How to test a used PC and its parts before you buy

Between buying a used machine and pricing it sits the step that most people rush: finding out what it actually is. A used PC is a box of claims until you test it. The seller’s “runs great” means nothing, your own guess from the specification means only a little, and the difference between a good buy and a bad one is usually invisible until you put the parts under load and read what the sensors say back.

Testing is also where the money is made. Half an hour of it turns “I think this works” into a benchmark score, a set of temperatures and a clean bill of health for each drive. That is the same information you need to work out what the machine is worth, to decide whether to sell it whole or part it out, and to list it without inviting a haggle. Do it once, properly, and every later decision runs off it.

This is the guide to doing it: the free tools, how to read temperatures and throttling, and, part by part, what is a cheap fix and what means walk away.

What do you test a used PC with?

You do not need to spend anything. The core toolkit for a gaming machine is five free programs, each aimed at a different part.

• 3DMark for the graphics card. The free version on Steam is all you need. Run Time Spy or Steel Nomad and you get a score you can compare against thousands of identical systems, so “in line with a stock RTX 4070” replaces “seems fast”. A loop of it is also a stress test, because a card that artifacts or crashes under one was going to do the same for a buyer in week two.
• Cinebench for the CPU. It pushes every core to full load, which is both a performance number and the fastest way to see how hot the processor gets and whether it holds its clock speed.
• HWiNFO for the sensors. This is the one you leave running underneath the others. It shows the temperature, clock speed, fan speed and power draw of every part, live, so you can watch what happens while a benchmark hammers the machine. Most of what follows is read off HWiNFO.
• CrystalDiskInfo for the drives. It reads the SMART health data out of every hard drive and SSD and tells you, in plain terms, how much life is left and whether the drive is reporting problems.
• MemTest86 for the memory. It runs from a USB stick before Windows loads and works the RAM until any faulty stick shows itself. Unstable memory causes crashes that look like a dozen other faults, so ruling it out early saves chasing ghosts.

Two more are worth adding when you want to be thorough: OCCT for combined stress testing and power-supply strain, and FurMark if you want to cook a graphics card specifically to see whether its cooling holds. Neither is essential on top of the five above.

How do you read temperatures and throttling?

Temperature is the single most useful thing testing tells you, but the number on its own is not the point. The point is throttling.

Modern CPUs and graphics cards are built to run hot. Seeing a processor touch the mid-80s Celsius under a full Cinebench load, or a graphics card sit in the low 80s during a 3DMark loop, is normal and not a fault. What you are watching for in HWiNFO is the moment the temperature climbs into the 90s or towards 100 and the clock speed starts dropping to compensate. That drop is throttling: the machine protecting itself by slowing down, and quietly losing the performance you are paying for.

Throttling is a signal, not a verdict, and where it comes from decides what it costs you to fix.

If the whole machine runs hot and throttles, the cause is almost always airflow. Dust packed into the heatsinks and filters, fans that have given up, a cramped case with no intake, or a build where the cabling strangles the airflow. All of it is cheap. A blow-out with an electric duster, a fan or two, or moving the parts into a case that actually breathes can turn a throttling machine into a quiet, fast one for very little money. A machine that throttles because it is filthy is an opportunity, not a problem, and it is one of the reasons a badly kept PC can be a good buy.

If one part runs hot on its own, look at that part. A graphics card that runs far hotter than its normal figure while the rest of the machine stays cool has usually dried out: the thermal paste between its core and heatsink has aged, and on an older card the thermal pads on the memory have too. Replacing the paste, and the pads if needed, is an afternoon’s work and a few pounds or dollars in materials, and it often drops the temperature by a large margin. The same is true of a CPU that spikes instantly to its throttle point under load, which is the classic sign of old, hardened paste or a cooler that has worked loose.

The rule of thumb: hot because it is dirty or dry is cheap to fix. Hot with no explanation, or hot and unstable even after a clean and fresh paste, is a part telling you it is on the way out.

What do the drives tell you?

Storage is where the clearest signals live, because the drives keep their own records and CrystalDiskInfo just reads them out.

For an SSD, look at two figures. The health or life percentage is an estimate of how much of the flash’s write endurance is left; a drive at 90% or above has years in it. The total host writes, compared against the drive’s rated endurance for its capacity, tells you how hard it has been used. A drive with a fraction of its rated writes gone is barely broken in. What you avoid is a caution status, a health figure that has dropped noticeably, or any reallocated sectors, because an SSD that has started to fail does so without much warning.

