Repair & restoration
Repair and restoration — power supplies, capacitors, keyboard membranes, dead machines, diagnostics, recapping.
183 threads · 1385 messages · 2002-11-12 → 2026-06-03
Repair and restoration is one of the most active threads on the list across both eras, but the cast and tooling changed dramatically between Yahoo and groups.io. In the Yahoo era (2001–2019) the resident diagnostician was Eduardo Fuentes in Argentina, and the entire fault tree the group still uses for a dead 2068 was first laid out by him: suspect a faulty UA78L12 regulator near the video modulator or a bad LM1889 for missing video; piggy-back a known-good 4416 onto U12/U13 (8000–BFFF) or U17/U18 (C000–FFFF) to isolate bad upper RAM; turn the machine on and feel each lower-RAM chip with a fingertip — the hot one is the culprit. The 2013 "TS 2068 Video Problem" thread already had members measuring 19.34V on the supposedly +15V expansion-bus pin with a Fluke and asking on-list whether that was normal. The other defining Yahoo-era contribution came from vbtinfo (Vsevolod) in 2008–2009: scoping the +5V rail he found small switching pulses leaking from the internal SMPS into U3 (the SCLD), proved it was a design flaw rather than a fault, and published the mod that adds a Radio Shack 7805 to power the SCLD directly off the pad near pin 43 (lifting the original jumper to C67). That mod is the direct ancestor of every modern "clean up the 2068 video" recipe.
In the groups.io era (2020–present) the bench work moved to Carl Miles in New Mexico, who functions as the group's de-facto repair shop — diagnosing and reviving dead machines mailed in by Ryan Gray, Ian, Randy Kindig, the Bill Miller collection and others, and posting step-by-step narratives. Eduardo Fuentes still appears (he gives the touch-the-RAM tip again in the 2023 "Toasted TS2068" thread, 15 years after he first wrote it on Yahoo), and Carl's fault-tree for a machine that boots to bars but runs no code — bad lower 16K RAM, dead crystal, CPU, SCLD, ROM/EXROM, or shorted address/data bus — is the same one Eduardo published, just expanded. The canonical modern example is Carl replacing all six memory chips and the Z80 in Ryan's original 1983 2068 (thread 5242), and then rebuilding Ryan's 30-year-old dead donor unit into a working test mule. Recurring diagnostic wisdom that survived from Yahoo: piggy-back-then-replace for RAM, watch for invisibly bent socket pins (Mike Druckenmiller), and use a scope that can lock on a specific scan line for video work (David Anderson).
Power supplies are the single most-discussed failure mode in both eras. The Yahoo archive carries the reprinted UPDATE Jan. '89 warning blaming voltage-regulator failures for roughly 150,000 abandoned TS-2068s, and the groups.io consensus is built on top of it: move to a modern switching brick, replace the original 5V switching regulator with an R-78E5 drop-in (which also kills the audio hiss Vsevolod first traced to SMPS noise), and use Jameco #191652 (15VDC 1A) as the canonical replacement adapter. A live caution thread across both eras: peripherals like the ZX Printer, Interface 1 and Microdrives pull the unregulated 9V line directly, and the internal DC-DC converter derives +12V/-12V/-5V from it, so a higher unloaded modern adapter voltage matters. The TS2040 thermal printer is its own case — 24V AC at ~1.2A, with print darkness tracking line voltage (RichJ uses a variac to push to 125 VAC for dark prints). The TS1500/TS1000 ULA story closes a long-running gap: when a ZX81-family ULA dies, the modern answer is the VLA81 from vretrodesign, and snapped/delaminated mylar keyboards get replaced with new membranes from RWAP Software and ZX Renew.
On recapping the group reached an evidence-based position the Yahoo era never had the tools for: Carl's LCR-meter survey of Adam's 40-year-old Samyoung electrolytics found them essentially all good (only C41 and C60 marginal), so the modern rule is test before you replace, and the cap spreadsheet was deliberately archived as a reference. The same meter then caught a genuinely failing C20 in Mowgli's TS1500 — proving the rule cuts both ways. Still-open hard problems carried forward: sourcing the custom TS2068 SCLD (no modern equivalent exists; a CATS newsletter lead on ~100 spares is the current best hope), the lack of a dedicated 2068 diagnostic cartridge (Brendan Alford's testrom.bin and a proposed TS-Pico autoboot diag are the workarounds), and the open question raised by Josef Jelinek's "Any TS2068 PCB updates?" thread of whether a faithful 2068 clone (Claudius's WiP board, Don Dindang's Harlequin-2048) should match the TC2048/2068 /INT timing — shifted 15 T-states from Spectrum timing, which is what makes top-border demos crash on real Portuguese hardware.
Key threads
- Video Blurry; was: TS 2068 - Sound/Hissing Problem (archive.org) — Yahoo-era origin (2008-2009) of the modern video-cleanup mods: vbtinfo scopes the +5V rail, finds SMPS switching pulses leaking into the SCLD, and publishes the 7805-direct-to-U3 fix that everyone still references.
- TS 2068 Video Problem (archive.org) — Yahoo-era (2013) statement of the dead-video fault tree — UA78L12 regulator, LM1889, then 4416 upper RAM piggyback testing — by Eduardo Fuentes, the diagnostic method Carl Miles still teaches in 2023+.
- TS2068 Repairs Dec 2023 -- Ian's Machine and Ryan's two TS2068 machines (#5242 on groups.io) — Carl Miles's narrated bench log of reviving multiple 2068s in the modern era — socketing/swapping all six RAMs and the Z80 — the closest thing to a current 2068 troubleshooting tutorial.
- My Alternate T/S 2068 Computer - Recapping? (#7677 on groups.io) — Carl's LCR-meter measurements show 40-year-old Samyoung caps are mostly fine, establishing the modern 'test before recapping' consensus the Yahoo era never had the tools to reach.
- T/S 2068 Power Supply Replacement (#773 on groups.io) — Definitive modern PSU answer: Jameco #191652 (15V 1A), upgrading internal regulators, and the R-78E5 drop-in that ends the 5V switching-regulator hiss Vsevolod first identified in 2008.
- Toasted TS2068 (#4275 on groups.io) — Modern restatement of the bars-on-screen fault tree (RAM/crystal/CPU/SCLD/ROM/bus short) — and Eduardo Fuentes himself reappearing 15 years after his Yahoo posts with the same touch-the-RAM tip.
- A sad day... (#23 on groups.io) — Origin of Carl Miles as the groups.io-era repairman — the first 'dead 2068' plea from a Yahoo-list lurker that established Carl's standing offer to fix machines mailed to him.
- Slightly off topic… TS1000 Help (#6651 on groups.io) — Modern ZX81-family repair playbook: VLA81 modern ULA replacement, RAM expansion gotchas (Memotech "ZX81 killer"), and how to tell whether a silent TS1000 is actually alive.
- TS1000 keyboard repair options? (#1956 on groups.io) — Definitive answer on snapped/delaminated keyboard mylar across both eras: RWAP Software and ZX Renew sell new ZX81-compatible membranes that fit the TS1000.
- Any TS 2068 main PCB updates? (#6932 on groups.io) — Hardware-fidelity thread: Claudius's WiP replacement-board schematic and Don Dindang's Harlequin-2048 clone surface the /INT-timing decision (Spectrum vs real TC2048/2068, 15 T-states apart) for any faithful 2068 clone.
All threads (183)
↳ marks a thread where this topic begins mid-conversation — the link jumps to the message where it starts.