Low-side capacitor production: has it changed how you choose recap values?

[EchoForge]

There have already been several discussions here about modern capacitors being manufactured toward the low side of their tolerance range, including Blue Shadow’s thread on large electrolytics.
I’m curious about the practical consequence for recap work.

In my own measurements, this is becoming difficult to ignore. 2025 Examples include:

• Nichicon LLS 10,000 µF: 8,440 µF average (-15.6%)
• CDE 380LX 10,000 µF: 8,635 µF (-13.7%)
• Chemi-Con SMH 10,000 µF: 8,630 µF (-13.7%)
• Rubycon TXW 100 µF: 82.55 µF (-17.5%)
• Nichicon UKW 100 µF: 89.73 µF (-10.3%)
• Panasonic FR-A 47 µF: 43.0 µF (-8.5%)

I’m also seeing the same tendency with film capacitors. A recent batch of 2.2 nF ±5% parts was heavily concentrated around 2.108 nF (-4.2%).

What interests me is that the parts within a lot are often very tightly matched. Whether I measure a few pieces or dozens, the spread can be small while the whole lot is centered toward the lower side of the allowed tolerance. So has this changed the way you actually recap equipment?

For example, if the original circuit specifies 10,000 µF and current production consistently measures around 8,500–9,000 µF, do you still order 10,000 µF, hand-select closer parts, or move to the next standard value?
And what about smaller capacitors used in coupling, bypass, tone, loudness or filter networks? Do you treat those differently and stay closer to the original nominal value?

I’m especially interested in actual shop practice from people doing a lot of recap work today, rather than whether a capacitor is technically “within tolerance.”

 

[ConradH]

Yes, I often go up a value.

 

[gadget73]

Same, particularly with large supply caps. I often end up bumping them to get something that fits the lead spacing or clamps anyway. Having them clock in on the low end of tolerance means for all practical purposes I'm not actually increasing the values by significant amounts.

 

[daybreak2350]

Good question and one only the manufacturers/engineers could explain but I'm not sure how you'd get it out of them. In PS applications I understand it's OK to go up--like those big boys the size of a soup can and all. But I wonder how much an out-of-spec replacement cap in a tuned circuit would affect performance of a unit. I think I'll be checking replacement caps more carefully now and get more use out of those expensive capacitance/ESR meters!

 

[pac1085]

What are your test conditions, which LCR meter and what frequency? Are you testing them at 120 Hz?

 

[EchoForge]

I’m using a DER EE DE-5000, measuring the electrolytics at 120 Hz in Cs mode. For example, this 22,000 µF / 63 V capacitor measures 18.91 mF (18,910 µF), with D = 0.004.

 

[EchoForge]

Same, particularly with large supply caps. I often end up bumping them to get something that fits the lead spacing or clamps anyway. Having them clock in on the low end of tolerance means for all practical purposes I'm not actually increasing the values by significant amounts.

That distinction is exactly what I’m trying to pin down. With large power-supply filter caps, going up one nominal value often seems harmless or even practical. But I’m more concerned about doing the same thing in circuits where the capacitor value directly sets the frequency response or a time constant.
For example : RIAA equalization networks, tone-control networks, filters, coupling networks, etc. In those cases, if a modern capacitor consistently measures 10–15% below nominal, do you still go up one value, or do you select parts to get as close as possible to the original intended capacitance?

 

[pac1085]

I wouldn't worry too much about those circuits since they usually use other types of caps (like polystyrene, film or ceramic) in the critical positions. Electrolytics were never very tight tolerance so the circuits seem to be designed around that. Plus these new caps, even reading on the low side are still technically within tolerance.

I've done quite a few restorations using modern caps without any issues. I test the devices on an audio analyzer (Rohde and Schwarz UPL) after the work is done and validate that everything meets spec, including the various filters, rial equalization, etc.

 

[Warren-Oates]

Signal path capacitors, small values .47 to 22, tend to measure closer to the label. Film caps are usually very close that I have tested,

As you have noted the larger electrolytics come in at below the label, within spec though. As another noted going up a notch in UF often will get you you the desired lead spacing. For example a 63V 100uf cap with 7.5mm lead spacing had no results for me at Digikey. However going to a 120uf (which test near 100) cap I could get exactly what I wanted.

Sometimes the oddball values are quite cheap too, as they don't move as fast maybe???

If you are looking for drop in caps, with slightly higher uf ratings (for power reservoir duty), at digikey or wherever, explore the oddball uf ratings. Such as; 120, 150, 180, 270, 390, 560, 680, 820. You will find many great selections such as Nichicon UPW, UPM, HEM. As noted changing uf in locations that are not stores of power is a bad idea unless you do the math and understand what you are doing, which I'm not smart enough so I don't.

Just FYI another variable that will help you find the correct lead spacing is temp rating. You will find the 125-135c rated caps often have the wider lead spacing.

 

[EchoForge]

I wouldn't worry too much about those circuits since they usually use other types of caps (like polystyrene, film or ceramic) in the critical positions. Electrolytics were never very tight tolerance so the circuits seem to be designed around that. Plus these new caps, even reading on the low side are still technically within tolerance.

I've done quite a few restorations using modern caps without any issues. I test the devices on an audio analyzer (Rohde and Schwarz UPL) after the work is done and validate that everything meets spec, including the various filters, rial equalization, etc.

