A complete replacement capacitor set for a Roland Juno-106 costs 64 euros. The labour to fit it costs between six and twenty times that, and most of that labour is not soldering. It is disassembly, cleaning, calibration, and the two hours spent finding out the fault was never a capacitor.
That gap is the subject. Recapping is cheap in parts and expensive in hands, which is exactly the shape of a job that gets sold when it is not needed and skipped when it is.
When does a vintage synth need recapping, honestly
Almost never as routine. Frequently as a repair. Those are different jobs and only one is worth paying for.
A blanket recap, every electrolytic on every board replaced because the machine is old, is the version sold to nervous buyers. A targeted recap, the power supply reservoir and regulator capacitors plus anything measured out of spec, is what a tech does. On a forty-year-old polysynth it takes an hour or two and fixes most of what the blanket job would have.
Which capacitors actually degrade, and which do not
Aluminium electrolytics: the ones that genuinely wear out
The only capacitors in most gear with a published wear-out life, because they are the only ones with a liquid in them. Nippon Chemi-Con's failure-mode document is blunt: the electrolyte "has gradually evaporated and diffused out of the capacitors through the rubber seal materials with time, which leads to decrease in the capacitance and/or increase tanδ". Capacitance falls, ESR rises, leakage climbs. Chemi-Con confirms the rule every tech quotes, that life roughly halves for each 10°C rise in ambient temperature, and caps its own lifetime estimates at fifteen years.
Fifteen years. Every electrolytic in your Juno-60 is more than twice past the number its own manufacturer would calculate to. That is the honest case for servicing, and it is about the power supply, not about tone.
Tantalums: they do not drift, they short
Solid tantalums fail the other way: low insulation resistance or a dead short, from crystalline defects growing in the tantalum oxide until the dielectric cracks. They do not gently lose value. They go from fine to short, often taking a regulator with them. So "it worked last week and now a rail is down" points at a tantalum, while "it has got hummier over five years" points at an electrolytic. Same board, opposite diagnostics.
Film and ceramic: mostly leave them alone
This is the part the blanket-recap pitch skips. Film capacitors, the polyester and polypropylene parts used for oscillator timing, filter poles and coupling, have no electrolyte to lose and routinely outlast electrolytics by decades. Replacing them wholesale on a working machine is not maintenance, it is changing the circuit and hoping.
One loud exception, and it is a safety part. Metallised paper X2 mains suppression capacitors, the translucent yellow RIFA parts across the mains in a great deal of 1970s to 1990s European gear, fail with what Hackaday calls "a bang, brown goo and a nasty stink that hangs around for a week". The mechanism is contested, moisture ingress against plasticiser breakdown, but it does not matter much: if you open a Revox PR-99 or a Copicat and see one, replace it before you plug the thing in again.
The bad batches, and how to date them
Two episodes produce capacitors that fail early rather than late, and both have dates you can check a serial number against. The capacitor plague covers parts made roughly 1999 to 2007. The familiar story is that a materials scientist stole Rubycon's water-based electrolyte formula for its ZA and ZL series, defectors copied an incomplete version, and Taiwanese makers sold the result cheap. Wikipedia supports the chemistry: the bad electrolyte was alkaline at pH 7 to 8 where Japanese parts sat near pH 4, phosphate ions were missing, and without a depolariser the capacitors made hydrogen and bulged or vented. Hackaday's 2025 piece doubts one stolen recipe explains the scale, pointing at low-ESR designs, hotter boards and counterfeits arriving together. Both can be true. Practically: anything digital from that window, including the 2000s racks on your Akai sampler shortlist, gets its electrolytics checked first.
The earlier problem is less famous and worse. Surface-mount electrolytics from the late 1980s and early 1990s leak invisibly under their own bodies and eat the traces beneath. No bulging tops to spot: a faint brown halo around an SMD can means the damage is already done.
Does recapping change the sound?
Sometimes. Rarely the way it is sold.
Coupling capacitors sit in series with the audio and set a high-pass corner with whatever resistance follows. Lose half the capacitance and that corner moves up: you lose low end. Real, measurable, worth fixing. Go leaky and DC creeps into the next stage. Also real.
What is not real is that fitting better coupling capacitors makes a machine sound better. Ethan Winer measured 24 capacitors across four types and six values for audioXpress in May 2020 and found film, aluminium and tantalum all sitting near the residual distortion of his test rig at audible frequencies. Only ceramics misbehaved. Above the roll-off point, type stops mattering.
