Question High memory usage even after a clean install of Windows 11 ?

[Sapling09]

Hey.

I'm dealing with unusually high memory usage on Windows 11, occurring immediately after boot.

As a developer, I maintain a clean system with minimal running applications. Despite this, my memory usage is abnormally high, and more frustratingly, I can't trace the source.

I recently performed a complete drive wipe and Windows 11 reinstall specifically to address this issue. Surprisingly, the high memory usage persisted.

This leaves me considering several possibilities:
- Driver memory leak (running only AMD chipset and NVIDIA drivers)
- Recent Windows 11 update affecting memory caching behavior
- Persistent malware surviving the wipe (possibly BIOS level)
- Normal behavior that I'm overthinking

While I understand Windows using memory for system optimization and caching, the inability to trace memory allocation is particularly concerning.

I will attach screenshots of what I see in the task manager:






My user processes account for only 3.5GB. The remaining usage is unaccounted for in any meaningful way.

The historical opacity of Windows memory management makes troubleshooting these issues particularly challenging. There seems to be no reliable method to track down memory allocation sources.

Linux is looking more appealing by the day.

Has anyone encountered similar issues or can offer guidance? Standard troubleshooting resources haven't yielded results.

EDIT:
My full specs are:
Processor AMD Ryzen 9 7900X 12-Core Processor 4.70 GHz
Installed RAM 32.0 GB (31.1 GB usable) 5200mhz
System type 64-bit operating system, x64-based processor
GPU: Nvidia Rtx 4080
CPU-Cooler: ARCTIC Liquid Freezer III 240 Black
Case: Fractal Design Meshify 2 RGB Black TG Light Tint
Motherboard: ASUS PRIME X670-P WIFI
Memory: 32GB PC5-41600 DDR5 SDRAM Kingston
HDD : WDC WD10EZRX-00L4HB0 1tb - Sata
SSD: WDC WDS100T2B0A-00SM50 1TB - Sata
SSD-M2-NVME: KINGSTON SKC3000D2048G 2TB, WD Blue SN580 2TB, Lexar 512GB SSD
PSU: I don't remember. It is some kind of 850W Platinun Seasonic
Monitor: Some LG 1080p 144hz Gsync 27 inch monitor.

 

[Lutfij]

I recently performed a complete drive wipe and Windows 11 reinstall specifically to address this issue.
Did you recreate the bootable USB installer using Windows Media Creation Tools? Did you install the OS in offline mode? If so, did you connect to the www to let the OS install drivers?

System specs:
When posting a thread of troubleshooting nature, it's customary to include your full system's specs. Please list the specs to your build like so:
CPU:
CPU cooler:
Motherboard:
Ram:
SSD/HDD:
GPU:
PSU:
Chassis:
OS:
Monitor:
include the age of the PSU apart from it's make and model. BIOS version for your motherboard at this moment of time.

Installed RAM 32.0 GB (31.1 GB usable)
You might want to also disable the iGPU in BIOS if you're not using the onboard display output ports on your motherboard. You might also have AMD's E.X.P.O disabled in BIOS.

 

[geofelt]

Everybody sees some 80% ram usage.
Task manager can be misleading if it is used to assess ram use.
Windows stores unused code in ram in anticipation of quick reuse.
If an app tries to access code that is not currently resident in ram, it needs to fetch it from the page file and perhaps write some older data to the page file to make room. That is called a hard page fault and while the fault is not resolved, the app stops dead.
Resolution can be painful if the page file is on a HDD.

When actively running, open task manager, open the resource monitor/memory tab.
Look at the hard fault rate column.
If you see anything much more than zero, you can use more ram.

 

[Sapling09]

I recently performed a complete drive wipe and Windows 11 reinstall specifically to address this issue.
Did you recreate the bootable USB installer using Windows Media Creation Tools? Did you install the OS in offline mode? If so, did you connect to the www to let the OS install drivers?

System specs:
When posting a thread of troubleshooting nature, it's customary to include your full system's specs. Please list the specs to your build like so:
CPU:
CPU cooler:
Motherboard:
Ram:
SSD/HDD:
GPU:
PSU:
Chassis:
OS:
Monitor:
include the age of the PSU apart from it's make and model. BIOS version for your motherboard at this moment of time.

Installed RAM 32.0 GB (31.1 GB usable)
You might want to also disable the iGPU in BIOS if you're not using the onboard display output ports on your motherboard. You might also have AMD's E.X.P.O disabled in BIOS.

Yes,
No, the OS was installed with the ethernet plugged in.
Yes it did install the drivers remotely from the internet.
In addition to the drivers it detected on its own, I installed my motherboards:
LAN drivers, Audio drivers, Chipset drivers
I also installed Nvidia drivers through nvidia experience.

