As I understand it, most switch-mode regulators for processor core voltage do not depend on a minimal ESR of the output capacitors, at least not knowingly by design. There is a slight chance that you operate the regulator under conditions that were never tested, so maybe the regulator starts misbehaving, that's why the general recommendation to stay in the same region of ESR is basically good. When I fix stuff, the ESR is actually the second value I look at, if at all. I consider the maximum allowed ripple current the primary factor that decides whether the replacement cap will last as long as the broken one. Caps with a sufficiently high ripple current rating tend to have low enough ESR, as higher ESR means more waste heat at the same ripple current, and heat is the last thing you need for longevity.

If it were my system, I wouldn't worry too much about too low ESR, and just test it with the ZLJ caps, but your mileage may vary.

Real reason is need for ripple current designed into the motherboard back in the day because of good access to ultra low ESR electrolytics. If the capacitors is too few, I'd split the capacitors using lowest ESR electrolytic, say the motherboard uses 6 for the processor's v core supply, I'd split the capacitors to 4 electrolytics and two polymer capacitors to make up for ripple current of modern capacitors.
Also mixed capacitors blends the frequency responses more broad too.

You could make extreme ultra low ESR using modern ceramic MLCC capacitors in parallel on a home brewed PCB and put one in too for the V-core.

Cheers,

A typical VRM has approximately 8 filter capacitors (eg GA-BX2000).
So calculating the combined ESR gives (theoretical value without considering wire resistance):
YXF: 0.16ohm / 8 = 0.02ohm
ZLJ: 0.059 / 8 = 0.007375 ohm

In LDOs, tantalum capacitors and electrolytic capacitors cannot be replaced with MLCCs and polymer capacitors due to their phase margin (having ESR limits), but what about VRMs?

Also(1): I don't think it oscillates the signal by replacing the filter capacitors (for PI and EMC) far from the VRM with any low ESR. However, it improves emissions but worsens immunity.

Also(2): Can I use one or two SEPC 6.3V 1000uF (0.018ohm) instead of YXG 6.3V 1000uF *8?
SEPC ripple current is 3530mArms and YXG is 840mArms.

Is my thinking correct?
Thank you,

maekawa wrote on 2022-12-25, 18:14:

According to TheMobRules on this topic, S7/Slot1 and S370 require different ESR caps!

where exactly did you read that? because its not in the link you provide

maekawa wrote on 2022-12-25, 18:14:

Should I buy the correct one immediately?

you already did

maekawa wrote on 2022-12-25, 20:11:

A typical VRM has approximately 8 filter capacitors (eg GA-BX2000). So calculating the combined ESR gives (theoretical value wit […]
Show full quote

A typical VRM has approximately 8 filter capacitors (eg GA-BX2000).
So calculating the combined ESR gives (theoretical value without considering wire resistance):
YXF: 0.16ohm / 8 = 0.02ohm
ZLJ: 0.059 / 8 = 0.007375 ohm

In LDOs, tantalum capacitors and electrolytic capacitors cannot be replaced with MLCCs and polymer capacitors due to their phase margin (having ESR limits), but what about VRMs?

The term VRM isn't the best fit for what you mean. "VRM" is a "voltage reduction module", and early VRMs (on 486 boards) often were in fact impemented using linear regulation for example with the LM1086 LDO regulator chip. You can express more clearly what you mean by using the term "(step down) switch-mode regulator". As far as I know, there are no known minimum ESR values for most switch-mode implementations, so you are most likely fine.

maekawa wrote on 2022-12-25, 20:11:

Also(1): I don't think it oscillates the signal by replacing the filter capacitors (for PI and EMC) far from the VRM with any low ESR. However, it improves emissions but worsens immunity.

Exactly so. Caps "far away" from a switch-mode regulator would not cause instability, because the trace resistance and inductance adds to the apparent ESR/ESL at the regulator. So even in case they would use one of the ESR-sensitive LDOs to generate the rail, instability is unlikely. You should also keep in mind that the ESR issue about instability only applies on the output of the regulator, not on the input. Caps to smooth the 5V or 3.3V (on ATX boards) supply, that get directly fed from the PC power supply can be as low ESR as you want. The most delicate position is the output filter caps of the core voltage regultor.

maekawa wrote on 2022-12-25, 20:11:

Also(2): Can I use one or two SEPC 6.3V 1000uF (0.018ohm) instead of YXG 6.3V 1000uF *8?
SEPC ripple current is 3530mArms and YXG is 840mArms.

While you get good enough ripple current handling capability and low enough ESR with those caps, dropping the capacitance to a quarter of the original value is something I would not blindly recommend. The factor that determines the amount of output ripple is acutally the total impedance of the capacitor bank, which is the result of of the ESR, the ESL (the lower, the less impedance) and the capacitance (the higher, the less impedance). You could do a rough calculation on how much the capacitance of 8mF or 2mF contributes to the total impedance at around 500kHz (a typical switching frequency for the core voltage regulator) to determine whether dropping the capacitance is significant. Ignore the contribution of the ESL (equivalent series inductance) for the moment.

In a nutshell: Physical fact: ESR + capacitance determines the amount of ripple, and ripple current rating determines the longevity of the replacement caps. Personal experience: If you find brand caps with sufficiently high current rating at the original capacity, you nearly always also match ESR requirements.

A couple of things:

8x16mm caps on a Slot1 motherboard might interfere with CPU insertion. Ideally you want 8x11.5mm caps there. Not exactly that common, but you can use 6.3V 820uF parts in 8x11.5mm, and it should be just fine.

