard PC power supply in a server envirornment

I don't think they actually intended for the fan to not be spinning at any time including when it first turns on and the thermistor would be at the same temperature then whether it was still clipped to the heatsink or not. More likely the fan was chewing up the bearing and needed an abnormally high startup current to break free of the bind it was in.

The biggest risk I see with the Adda fans is turning a system off. You might have a fan run for many years but then if there's a power outtage or it's turned off, the fan won't spin anymore even if it might have stayed spinning a long time if the system hadn't been turned off.

Every now and then I'll come across PSU that will shut themselves off if they don't detect the fan is spinning. One comes to mind that had a design where it didn't even need a fan with an RPM or rotor signal lead to do so, it just sensed the pulsing current on the fan DC power feed. I wish they'd all do that, but it's too much to hope for if the PSU itself seems to be worth little more than the box and power cord that came with it.

[Wester547]

Re: Reliability of a standard PC power supply in a server envirornment

The fan wouldn't start up because the temperature of the air inside the power supply is not warm enough to make the fan spin. Look at these pictures. The thermistor should be inside the clip, not sitting above it. Check that and the fan. The sleeve bearing Adda fans that Hipro uses are known to seize up. The power supply in that post seems to have the oldest working Adda sleeve bearing fan in the world. It still spun freely and the bearing still had plenty of oil, and that power supply is now nine years old and ran 24/7 for its entire life.

The HP-P2507F3P's fan is an ADDA sleeve bearing? I wonder how powerful it is (being a 80mm fan). ADDA fans are a recipe for disaster, a sleeve bearing certainly doesn't give aid to that. At least the fan is vertically mounted and not horizontally. Still, for an ADDA sleeve bearing, it seems the fan in that PSU lasts quite a while. So do the aforementioned inferior capacitors as barring the fan, capacitors, and thermistor, I feel it's one of the best PSUs for its time and stature. And honesty to tell, the "Q-tip" method I used makes the fan work for now, so I'll just go with that until the fan stops completely. ^^; (at which point I'll immediately perform a cold shutdown) But I appreciate the information.

Every now and then I'll come across PSU that will shut themselves off if they don't detect the fan is spinning. One comes to mind that had a design where it didn't even need a fan with an RPM or rotor signal lead to do so, it just sensed the pulsing current on the fan DC power feed. I wish they'd all do that, but it's too much to hope for if the PSU itself seems to be worth little more than the box and power cord that came with it.

My HP-P2507F3P did not shut down when the fan was seized and it was on for 30 minutes after, though again, perhaps the weather was too cold for the PSU to draw close to overheating even without a working fan.

This. Many Fortrons (usually full of dust TBH) I boguht were already bloated. Not that I won't change the caps anyway just because of their unreliability which I won't cover with warranty this way.

If you put aside the Fuhjyyus in that PSU it was very good and very powerful for a power supply rated at 300W. It remained functional for a while, for a PSU that had bad capacitors. And that's an interesting debate to spark, whether or not a fan is more worn down by keeping it on 24/7 or putting it through too many off/on cycles...

[momaka]

Re: Reliability of a standard PC power supply in a server envirornment

Well, I'm still sure it puts more strain on a PSU to have, say, 5,000 power cycles and 20,000 power on hours than 500 power cycles and 20,000 power on hours. Aren't heat-cold cycles very pernicious to the soldiered connections and don't, say, motors need more power when first they start (thus pulling more power from the PSU upon boot)? Of course, because these are older units that we are speaking of, perhaps the leaded solder can take it.

The stress from power cycling isn't that much, even though you do have quite a few system components pulling much more power.
Lead-free is indeed a bit more fragile, but in applications such as a PSU, it can still take quite a bit of stress just fine.

Being somewhat overbuilt, it feels more like a 275-300W PSU to me since I was able to run a Pentium 4 1.7 GHz Willamette (66W TDP) on it, along with 512MB of PC800 RDRAM (two DIMMs/four slots filled up), a Radeon 9500 Pro, a Sound Blaster Audigy LS, two 7200RPM HDDs, a USB 2.0 card, a dial-up card, and a LAN card in it, along with two DVD-RWs at full load without any issue from Spring of 2003 til' Summer of 2009.

