need to replace power supply on legacy silent pc

[ginahoy]

Well, you did say I needed a thermal camera in response to my comment about probe placement. I thought perhaps you were using sarcasm to make your point, which is why I asked if you were serious. Again, your reply sounded like you were being dead literal in your recommendation. So now you've make it clear that checking internal PSU temps would be a waste of time/energies. Thanks for the clarification. I apologize for the misunderstanding.

What's the difference?

Big difference. You made the point that my fan swap is a trade-off between noise and longevity. I could care less if I reduce remaining life from, say, 3 years to 2-1/2 years. But I do care if the modded power supply overheats to the point that fails in the near term, especially since there's apparently no way to anticipate failure, such as by monitoring the voltage levels.

[quest_for_silence]

Well, you did say I needed a thermal camera in response to my comment about probe placement. I thought perhaps you were using sarcasm to make your point, which is why I asked if you were serious. Again, your reply sounded like you were being dead literal in your recommendation. So now you've make it clear that checking internal PSU temps would be a waste of time/energies. Thanks for the clarification. I apologize for the misunderstanding.

There's no problem, gina, I am sorry for my part (of the misunderstanding) too, so please accept my apologies: I assumed my thoughts were clear just by my words, and it's never so even if I explicitly used the expression "I guess" in front of them.

Actually I didn't imply you do need only that instrument to carry on such an investigation/measure, but that the use of a point to point measures requires a well thought/complex methodology, which may not be easy to implement if you (royal you) don't know that much about that PSU design: OTOH a FLIR camera could have offered a more comprehensive, whole picture, to let you (royal you) visually assess which parts may suffer more of the reduced cooling.

Broadly speaking, like people PSUs wear out over time: a nice article was made a while ago by Paul/Spectre on HardOCP (this one); albeit it is not a general rule, it may be an useful insight knowing that after about 7 years that PSU lost about 1/4 of its capacity (I'm simplifying: actually the performances degraded and at nominal capacity are out of specs).

With reference to your specific PSU, we (you and me) don't know how it grew old: we know it didn't worked that hard, but to which degree that wear out it itself isn't clear, and therefore we can't predict once more how a reduced cooling prowess may affect the relevant performance and reliability.

Still broadly speaking the most prone to fail parts due to heat are likely rectifiers/mosfets, so those are good candidates to place any probe: OTOH unlikely it's easy to stick something on those chips (at least given how many they are and where they are located), or so I guess.
Even if it were done, than we should know how the specific chips react/derate with reference to temp, so we should know how to read the relevant datasheet, with the usual caveat the datasheet is referred to a new specimen, not to an 8 or 10 years old one: here having a PSU designer as a friend may come handy (I can't help in case). But that's not the whole story, of course.

What's the difference?

Big difference. You made the point that my fan swap is a trade-off between noise and longevity. I could care less if I reduce remaining life from, say, 3 years to 2-1/2 years. But I do care if the modded power supply overheats to the point that fails in the near term, especially since there's apparently no way to anticipate failure, such as by monitoring the voltage levels.

Sorry, there's no such a difference, IMVHO.

Shit happens, long term reliability is a statistical value, you can't know what tide would bring some day.

That's why I said I'd swap the fan and call it a day: whether the PSU had to die, then I'd just replace it.

Perhaps, in order to minimize the risk, I'd buy a Scythe KAZE-JYUNI DB (SY1225DB12M or SY1225DB12H which offers a compatible start up voltage), because the KAZE-JYUNI design has a greater airflow/rpm ratio than the venerable SFF.

At any rate, enjoy your "new" (i.e. fan swapped) PSU, and have a nice time!

[quest_for_silence]

What's the difference?

Big difference. You made the point that my fan swap is a trade-off between noise and longevity. I could care less if I reduce remaining life from, say, 3 years to 2-1/2 years. But I do care if the modded power supply overheats to the point that fails in the near term, especially since there's apparently no way to anticipate failure, such as by monitoring the voltage levels.

Sorry, there's no such a difference, IMVHO.

Shit happens, long term reliability is a statistical value, you can't know what tide would bring some day.

Thinking better, I probably didn't address your concern properly/clearly.

The moot point was "to the point that": well, in my experience even the tiniest airflow didn't lead to near term fail on new units.

To say, I cooled an Antec EarthWatts (Seasonic made) since new with an about 500rpm 80mm fan for more than 6 years without issues.
Average load was somehow similar to your one, probably something more (oc'ed rig), while the original ADDA fan likely put more than twice the flow.

Said that, YMMV: because the unit is old and because shit happens. But you may test the PSU with a multimeter, occasionally: that surely helps.