Power supply capacitors are not in the audio path at all. They are a reservoir. When they fail you get hum at twice mains frequency, rail sag under load, instability, motorboating, and on a polysynth, voices misbehaving together rather than one at a time. You do not get "dullness". Anyone telling you a tired supply is making your Xpander sound veiled is describing a symptom that does not follow from the mechanism.
Signs your synth needs new capacitors
| Symptom | Most likely cause | Is it a recap? |
|---|---|---|
| Constant hum at 100Hz or 120Hz, unaffected by volume | Reservoir capacitors, high ESR | Yes, targeted |
| Slow putt-putt oscillation (motorboating) | Decoupling or reservoir failure, shared rail | Usually |
| A rail dead, fuse intact, no smoke | Shorted tantalum or failed regulator | No, a part replacement |
| Low end gone on one channel | Coupling capacitor value drop | Yes, targeted |
| One voice dead or crackling as it warms | Voice chip or module | No |
Two things people call a recap that are not one
First, the Juno-106's 80017A voice modules: potted hybrids carrying the filter and VCA circuitry. When they fail you get a dead voice or one that crackles more as the synth warms. Not a capacitor fault, and a recap will not touch it. Sources disagree on the mechanism. polynominal.com blames the resin coating degrading under the module's own heat; SynthTweaks says the resin is a symptom and the real fault is dry solder joints underneath, which is why the acetone trick works for a while and then stops. Analogue Renaissance sells a replacement at 46 euros each. More in the Juno-60 versus Juno-106 comparison.
Second, and more urgently, the battery.
The battery is the emergency, not the capacitors
Battery-backed patch memory was standard from the early 1980s, in two flavours. A CR2032 lithium coin cell is the benign one: it dies, you lose your patches, you fit a new one. Attack Magazine names the Yamaha DX7, Korg Poly-61 and several Rolands, including the Juno-106, JX-3P and Alpha Juno-2, as ones to check on arrival.
The dangerous one is a soldered-in rechargeable nickel-cadmium cell, permanently under charge, for forty years. The Korg Polysix is the famous case. When that cell lets go, the electrolyte spreads across the KLM-367 board and corrodes the large ground trace and the plated-through holes, which as Old Crow's repair guide notes "are the first thing to fail". The fix means cutting out dead traces, drilling out ruined vias and soldering in wire. The Prophet-600 is in the same class.
A leaking battery is the only fault on this page that gets worse while the machine sits in a cupboard doing nothing. If you own an early-80s polysynth and have never had the lid off, open it this week. Not for the capacitors. For the battery.
Vintage synth repair cost: what a bench actually charges
Published rates, gathered September 2026, currencies as each shop states them.
| Shop | Territory | Rate | Minimum or assessment |
|---|---|---|---|
| Bothners | South Africa | R360 per half hour incl VAT (R720/hr) | R450 assessment |
| Studiocare | UK | £90/hr ex VAT | £90 assessment, waived on repair |
| dBm, Synth Tailor, Retro Pro Audio | US | USD 150/hr | 1hr min, up to USD 300 prepaid on vintage |
| Rosen Sound | US | USD 200/hr from 1 September 2026 | 1hr minimum |
The South African rate is the interesting one. At the 16.10 USD/ZAR quoted by Trading Economics on 11 September 2026, R720 an hour is about USD 45. Local labour is cheap by world standards and local parts are not, because parts land at import prices with freight and VAT on top. That inverts the arithmetic you would do in London.
How much to service a Juno 106
Two US shops publish the same flat figure. Bell Tone Synth Works charges USD 600 for a Juno-106 Essential Service: a full set of Analogue Renaissance voice chips, the battery replaced with patches backed up and restored, power supply filter capacitors, solder reflow on the heat-sinked parts, and full calibration. Retro Pro Audio lists the same at USD 600. Front panel work is separate, USD 200.
Note what is in and what is not. Voice chips, battery, power supply filters, yes. A blanket recap, no, because the people doing this for a living do not think it necessary.
