Prime95 Errors Explained: Hardware Failure Detected and Fixes

A Prime95 error isn’t the test being broken. It’s the test doing its job: it found a calculation that came out wrong, and it stopped that worker to tell you. This guide explains the common messages, what usually causes them, and a step-by-step order for fixing them without guessing.

What do Prime95’s error messages mean?

Prime95 runs maths that has an exactly known right answer. Internally it checks the rounding error of each calculation against a threshold. If a result drifts past that threshold, the calculation was corrupted somewhere between the CPU core, its cache and memory, so the program reports an error and stops the worker that produced it.

You’ll typically see one of these messages, often with a pointer to a log file:

Forum threads show errors with values such as 0.5, 16 and very large numbers. The bigger the number, the more serious the error. These are descriptions from community threads rather than official documentation, so treat the exact thresholds as indicative.

What causes Prime95 errors?

Threads on HardForum, Overclockers Club and ComputerBase repeat the same list of causes. In rough order of how often they come up:

1. An unstable overclock. Core clocks run too high for the voltage supplied. This is the classic cause.
2. Too little core voltage. Sometimes even at stock settings, if the motherboard under-supplies voltage or an undervolt is too aggressive.
3. Memory problems. RAM running too fast, on too little voltage, or an XMP or EXPO profile the system can’t hold.
4. Heat. An overheating CPU or memory module produces errors even when settings are otherwise fine.
5. A weak power supply. Voltage droop under full load causes sporadic errors.
6. A faulty core or component. Occasionally, one worker fails repeatedly at stock settings with good temperatures.

Those same threads give consistent advice: if you want to know why you got the error, the first suspect is your overclock, and if you aren’t overclocking, the likely culprits are an overheating CPU or memory that isn’t quite up to specification.

How to fix a Prime95 error, step by step

Work through this order and change one thing at a time. Changing several settings together leaves you guessing which one mattered.

1. Return to stock. Reset BIOS overclocks, undervolts and memory profiles to defaults, then retest. Every troubleshooting thread starts here, because it separates “my tuning is unstable” from “my hardware is faulty”.
2. Check temperatures. Rerun the test with HWiNFO or Core Temp open. If the CPU reaches its limit, fix cooling first. See safe Prime95 temperatures.
3. Split the test. Run Small FFTs alone, then Blend alone. Where it fails tells you where to look (next section).
4. Loosen memory. Disable XMP or EXPO and retest at the memory’s base speed. Forum users report memory that passes other tests but still fails Prime95 Blend.
5. Adjust tuning in small steps. If you overclock, raise voltage a little or lower the clock slightly, then retest. Don’t jump by large amounts.
6. Check the power supply and motherboard. If the system is stock and still failing, suspect power delivery.

CPU or memory? Narrowing it down

This is the question the test mode can answer. Small FFTs stays inside the CPU cache and barely touches RAM, while Blend uses most of your memory. For a fuller explanation, see Small FFTs vs Large FFTs vs Blend.

This is a guide, not a guarantee. Blend itself can’t tell you whether the CPU or the RAM is responsible, and community members point out that exact drawback. A memory-only tool such as MemTest86 or a second stress test helps confirm.

What if errors happen at stock settings?

It happens, though less often. If you’ve reset everything to defaults, temperatures are fine, memory is at its base speed, and a worker still fails repeatedly, the cause is more likely a hardware fault. Reported culprits include a motherboard that under-supplies voltage, a failing power supply, or a faulty CPU core. At that point:

• Note whether the same worker (core) fails each time. Repeated failure on one worker points to a specific core.
• Test with a different memory kit or a single stick if you can, to rule out RAM.
• Test with a different power supply if you can.
• Check whether the CPU is still under warranty and contact the manufacturer or retailer.

Other symptoms that point to the same problems

• Random game or app crashes under load, especially after a BIOS change.
• Blue screens with WHEA errors, which often indicate hardware-level instability.
• Corrupted files or failed archive extraction, which can happen when memory is unstable.

If you see these in daily use, don’t dismiss a Prime95 error as a quirk of an extreme test. It’s an early warning of the same fault.

Should one error worry you?

Yes. An error means the system produced a wrong result under load. In a real workload, that could become a crash or silently corrupted data. Don’t treat a single failure as noise, especially if it repeats. A long run that ends with one error at the 11th hour still tells you the setting isn’t fully stable. It may be “stable enough” for a gaming PC and not for a workstation, but that’s a decision you should make knowingly.

Extra tools for confirming

• MemTest86 tests memory outside Windows, which separates a RAM fault from a CPU or software one.
• OCCT or AIDA64 give a second opinion with a different load. See Prime95 vs OCCT vs AIDA64 vs Cinebench.
• The online CPU stress test reports calculation errors from a lighter, browser-based load. Errors there are a strong hint of serious instability.

Error message glossary from the release notes

Prime95’s own Prime95 release notes mention several of the messages you may meet. They’re a useful reference because they come from the developer, not from forum memory.

The practical lesson is that a few of these messages in a row usually means you’ve got a real instability, and a hard stop after 100 errors is Prime95 telling you there’s no point continuing.

A proven troubleshooting order from PlayTool

The most detailed troubleshooting walkthrough that’s been published for years is on PlayTool. It reads like a checklist, and the order is the interesting part. In summary:

1. Test RAM first. The guide says that when Prime95 fails in any way, the first thing to do is run a memory test such as MemTest86 to rule it out.
2. If RAM passes, suspect the CPU or the CPU-to-RAM interface. On systems where the memory controller is inside the CPU, the CPU is the likely culprit.
3. Isolate with Small FFTs, since that test accesses RAM very little. A failure there points more strongly at the CPU, though not with certainty.
4. Check cooling. Overheating is a common cause. Clean dust from the heatsink and fan.
5. Monitor temperatures and voltages with a tool, and compare against published figures for your CPU.
6. Adjust voltage or clock speed if you’re experienced. The guide notes that raising CPU voltage can stabilise a CPU but adds heat and shortens lifespan, and too much can destroy the chip.
7. Check the power supply last. Watch the 12 V rail when Prime95 starts. Specifications allow about ±5%, and the guide advises avoiding the extremes, adding that weak supplies rarely cause these failures.

That ordering differs from the one at the top of this page in one place: PlayTool puts memory first. If you suspect your memory is the problem, test it with MemTest86 before changing CPU settings, because it quickly narrows the search. The guide also notes that some unstable machines crash or hang instead of reporting an error, so a freeze under load counts as a failure.

Make sure the test is actually loading every core

A subtle reason people see no errors is that the test wasn’t at full strength. PlayTool advises checking Task Manager and aiming for every CPU at 100%, noting that anything lower means the test isn’t at full strength. Close other programs, since they lower CPU load, and avoid disk activity during Blend. If the hard drive light is on constantly in Blend, the test has allocated too much memory, and you should reduce the “Memory to use” value under Custom until it runs properly. A test that isn’t fully loading the system can pass when a proper one would fail.

Frequently asked questions

Related guides