Find the actual failing part, not a guess

A computer that will not start is rarely silent. Desktops with a speaker header still produce classic patterns, one long and two short beeps for a video fault on AMI and Award-derived firmware, repeating short beeps for memory. Business laptops replaced beeps with blink codes: Dell uses an amber and white power LED sequence where the counts map to a specific subsystem, HP uses caps lock or power LED blinks in groups, Lenovo signals through a combination of LED behavior and its own diagnostics. Reading the pattern correctly before opening anything narrows the field to one subsystem in about two minutes, and it is astonishing how often it gets skipped.

Memory, and why one pass is not a test

Bad RAM is the most misdiagnosed fault in computing, because it produces symptoms that look like everything else: application crashes, file corruption, blue screens with different stop codes each time, and installers that fail at random points. A quick memory check that passes proves almost nothing. We run MemTest86 or an equivalent for multiple complete passes, usually overnight, because marginal cells frequently only fail on specific test patterns or after the modules have reached full operating temperature.

When errors do appear, we isolate by removing modules and testing individually in a known-good slot, which distinguishes a bad stick from a bad slot or a damaged memory controller. And we check memory configuration before condemning the hardware: an XMP or EXPO profile running at rated speed is technically an overclock, and a kit that is stable at its default JEDEC speed but crashes at its advertised speed is an instability problem, not a dead module.

Heat and power, the two quiet killers

Thermal faults do not usually announce themselves as heat. They present as a machine that works fine for twenty minutes and then stutters, drops frame rates, or shuts down under sustained load. We log core temperatures, package power and clock speeds during a controlled stress run and look at the delta over ambient rather than the absolute number, because a workstation in a warm Chandler office in July has a very different baseline from the same machine in February. Pump-out of thermal compound after three or four years, a heatsink matted with fine Arizona dust, a failed fan bearing and a heat pipe that has lost its working fluid all produce the same throttling curve and need different fixes.

Power supplies are the other reliable source of mystery behavior. A failing PSU rarely dies cleanly. It sags under transient load, so the machine reboots when the GPU ramps or a drive spins up, and everything else gets blamed first. Shorting the green wire to ground proves only that the supply turns on, which is nearly worthless as a test. We load-test properly and check the rails under demand. Bulging or vented capacitors, a fan that has stopped, or a supply that smells faintly of fish are all condemnations on sight. In a small business this matters more than people expect, because an unstable supply in a server can corrupt data long before it fails outright.

Storage gets read at the source rather than guessed at. SMART attributes tell us reallocated and pending sector counts on mechanical drives, and percentage used, available spare and media wear on SSDs and NVMe. A drive throwing pending sectors is not a repair candidate, it is a data preservation job, and the correct next step is imaging it before anything else, not running a repair utility that hammers a dying disk with retries. Where a drive is already failing badly, our data recovery work takes over.

That leaves the board. Motherboard faults split into two categories with very different economics. Component-level failures, a blown MOSFET in the VRM, bulging capacitors, a burnt trace from liquid ingress, a damaged DC input circuit, are often genuinely repairable on the bench, and on a laptop where the board carries a soldered CPU and RAM, a component repair can be a fraction of the cost of a replacement board. Failures involving a BGA package under the chip itself are a different story. Reflowing a ball grid array is a temporary trick that fails again in weeks or months, and we will not sell it as a repair. Where a board is genuinely finished, we say so.

Then comes the arithmetic. Our rule of thumb is straightforward: if the repair cost approaches half the replacement cost of an equivalent current machine, and the machine is more than three or four years old, replacement is usually the better spend, because you are buying remaining life as well as function. A younger machine with a single failed component almost always deserves the repair. Business machines shift the math further toward fast resolution, since downtime and technician hours outweigh the parts difference. Our free fix or replace calculator runs the same numbers we do, and the upgrades page covers the cases where the right answer is neither.

Swapping parts until it works is not diagnosis. It is shopping.Orca IT, Gilbert AZ