That makes sense for checking the RIAA curve. But have you ever tested whether the actual capacitance value makes any measurable difference to the noise performance?
For example, comparing two electrolytics of the same nominal value and series, one measuring near the low end of tolerance and the other near the high end, in the exact same circuit, then comparing the noise floor or SNR ?

 

[whoaru99]

Good question and one only the manufacturers/engineers could explain but I'm not sure how you'd get it out of them. In PS applications I understand it's OK to go up--like those big boys the size of a soup can and all. But I wonder how much an out-of-spec replacement cap in a tuned circuit would affect performance of a unit. I think I'll be checking replacement caps more carefully now and get more use out of those expensive capacitance/ESR meters!

Wouldn't one presume if the circuit was designed with +/- X% capacitors, as long as the capacitors fall within that +/- % spec the circuit should work as expected. And wouldn't be out of spec if the value was within the spec.

I mean if a value was critical to a circuit, e.g. your tuned example, one probably wouldn't be spec'ing a +/- 20% value component.

 

[whoaru99]

It would be an interesting challenge though in context of the consistency of measured low values.

"Manufacturing tolerances" as a reason doesn't really hold water. If it was that wouldn't one expect a somewhat equal distribution on the + side and - side, rather than apparently consistently just low.

 

[EchoForge]

Wouldn't one presume if the circuit was designed with +/- X% capacitors, as long as the capacitors fall within that +/- % spec the circuit should work as expected. And wouldn't be out of spec if the value was within the spec.

I mean if a value was critical to a circuit, e.g. your tuned example, one probably wouldn't be spec'ing a +/- 20% value component.

That makes sense when replacing a single capacitor in a circuit designed around a ±20% part. But what about a full recap, where most or all of the replacement electrolytics happen to be consistently below nominal?

Each individual capacitor is still within spec, but now the whole population may be biased toward the low side rather than randomly distributed around nominal. Would you still expect that to be equivalent to the tolerance variation the original circuit was designed around?

 

[whoaru99]

That makes sense when replacing a single capacitor in a circuit designed around a ±20% part. But what about a full recap, where most or all of the replacement electrolytics happen to be consistently below nominal?

Each individual capacitor is still within spec, but now the whole population may be biased toward the low side rather than randomly distributed around nominal. Would you still expect that to be equivalent to the tolerance variation the original circuit was designed around?

Why not? If, for example, all +/- 20% tolerance was spec'ed originally, isn't that what an engineer determined was close enough?

To me it would be questionable engineering to spec +/- 20% parts yet count on them to be all close to nominal value or all balance out in the end. If the design can't tolerate all -20% part maybe the tolerance spec should have been +/-10% or 5% or whatever to satisfy the overall intent.

 

[EchoForge]

Why not? If, for example, all +/- 20% tolerance was spec'ed originally, isn't that what an engineer determined was close enough?

To me it would be questionable engineering to spec +/- 20% parts yet count on them to be all close to nominal value or all balance out in the end. If the design can't tolerate all -20% part maybe the tolerance spec should have been +/-10% or 5% or whatever to satisfy the overall intent.

Could that actually be part of the reason manufacturers specified performance as a worst-case limit rather than an exact figure? In other words, could two units populated near opposite ends of the allowed component tolerances show measurably different performance, while both still comfortably meet the published spec?

 

[whoaru99]

It seems reasonable, to me anyway, to conclude a design has to tolerate the variability of all components, high, low, or nominal, to meet the design intent.

 

[EchoForge]

It seems reasonable, to me anyway, to conclude a design has to tolerate the variability of all components, high, low, or nominal, to meet the design intent.

I wonder if there’s another aspect to this. Did the engineers actually choose ±20% because that was the tolerance they considered appropriate, or were they simply designing around the tolerances the capacitor industry could economically provide at the time?

If it was largely the latter, that could also help explain why equipment specifications were given as guaranteed limits rather than exact expected performance. The design had to meet those limits despite the component variation available in production.

 

[whoaru99]

I wonder if there’s another aspect to this. Did the engineers actually choose ±20% because that was the tolerance they considered appropriate, or were they simply designing around the tolerances the capacitor industry could economically provide at the time?

In the case of the latter, isn't that really a different matter albeit doesn't conceptually negate the other?

Whether price or performance or availability, it still has to work as expected/to design intent with those tolerances.

 

[EchoForge]

In the case of the latter, isn't that really a different matter albeit doesn't conceptually negate the other?

Whether price or performance or availability, it still has to work as expected/to design intent with those tolerances.

From the standpoint of simply restoring the unit so that it meets its original specifications, I think your reasoning is very pragmatic.

Where I’m more curious is in the measurable performance differences between units, and in what we consider the objective of a restoration. Is the goal simply to bring the unit back within its original specifications, or is there also value in improving measurable performance where possible, while preserving the original character and intent of the design as much as possible?

And in that context, what is the real impact of the apparent tendency for modern electrolytics to measure consistently below their nominal value? Are those differences large enough to produce a meaningful change in the final measured performance after a full recap?

 

[whoaru99]

Certainly a simple before/after comparison of a full recap doesn't catch the essence of that point because it's old caps vs new caps in addition to whatever else.

Fully recap a unit with new parts, carefully selected around nominal values, test it, then recap it again with parts that all fall to the low side then test it again. I'd suspect at least some measurements could show differently. Whether is of anything of merit more than different on paper, so to speak, is the age old question.

But, my interest fades pretty quickly in a bunch of what-ifs that stand little chance of being rigorously implemented and tested, so...