For a hard drive, the numbers matter and so do your ears. In CrystalDiskInfo, reallocated sector counts, pending sectors and a high power-on hours figure are the warnings. A drive with thousands of hours and a handful of reallocated sectors is living on borrowed time. Then listen to it: a healthy mechanical drive hums, and a dying one clicks, ticks or grinds. A clicking hard drive is a walk-away, and it is never worth the data-loss risk to a buyer even at a low price. Mechanical drives are cheap enough that a suspect one should simply be replaced rather than sold on.

Either way, a failing drive is rarely a reason to reject a whole machine, because a drive is the easiest part to swap. It is a reason to knock its value off the price and plan to fit a new one.

What do you look for in a graphics card?

The graphics card usually carries most of the value, so it deserves the closest look, and its failures are the ones that separate a cheap fix from a dead part.

Artifacts are the verdict. Run the card under a 3DMark loop and watch the screen. Coloured dots, flickering, strange textures, blocky patterns or a crash to a black screen all point to failing video memory or a dying core. That is a card at the end of its life, and no amount of cleaning or repasting brings it back. If you see artifacts, price the machine as though the card is worth little or nothing, and lean towards a part-out of everything else.

Heat and noise are usually fixable. A card that runs hot but renders cleanly almost always just needs new thermal paste and, on an older model, new thermal pads, as covered above. Fans that grind, rattle or wobble are a bearing wearing out, and replacement fans for popular cards are cheap and easy to fit. Neither is a reason to walk.

Check the physical signs too. Sag that has strained the slot, scorching or a burnt smell around the power connector, or bent pins on the connector itself are all worth a close look before you commit. And be honest with yourself about history: a card that was mined on is not necessarily more worn, since mining runs at steady temperatures rather than the heat cycling of gaming, but the market prices it lower and you should too.

What about the CPU and the memory?

Processors are the most reliable part in a PC and rarely the problem, which is why they are worth checking quickly and then trusting.

The CPU either works or it does not. If the machine POSTs, boots and holds a Cinebench run without crashing, the processor is almost certainly fine. The two things to watch are temperature, where an instant spike to the throttle point points to paste or a loose cooler rather than a bad chip, and stability under load, where random crashes can be the CPU but are far more often the memory or the power supply. On an AMD platform, look at the socket pins when the cooler is off, because a bent pin is a common and sometimes fixable fault on a chip that otherwise looks dead.

The memory is worth ruling out early precisely because its faults imitate everything else. A bad stick causes crashes, freezes and blue screens that look like a failing drive, a dying graphics card or an unstable processor, and you can waste an afternoon chasing the wrong part. Run MemTest86 from a USB stick, let it complete a pass or two, and you either get a clean result or a specific stick to pull. Bad RAM is cheap to replace, so it lowers the price rather than killing the machine.

What is fixable, and what means walk away?

Testing sorts every fault into one of two piles. The cheap-fix pile is where the margin lives, because it is full of machines that look worse than they are.

Fixable, and often the reason a machine is cheap:

• Overheating and throttling from dust, dead fans or bad airflow
• A part running hot from old thermal paste or dried-out pads
• Grinding or rattling fans, on the case or on the graphics card
• A failing or worn hard drive or SSD, which is a swap
• A single bad stick of RAM
• No display because a cable is in the motherboard rather than the graphics card

Reasons to walk away, or to buy parts-only at a price that assumes the worst:

• A graphics card that artifacts or crashes under load
• A drive that clicks or grinds, if it is the machine’s only proof of working
• Instability that survives a clean, fresh paste, known-good memory and a known-good power supply
• Liquid damage, bulging capacitors, or scorching around a socket or connector
• A machine that will not POST at all and cannot be brought to life with a spare power supply and the memory reseated

One part deserves its own note: the power supply. It is the component that ages worst and the one that can take others with it when it fails, and it is not something a benchmark tells you much about. When in doubt, budget for a new one rather than trusting a tired unit, and never sell a machine on a no-name supply that came free with a case.

Doing this more than once

Test one machine and the results live comfortably in your head. Test them by the dozen and the admin becomes the hard part.

Once you are buying and flipping regularly, the thing you keep losing track of is not how to test a part but what each part’s test said, what you paid for it, which build it went into, and what it eventually sold for. The 3DMark score for that card, the drive health for those two SSDs, the fact that this graphics card needs repasting before it goes anywhere: that is exactly the information that decides your prices, and it is the first thing to get lost when parts move between builds and sit around for weeks.

That is what PartPlanner is for. Parts, builds, condition, cost and sale outcomes in one place, so the work you put into testing a machine is still there the day you decide what to do with it.