[CA_Steve]

Still broadly speaking the most prone to fail parts due to heat are likely rectifiers/mosfets

I would have gone for secondary side caps.

[quest_for_silence]

Still broadly speaking the most prone to fail parts due to heat are likely rectifiers/mosfets

I would have gone for secondary side caps.

With reference to ginahoy's own concern, silicon are the weakest point: you may use 125°C japanese everywhere but that won't save you from not decent/degraded silicon.

With reference to proper long term reliability (I mean, wearing out from new state), bad silicon still wear out faster than caps.
The second failure reason are maybe badly programmed housekeeping ICs (not properly working protections, particularly OTP which make OPP/OCP useless).
The other competitor for the place is the fan, if sleeve bearing (or derived).
Then come the caps: the Antec EarthWatts Green is perhaps a perfect example, it uses shitty Ltec ecaps (chinese garbage according to badcaps.net) but it's the Antec with the lowest failure rate ever.

[NeilBlanchard]

So, $41 (shipped) for this PSU:

https://www.newegg.com/Product/Product. ... 6817151086

It is (probably) significantly more efficient than the current unit, and it is new, so the worry of the current unit failing even after getting a new fan - seems to be worthwhile to me. And no need for a 24/20 pin adapter - the extra 4 pins just split off.

[CA_Steve]

With reference to proper long term reliability (I mean, wearing out from new state), bad silicon still wear out faster than caps.
The second failure reason are maybe badly programmed housekeeping ICs (not properly working protections, particularly OTP which make OPP/OCP useless).
The other competitor for the place is the fan, if sleeve bearing (or derived).
Then come the caps: the Antec EarthWatts Green is perhaps a perfect example, it uses shitty Ltec ecaps (chinese garbage according to badcaps.net) but it's the Antec with the lowest failure rate ever.

If it were custom silicon like a gpu, sure, I'd go for this. Especially with the limited expected life of a gfx card and the assumed planned obsolescence. But, power ICs tend to be sturdier animals...and I'd assume the main failure mechanism of an IC would be electromigration. Assuming the OP's max power use is in the 150-160W range and the PSU is rated for 330W. So, half the power and ~ half the temp rise over ambient that you'd see at max load with a static fan profile.

<handwaving and many assumptions ensue>

If you look at Black's equation, compare the effect of current density vs temp on mean time to failure. Current density is a squared function. So, half the expected current density and the MTTF quadruples. Temperature is a e^-T function. What we don't know is the junction temp of the power ICs...So, here's a couple of guesses with 25C ambient.

1) Raise the junction temp from say an expected 75C to 100C and the MTTF is quartered. But, if the power and current density is halved, then the MTTF stays the same.

2) Raise the junction temp from an expected 100C to 125C while maintaining 50% power and the MTTF goes up 20%.

3) Raise the junction temp from an expected 75C to 125C while maintaining 50% power and the MTTF drops to 0.3 of the baseline.

I guess what I'm (probably unsuccessfully) trying to show is since the PSU is running well below it's rated maximum power, the reduction in current density will mitigate most/all reductions in electromigration based lifetime caused by raised junction temperature.

[quest_for_silence]

But, power ICs tend to be sturdier animals...

...with an hard work.

Very recently Corsair had an entire PSU batch with an high failure rate: eventually, switching the sturdier animals within the relevant AVL, they solved the issue.

In the OP case, those S12 are likely the most reliable Seasonics (MUCH MORE than their acclaimed KM/XP), but nonetheless, as you know, those chips have their derating curve and expected lifetime, and the reduced cooling (albeit Seasonic original one were rather conservative) would put more stress on them.

I guess what I'm (probably unsuccessfully) trying to show is since the PSU is running well below it's rated maximum power, the reduction in current density will mitigate most/all reductions in electromigration based lifetime caused by raised junction temperature.

Yes, your guesswork sound not unreasonable at all, but those ICs are about nine years old: how much of their MTTF has gone? As you may see in the HardOCP Silverstone Redux linked above, after 7 years the PSU "lost" 1/4 of its nominal capacity: so what about those values for A, n, and Q found by fitting the model to experimental data? I can't help.

[CA_Steve]

Well, the OP's quadrupled the estimated MTTF of the power ICs just by running the PSU at half (or less) of the rated power up until this point. So, if the thing's got a 5 year warranty for running at rated power, then I'm guessing there's a bit left. Until a cap craps out

[Olle P]

Seriously?! Those images were taken with a $3,000 FLIR E4 320x240!

A CAT S60 is more like $600, and then you get a good smartphone with it.