Tom Verbeure found a Tektronix TDS7104 at the Silicon Valley flea market. The seller wanted $700, then dropped to $300. For a 1 GHz, four-channel oscilloscope with 10 GS/s sampling, that is a good price if you can get it running, and the scope booted fine at first, then died at the BIOS screen with a CMOS battery error a few hours later. His write-up of the full repair is worth reading for anyone who owns aging test gear, because almost every failure he hit is a known failure mode for this entire series.
What a TDS7104 actually is
Introduced in mid-2000, the TDS7104 is a Windows PC wearing an oscilloscope costume. A 4-channel scope with 1 GHz bandwidth and up to 10 GS/s on a single channel (dropping to 2.5 GS/s when all four channels are live). Even next to current hobbyist gear from Rigol or Siglent, the specs hold up, though you pay for it in bulk: 39 pounds and a chassis the size of a small microwave. The little brother, the TDS7054, has 500 MHz in the same box.
Inside there are three computers. The PC motherboard is an NLX-form-factor RadiSys board with a Socket 370 Celeron on a 100 MHz front-side bus, originally a 550 MHz part. A separate PowerPC board running VxWorks handles acquisition control, and the acquisition board itself is the third system. The 640×480 LCD has resistive touch, which matters because you can drive the whole advanced feature set without hunting for a mouse. Unlike the higher-end TDS7xxx models, it uses plain BNC connectors instead of Tektronix’s custom interface, which matters a lot when replacement probes cost more than the scope does today.
The four ways these scopes die
Community repair threads on EEVblog have converged on four common failures: the PC motherboard’s CR2032 backup battery, the PowerPC board’s backup battery, a dead hard drive, and leaking power supply capacitors. Verbeure got lucky and only dealt with the first and third. The second one is the nasty one. A dead PowerPC battery means disassembling further, replacing the battery top on a Dallas DS9034 NVRAM, connecting over RS-232, and hand-entering values to reprogram it. The exit-failure GitHub repo documents the full NVRAM procedure, and the Feedbackloop repair video shows an FRAM swap that removes the battery from the equation entirely.
Image the drive first. Always.
The most transferable lesson is procedural. Whether the machine boots or not, image the hard drive before touching anything else. In this case a 6 GB IBM Travelstar, and it turned out to be on its last legs, throwing corrupt sectors during the copy. The damage landed on a French Windows 2000 manual of all things, so the repair survived, but that was luck. The drive sled pulls out the back of the chassis without opening the case, which nobody figured out until after opening the case.
Where the repair gets weird
After replacing the CR2032, the scope booted with corrupted waveforms. The cause was the Chips & Technologies 69000 graphics card, which drives the LCD and has dedicated hardware to render waveforms in overlay mode, sitting in 256-color palette mode instead of 24-bit True Color. A small SSD swap attempt also failed in an instructive way: installing the wrong driver made Windows silently fall back to the Intel integrated GPU once the GUI started, and then refuse to use the correct driver even after it was selected, with no error anywhere. The fix was manually deleting the bad driver files.
The OS reinstall path is its own adventure. Windows 2000 Pro has no USB boot support, so the route was: build the installation in a virtual machine, burn the disk image to a USB stick, and boot the scope’s ancient hardware with the Plop Boot Manager to work around its missing USB boot capability. Then the Tektronix firmware, the TDS7104-specific display drivers, and TekFonts, followed by re-enabling the licenses that were already burned into the machine.
Why this matters beyond one scope
Most of this post generalizes to any embedded Windows machine from that era, and there are a lot of them still in service. Medical devices, industrial controllers, old ATE racks. The playbook is the same: image the storage before anything else, know which backup batteries keep state where, expect driver hell on bespoke graphics hardware, and assume the vendor’s restore media is long gone so you need a VM pipeline to rebuild it. The licensing wrinkle is worth noting too: these scopes carried software licenses (jitter analysis, USB certification) locked to the hardware, and getting them re-enabled after an OS rebuild is the difference between a working scope and a $300 paperweight.
Verbeure ends with an honest assessment: the specs still exceed hobbyist scopes, but for most real work a modern compact scope is more practical, so this one may end up displacing his Agilent 54831 rather than getting daily use. As a documented survival guide for a whole class of aging Windows-based instruments, though, the post earns its keep. It ends with a line that confirms the human behind it: every word in the post was written by a human, and after reading the SSD driver saga, nobody would doubt it.