Processor and graphics card specifications are live

Problems · How to check

How do I test my RAM speed?

There are two different things people mean, and they need different tools. To read the rated speed your memory is configured at, open Task Manager on Windows, choose Performance and then Memory, and the speed in megahertz is listed there. That takes ten seconds and tells you what the system thinks it is set to.

To measure what your memory actually delivers, you need a bandwidth and latency test, because the rated figure is a configuration setting rather than a result. That second measurement is the one that reveals a single populated channel, a profile that was never enabled, or a controller that is the real ceiling.

First, notice when it happens

From the first minute

Memory results are stable from the first second. Bandwidth and latency are properties of the configuration rather than of temperature, so unlike a processor test there is nothing to warm up and no curve to watch. A reading taken after ten seconds is as valid as one taken after ten minutes.

Only after minutes of use

If performance degrades over minutes, memory is not the explanation and you should be looking at heat and power limits instead. This is a useful thing to establish early, because memory upgrades are frequently bought to solve problems that were actually thermal.

The causes, ranked

1. You want the rated speed, which the operating system will simply tell you

most common

How to confirm it

You want to know whether the modules are running at the speed they were sold at.

The fix

On Windows use Task Manager, Performance, Memory. On macOS use About This Mac and then More Info. On Linux, run the system profiler for memory hardware. Compare what it reports against the rating printed on the modules.

2. The reported speed is far below the rating, meaning the profile is off

common

How to confirm it

Modules rated DDR5-6000 report as DDR5-4800, or DDR4-3600 reports as DDR4-2133. Memory runs at a conservative default until a profile is switched on.

The fix

Enable XMP on an Intel board or EXPO on an AMD one in the firmware setup, then re-check and test for stability. This is the most common reason fast memory performs like slow memory.

3. The speed is correct but the machine is still memory limited

common

How to confirm it

Rated speed looks right, but measured bandwidth barely rises when more threads read at once. That points at channel configuration rather than at the modules.

The fix

Check that more than one slot is populated and that the modules sit in the pair of slots the manual specifies. Moving two modules into the correct slots costs nothing and can add substantial bandwidth.

4. You are actually looking for faulty memory, which is a different test

occasional

How to confirm it

The symptom is crashes, corruption or blue screens rather than slowness. A speed or bandwidth test will not find this and can pass on failing memory.

The fix

Run a dedicated memory tester across the whole address range, for hours rather than minutes. Windows has a built-in memory diagnostic, and there are more thorough bootable testers for a serious hunt.

What a browser can and cannot measure here

A browser can measure the two things that actually matter for speed: how much data moves per second, from one thread and from all of them, and how long a single dependent read takes. Those are measured by doing the work and timing it, which is why they need no special access.

A browser cannot read the configuration: not the capacity, not the clock, not the timings, not how many slots are filled. So the useful output is not a specification sheet, it is a set of measurements plus what those measurements imply, stated as an implication rather than a reading.

The practical consequence is that the browser test and Task Manager answer different halves of the question, and it is worth doing both. Task Manager tells you what the machine is set to. The measurement tells you what that setting is producing.

Reading the result honestly

Compare single-thread against all-thread bandwidth first. A healthy memory system gives added threads more bandwidth. One that is already saturated by a single thread does not, and that is the signature worth acting on.

Treat latency as a secondary figure. It moves slowly between generations, it is rarely the thing you can fix, and a workload has to be unusual for latency rather than bandwidth to be its limit.

Do not compare an absolute bandwidth number against someone else's machine on a different platform and conclude anything. Compare against your own machine before and after a change, which is a comparison that controls for everything else.

Measure it instead of guessing

The memory stage measures read and write bandwidth from a single worker and from every logical core at once, then stops all of them and measures latency with a single-cycle random permutation over a 64 MiB array, so the processor cannot prefetch or overlap the reads. The one-thread to all-thread ratio is reported directly, and below about 1.35 times the report says the memory system rather than the cores looks like the ceiling.

What software cannot tell you here

  • Your module capacity, clock speed, timings, or how many slots are populated. None of it is exposed to a web page, so a configuration problem is inferred from behaviour and labelled as an inference.
  • Whether memory is faulty. Bandwidth measurement is not an error scan and will pass on modules that are failing.
  • Whether enabling a profile would be stable on your board. That has to be tested after the change, not predicted before it.

Asked, in people’s own words

How do I check my RAM speed on Windows 11?
Open Task Manager, select Performance, then Memory. The configured speed in megahertz appears on the right along with the slots in use. Compare that against the rating on the modules: if it is much lower, the XMP or EXPO profile has probably never been enabled.
Can I test RAM speed without installing anything?
You can measure bandwidth and latency in a browser, because both are measured by doing timed work rather than by reading hardware registers. You cannot read the rated clock or the timings that way, since no browser exposes them, so pair the measurement with what Task Manager reports.
Why is my RAM running slower than advertised?
Almost always because the advertised speed is a profile that has to be switched on. Memory boots at a conservative standard speed and stays there until XMP or EXPO is enabled in the firmware, so a kit sold as DDR5-6000 will happily run at 4800 indefinitely.
Does a memory speed test find bad RAM?
No. Speed and bandwidth tests measure throughput, not correctness, and will pass on memory that is producing errors. If you are chasing crashes or corruption, run a dedicated memory diagnostic across the full address range for several hours.

Name the cause. Three and a half minutes, in the tab you have open.