Next is cap longevity. Looking at each series you specify:

ZLJ: 8000 hours (ZLJ endurance rating varies with dimension and voltage)
YXG: 4000 hours
YXF (6000 hours - ditto on variation of endurance rating with dimension and voltage)

ZLJ is obviously the superior part here. This is not empirical however, and I don't know if ZLJ is really better than YXF. You'd have to test that in a lab.

YXG is decent enough, but they do tend to go out of spec after 20 years or so.

mockingbird wrote on 2022-12-26, 10:38:

Next is cap longevity. Looking at each series you specify: […]
Show full quote

Next is cap longevity. Looking at each series you specify:

ZLJ: 8000 hours (ZLJ endurance rating varies with dimension and voltage)
YXG: 4000 hours
YXF (6000 hours - ditto on variation of endurance rating with dimension and voltage)

True. Nevertheless, note that these values are at rated maximum ripple current and temperature. If you use a way below the rated ripple current and at lower ambient temperature, the longevity will increase. Of course, this applies to all those series, but as the end result, a "6000 hour" cap that is operated at half the maximum ripple current might easily outlive an "8000 hour" cap that is operated at the maximum permitted current.

For the same reason, a 4000 hour 125°C cap most likely outlives any 8000 hour 105°C cap, if all other parameters are equal. While it's not precise science, you can use "Arrhenius' law" for a first estimate of longevity at lower temperature: It claims that chemical reactions happen at approximately half the speed if temperature is reduced by 10°C. Assuming that the initial degradation is due to reactions that follow Arrhenius' law, a 4000 hout 125°C cap is to be specified for 16000 (round down to 15000 for some margin) hours at 105°C.

mkarcher wrote on 2022-12-26, 11:30:

While it's not precise science, you can use "Arrhenius' law" for a first estimate of longevity at lower temperature<snip>

Sorry to nitpick, but there is also the variable here of the aqueous series with a higher endurance rating which is not necessarily better than a non-aqueous series... So there's some really sophisticated chemical engineering here at play.

Thank you all. I am released from the crisis of buying medium-ESR again.

Q1. Why do switching power supplies that are not designed for low impedance loops reject a ultra low esr caps such as polymer?
Is it prohibited to use ZLJ in a switching power supply that uses medium ESR (like SXE, PL) for the secondary filter?

Q2. I would like to ask about the so-called polymod.
I think the consensus in Vogons is that it's not recommended to recap the P3 capacitors to polymer capacitors.

Why is this?

My guess:
1: Expensive.
At Mouser it's 192 yen ($1.44) @ 100 pieces, but at Akizuki Denshi, a local store in Japan, it's 90 yen ($0.68) @ 1 piece.

2. Because of difficulty to measure EMC immunity.
3. Avoiding unknown risks by no deviating significantly from the original design.

maekawa wrote on 2022-12-26, 17:49:

Finally, I would like to ask about the so-called polymod. I think the consensus in Vogons is that it's not recommended to recap […]
Show full quote

Finally, I would like to ask about the so-called polymod.
I think the consensus in Vogons is that it's not recommended to recap the P3 capacitors to polymer capacitors.

Why is this?

My guess:
1: Expensive.
At Mouser it's 192 yen ($1.44) @ 100 pieces, but at Akizuki Denshi, a local store in Japan, it's 90 yen ($0.68) @ 1 piece.

yes, 20x more expensive than normal low ESR caps that will outlive your passion for retro gaming 😀

https://www.tme.com/us/en-us/katalog/tht-elec … 1632190;&page=1

In the datasheet / application notes for a typical integrated step down regulator, the LM1576, you can find some information:

As for output filtering in general larger capcities, higher nominal voltage and lower ESR are reducing ripple, but there´s a limit where instabilities might occur.

I do the polymods using capacitors rescued from defective or useless motherboards, I'm not an expert in anything so I can't give many explanations, I just followed the explanations of a member of this forum and a few hours on YouTube to reach the conclusion that this way of doing it could work.
Although it is curious, this motherboard works correctly with an Athlon 1ghz but with a 133mhz bus, that is, 10x133 and 1.7v in the mosfets and an original AMD heatsink without any temperature problem or instability.
It's just a curious example.

Could you show me its waveform on an oscilloscope?

maekawa wrote on 2022-12-28, 02:30:

Could you show me its waveform on an oscilloscope?

You wouldn't know how much I would like to do that, but, I can't afford an oscilloscope so the only reference I have is the stability and temperature of the components.
I have used this method several times without problems and the only strange thing that I have noticed so far is that the voltage supplied to the CPU measured in the Mosfets rises slightly when changing the electrolytic capacitors for Polymers, +0.050mv of difference according to my multimeter, on this board and on an A7N8X-X.
As I said, I only show it as a curiosity.
I believe that there may be a distance between theory and practice and that when theory is put into practice, the expected result is not always obtained.
I am curious to know why in a modern motherboard eight 820uf polymer capacitors can keep the voltage stable for a processor that works at a maximum of 1.5v and that can consume more than 100w with much higher consumption peaks, and because six identical capacitors could not do the same with a processor that works at a minimum of 1,650v and has a maximum consumption of 70w.
I also added behind the motherboard in parallel to each replaced capacitor a Mlcc capacitor to further reduce ESR, used the ones usually found behind or inside the CPU socket, Donor Board was a GA- N650SLI that thanks to the nvidia chipset was destined for the trash.
As I said, I do not intend to say that this can always be done or that it is safe or that is perfect, I am just showing it as a curiosity and showing that it can be done.