The hidden costs of buying vintage gear, with the arithmetic
Worked on a Juno-106. Classifieds listing, R8,000, working with two voices that crackle. The purchase price is the smallest number here.
| Line | Rand | Where the number comes from |
|---|---|---|
| Purchase | 8,000 | The listing |
| Courier, 20kg boxed and insured | 400 to 700 | Published SA courier rates, R12 to R25/kg |
| Six AR80017A voice chips, landed | about 4,600 | EUR 245 at 18.71 EUR/ZAR, pre-duty |
| Capacitor set and battery | about 1,400 | EUR 64 set plus sundries |
| Bench, 8 hours | 5,760 | R720/hr incl VAT |
| Running total | about 20,400 | |
| Second visit: sliders, switches, the fault behind the first fault | 2,500 to 4,000 | Inferred, not published |
| Realistic all-in | 23,000 to 24,500 |
The second visit is not pessimism, it is normal: eight hours on a bench uncovers faults invisible until the first three were cleared. Then the resale line, which nobody includes. A serviced machine with a written bench report sells at the serviced price, because the next buyer inherits the work. A half-fixed one sells at classifieds price, because the next buyer assumes those eight hours are still ahead. That gap is usually larger than the service cost.
When the cheap one wins, because sometimes it does
- It genuinely only needs a battery and a clean. An hour of bench time and a R40 cell. If the seller shows you every voice sounding and every slider working, the downside is small.
- You do the work and your time is free. With a scope, a meter, an ESR tester and the service manual, the whole calculation collapses to parts. That is a real answer and we will not pretend otherwise.
- The machine needs no calibration, and a clone chip exists. A TR-808 is largely working or not; a polysynth with six voice cards to trim is not. Where a modern replacement part exists, the worst case has a ceiling.
Is recapping a synth worth it, and when is the repair not?
A recap on its own, rarely. A recap inside a service that also does the battery, the calibration and the actual fault, usually. And three tests end the job entirely. If working market value is close to the shop minimum, walk away: something worth R2,500 working cannot absorb a R450 assessment plus an hour minimum plus parts. If a custom LSI has failed and nobody makes a replacement, it is a parts donor. If battery electrolyte has spread across a large multi-layer board, the repair runs to tens of hours for something still worth only what the market says. We say no to all three regularly.
Buying from a private seller with no bench in sight
You will not get a proper assessment out of a Marketplace meeting. Ten minutes gets you most of the way.
- Insist it is powered on before you arrive, warm for twenty minutes. Faults that only show warm are the expensive ones.
- Smell it. Fishy means leaking electrolyte, acrid means something has already cooked.
- Get photos of the inside before you travel. You are looking for a battery and a stain around it.
- Play every voice on its own, one note at a time, all the way up. A held chord hides a dead voice.
- Move every control, then listen at idle. Hum tells you about the power supply for free.
- Price the worst case. Offer what it is worth if the three likeliest faults are all present. If that offends the seller, you have learned something.
The same logic, plus motors and heads, applies to buying a used tape machine, where a worn record head ends the conversation.
What we could not verify
The stolen-formula account of the capacitor plague is partly corroborated by chemistry but was never tested in court, and a serious source argues several causes acted at once. We reported the dispute rather than picking a side. The 80017A mechanism is contested between resin degradation and dry joints beneath the resin, and we named both. RIFA is the same. No published South African rate card for specialist synth work exists, so the rand figures use a national retailer's general electronics rate, which may understate specialist labour. The second-visit and resale lines are inferred, not published.
Sources
- Nippon Chemi-Con, Capacitor Failure Modes
- Wikipedia, Capacitor plague
- Hackaday, The Capacitor Plague Of The Early 2000s
- Hackaday, Why Do RIFA Capacitors Fail?
- audioXpress, Stop Worrying About Coupling Capacitors
- Polynominal, Roland 80017A
- SynthTweaks, Juno 106 80017a Chip Repair
- Old Crow's Synth Shop, Polysix Battery Repair
- Attack Magazine, Internal Batteries in Vintage Gear
- Bell Tone, Juno-106 Service Packages
- Studiocare, Outboard Repair
- Bothners, Service and Repairs
- synth-parts.com, Juno 106 Recapping Set
- UrgentGo, SA Courier Prices 2026
- Trading Economics, South African Rand
Everything here goes through that arithmetic before it goes on the site, which is why the S950, the Otari MX-5050, the Yamaha CA-2000 and the dbx 160 VU pair cost what they cost rather than what a classifieds listing costs. It is also why some things came through the door and went straight back out. For the same decision applied to a compressor, the dbx 160 decoder says which are worth a bench hour.