It is very overbuilt, IMO - but that's good and that's how it should be.
Coincidentally, I have that very same CPU from a computer I found in the dumpster. What's even more bizzare is that the board that came with that CPU uses the 5V to power it and not the 12V rail. And that CPU peaks at around 70-ish Watt, so that's 14A of current there on the 5V rail, not counting other devices. When I had that PSU power that computer, it ran cool even in 80F/27C weather.

If I may ask, how many power on hours does the Newton have?

Same as the HDD in that PC - just turned 10 000 two months ago. Basically not much power-on time and many power cycles.

Nevertheless, I would think that, again, to be more on-topic, so long as the server in question is powered down at least sometimes to clean the fans that a 200-Watt Newton Power should do fine as a 24/7 server PSU so long as it's enough to power the system as a whole.

+1

[Wester547]

Re: Reliability of a standard PC power supply in a server envirornment

It is very overbuilt, IMO - but that's good and that's how it should be. Coincidentally, I have that very same CPU from a computer I found in the dumpster. What's even more bizzare is that the board that came with that CPU uses the 5V to power it and not the 12V rail. And that CPU peaks at around 70-ish Watt, so that's 14A of current there on the 5V rail, not counting other devices. When I had that PSU power that computer, it ran cool even in 80F/27C weather.

I don't know how cool my PSU (same as yours, I believe - a HP-P2507F3P) ran when it was running fine because I never did feel how hot it became at the exhaust intake. ^^; It didn't help that my Dell Dimension 8200 had no thermal sensors so I had no way of knowing how hot my PC was running (barring my two HDDs which gave me their temperature) outside of opening the PC and feeling it for myself (after shut down, to be safer). But since I've been having fan issues with it, even when the fan works, it feels slightly warm at the intake, even in 60F weather. I don't know if that's because the fan isn't running at optimal speed anymore (it stops rather quickly when I power it off, so something is obviously wrong, whether it's starting to seize or the clip fell out of the thermistor). The fan in the PSU, I mean.

Interesting that it's 27C weather for you. My room temperature here is often 25C, even in 70F weather, but that could be because the room my computer is in is closer to my attic. And what board is that? Don't Pentium 4s draw power from the 12v rail on a PSU or does that depend more on the board? Does the (Socket 478) Intel 850 series of boards use the 12v or 5v rail for their CPUs? :O (since that's what I have in there)

Same as the HDD in that PC - just turned 10 000 two months ago. Basically not much power-on time and many power cycles.

To be honest, I consider 10,000 hours a long time for a PC to be on. But I don't have the financial resources to run my computer 24/7 so it's more half that or less, which means, for me, usually, one on and off cycle a day. That, for me, would roughly mean....

1 year of use = ~4,000 hours
2 years of use = ~8,000 hours
3 years of use = ~12,000 hours
4 years of use = ~16,000 hours
5 years of use = ~20,000 hours

And so forth. The most I've ever had on a PC power-on hour wise is 20,000 hours and 3,000 power cycles (though that computer had three different PSUs in it, all of which worked fine for the life of the computer). It would have been more had the KZGs on that particular board not died, but I suppose that was inevitable.

As far as the 200-Watt Newton for the server goes, I would recommend not loading the 12v rail much (I don't think there would be much reserve on that rail, on a PSU like that).

[999999999]

Re: Reliability of a standard PC power supply in a server envirornment

^ It was an Intel design guideline to use the 12V rail to power the CPU VRM subcircuit on motherboards. "Most" boards for P4 did, but manufacturers still had the final say. If a board uses the 4 pin 12V PSU connector, it's powering the CPU through that and if it doesn't, it's not, is using 5V rail instead.

Similar applied to later generations of AMD Athlon CPUs, by the Athlon XP era a lot of boards still used 5V for power but AMD had recommended 12V rail which became more popular at the end of that era.