Why Task Manager Says 95% Memory When Nothing Is Running

You open Task Manager on a machine that feels sluggish. Memory is at 94%. You scroll the process list and add up the big ones: browser 2 GB, everything else maybe 3 GB total. That is 5 GB on a 16 GB machine.

So where did the other 10 GB go?

This is the most common memory question I get, and the answer is almost never "you have a leak." It is usually that the number at the top of Task Manager and the numbers in the list below it are measuring different things, and neither one means what people assume.

Here is what each number actually is, which ones you can ignore, and the one that is worth worrying about.

The number at the top is not "used by programs"

On the Performance tab, the Memory panel shows four things that matter:

  • In use — pages currently held by processes, drivers, and the kernel
  • Available — memory that can be handed to a process right now
  • Committed — total private memory promised to processes, including what is backed by the pagefile
  • Cached — file data kept in RAM in case it is needed again

The one people read as "my RAM is full" is In use. But "In use" includes kernel and driver allocations that never show up in the process list. On a normal desktop that is several hundred megabytes. On a machine with a misbehaving driver it can be many gigabytes.

And Committed routinely exceeds your physical RAM. That is by design. It is a promise, not an allocation. A machine with 16 GB of RAM and 20 GB committed is not in trouble as long as the pages actually in use fit.

Why the process list never adds up

The Processes tab does not show "memory used by this process." It shows the active private working set — the pages a process has touched recently that are not shared with anything else.

Two consequences:

1. Shared pages are excluded. Every process that loads the same system DLL shares one physical copy. That copy is not attributed to any of them. 2. Pages the process has not touched recently are excluded, even if it still owns them.

So the column under-counts. Adding up the list will always give you less than "In use." That gap is not missing memory, it is memory the list is not designed to show you.

If you want the number that does add up, use the Details tab and add the Commit size column instead. That is private committed memory, and it is the closest thing to "how much this process has reserved."

The three things that fill memory without a process

Once you accept that the list will not reconcile, the question becomes: what is holding the rest? Three answers, in order of how often they are the cause.

1. Memory compression (normal, ignore it)

Since Windows 10, when memory gets tight the system compresses pages instead of writing them straight to the pagefile. Compressed pages stay in RAM but take less space than they would uncompressed.

Task Manager shows this as In use (Compressed). Seeing 1–4 GB compressed on an 8 or 16 GB machine is normal and is the system working as intended. It is faster than paging to disk and it is not a problem to solve.

You can confirm it is on:

Get-MMAgent | Select-Object MemoryCompression

True means enabled, which is the default. Leave it alone. (Get-MMAgent documentation)

2. The standby list (also normal, and often misread)

The standby list holds file data that was read from disk and kept in RAM in case it is needed again. It shows up under Cached.

This is the single most misread number in Windows. People see "Cached: 9 GB" and conclude that something is hogging memory. The opposite is true: standby pages are given up instantly when a process asks for memory. A large cache means the system is doing its job — using RAM you paid for instead of leaving it idle.

The only time standby matters is if it is full of pages that cannot be evicted (the "standby" vs "modified" distinction). RAMMap shows that breakdown if you ever need it.

3. Non-paged pool (this is the one to check)

Kernel memory comes in two flavours. Paged pool can be written to disk. Non-paged pool cannot — it must stay in physical RAM at all times.

Normal size on a desktop is roughly 200–600 MB. If it is sitting at several gigabytes and climbing every day, you have a driver leak, and that is a real problem that will not fix itself.

Check it:

Get-Counter '\Memory\Pool Nonpaged Bytes'

Divide the raw byte count by 1 GB. Run it now, leave the machine running for a day, run it again. A slow steady climb is the signature of a leaking driver.

To find which one, download RAMMap from Sysinternals, open the Nonpaged Pool tab, and sort by size. It groups allocations by pool tag, and pool tags map back to drivers. The tag is usually the first four characters of the driver name — NDIS for network, Proc for process management, and so on.

How to measure it properly

Task Manager is a summary. If you actually want to know whether you have a memory problem, three commands cover it.

Available memory — the only number that tells you if you are close to the edge:

Get-Counter '\Memory\Available MBytes'

Under about 500 MB sustained, with the machine paging, is genuine pressure. Above 2 GB, the percentage at the top of Task Manager is cosmetic.

Hard faults — how often the system actually has to go to disk:

Get-Counter '\Memory\Pages/sec'

A few hundred sustained means real paging. Occasional spikes are normal and harmless.

The top consumers by working set:

Get-Process | Sort-Object WS -Descending | Select-Object -First 10 Name, WS, PM

WS is working set in bytes, PM is private memory. If one process dominates both and keeps growing over hours, that is your leak — and it is an application problem, not a Windows problem.

What actually fixes it

If available memory stays above roughly 1.5 GB and pages/sec is low, nothing needs fixing. The percentage is high because Windows is using the RAM for cache, which is the point of having RAM.

If non-paged pool is large and growing, the fix is a driver update or removal. Check recently installed software first — VPN clients, antivirus, and virtual network adapters are the usual suspects. driverquery /v lists loaded drivers with dates if you need to spot something new.

If one application's working set grows without bound over days, that is the application. Restarting it is the workaround; reporting it is the fix.

What does not fix it

Memory "optimizers" and RAM cleaners. These work by forcing pages out to the pagefile to make the number in Task Manager go down. That number was never the problem, and now your data is on disk instead of in RAM. You have made the machine slower and the graph prettier. Uninstall them.

Disabling the pagefile. Some guides suggest this to "free up disk space" or "force Windows to use RAM." It removes the safety margin that lets committed memory exceed physical memory. When something asks for more than you have, the allocation fails and the application crashes instead of paging. Leave the pagefile managed by the system.

Adding RAM as a first move. It helps if you genuinely run out — available memory near zero with heavy paging. It does nothing for a driver leak or a misread graph, and it is the most expensive of the three options.

The short version

The percentage in Task Manager counts kernel and driver memory that the process list does not show, which is why the numbers never reconcile. A high percentage with healthy available memory is Windows using your RAM as cache — working as designed.

Check three things, in this order: available memory, pages per second, non-paged pool. If the first is above 1.5 GB and the second is low, close Task Manager and go do something else. If non-paged pool is in the gigabytes, start looking at drivers.

If your problem is disk space rather than memory, that is a different set of hidden consumers — I wrote about those separately: where the space actually went on Windows 11.

Comments

Popular posts from this blog

Why plugins corrupt your .xlsx files (and how to check before it happens)

Windows 11 Is Full and It's Not Your Apps: Where the Space Actually Went