My full specs are:
Processor AMD Ryzen 9 7900X 12-Core Processor 4.70 GHz
Installed RAM 32.0 GB (31.1 GB usable) 5200mhz
System type 64-bit operating system, x64-based processor
GPU: Nvidia Rtx 4080
CPU-Cooler: ARCTIC Liquid Freezer III 240 Black
Case: Fractal Design Meshify 2 RGB Black TG Light Tint
Motherboard: ASUS PRIME X670-P WIFI
Memory: 32GB PC5-41600 DDR5 SDRAM Kingston
HDD : WDC WD10EZRX-00L4HB0 1tb - Sata
SSD: WDC WDS100T2B0A-00SM50 1TB - Sata
SSD-M2-NVME: KINGSTON SKC3000D2048G 2TB, WD Blue SN580 2TB, Lexar 512GB SSD
PSU: I don't remember. It is some kind of 850W Platinun Seasonic
Monitor: Some LG 1080p 144hz Gsync 27 inch monitor.

AMD E.X.P.O is enabled.
iGPU is enabled as well. I will try to disable.

 

[Sapling09]

Okay, turns out that disabling my iGPU cut down the memory usage by 50%.

How is this possible? The iGPU was enabled, but my dedicated was used and the monitors were plugged into the dedicated GPU only.

Also, windows reported only 1gb HW reserved, so how could this have affected the memory so much?

 

[Lutfij]

No, the OS was installed with the ethernet plugged in.
Install the OS in offline mode;

View: https://www.youtube.com/watch?v=_GuFH9KdHC0

with the ethernet unplugged and the drive you wish to install the OS on, connected while all other drives are disconnected.

Yes it did install the drivers remotely from the internet.
With the internet disconnected, install all relevant drivers in an elevated command, i.e, Right click installer>Run as Administrator. Yes that means you download all relevant drivers with the latest versions and have them on a USB drive to use after the OS install is complete.

I also installed Nvidia drivers through nvidia experience.
You install said driver without GeForce Experience, downloading the latest version from here;

Driver Results | <dd~ProductName> | <dd~OSName>

Lookup results for <dd~ProductName> product and Operating System <dd~OSName>

www.nvidia.com

Installed RAM 32.0 GB (31.1 GB usable) 5200mhz
Please pass on a link to the ram kit. You're not taking advantage of your platform with that frequency on your ram kit.

Motherboard: ASUS PRIME X670-P WIFI
You forgot to mention the BIOS version for your build.

 

[Lolomid]

the main reasons for Windows 11 high memory usage is you are opening too many apps or services.

 

[CountMike]

Okay, turns out that disabling my iGPU cut down the memory usage by 50%.

How is this possible? The iGPU was enabled, but my dedicated was used and the monitors were plugged into the dedicated GPU only.

Also, windows reported only 1gb HW reserved, so how could this have affected the memory so much?

That IGPU has only 512MB of ram dedicated as default and only up to 2GB maximum. That minimum is dedicated even when not used but enabled in BIOS. W10/11 shouldn't use more than 3.5GB at true idle.

 

[LinuxDevice]

Okay, turns out that disabling my iGPU cut down the memory usage by 50%.

How is this possible? The iGPU was enabled, but my dedicated was used and the monitors were plugged into the dedicated GPU only.

Also, windows reported only 1gb HW reserved, so how could this have affected the memory so much?

An iGPU usually uses the system RAM. A dGPU normally has its own VRAM.

EDIT: The iGPU RAM must be contiguous as well, and so it typically reserves large blocks at startup.

 

[Sapling09]

An iGPU usually uses the system RAM. A dGPU normally has its own VRAM.

EDIT: The iGPU RAM must be contiguous as well, and so it typically reserves large blocks at startup.

I am well aware that the integrated gpu usually shares memory with the GPU, but I have never had this problem before.

I have been running this build for quite a while and only recently I started noticing high memory usage.
That is why I always had the iGPU enabled in the bios cause I never considered it to be a problem.

Why is it using so much?

 

[LinuxDevice]

I am well aware that the integrated gpu usually shares memory with the GPU, but I have never had this problem before.

I have been running this build for quite a while and only recently I started noticing high memory usage.
That is why I always had the iGPU enabled in the bios cause I never considered it to be a problem.

Why is it using so much?

I can't answer that. I will say though that some tools reporting memory also include "virtual" memory limits which are not actually used and which can even far exceed the system's memory. For example (at least in Linux), I've seen an iGPU on a system with only 4 GB or 8 GB of system RAM reporting 48 GB of RAM going to the iGPU. This was not actual RAM usage, but was instead a listing of what the iGPU was capable of using (if that much exists).

The part about needing contiguous memory is probably more important than it seems. I do not know why the memory is being used, but some new driver is added or changed (such as with a patch update) and that driver loads prior to the iGPU, then the remaining memory might no longer be available due to fragmentation.

I don't know under Windows exactly how you would find out why the iGPU is using more memory. One debug step which you might take is to add a new user login (perhaps a non-admin login, plus one admin login); then log in to those accounts and see if the memory use is still more than expected. If memory use of the iGPU goes down, then likely the issue is due to something which auto starts upon login of the high memory use account. If memory use remains high, then likely this is related to some driver or other issue which occurs before login. If this latter is the case, then you might try a similar test with networking unplugged to see if access to a network is related.

 

[helper800]

I am well aware that the integrated gpu usually shares memory with the GPU, but I have never had this problem before.

I have been running this build for quite a while and only recently I started noticing high memory usage.
That is why I always had the iGPU enabled in the bios cause I never considered it to be a problem.

Why is it using so much?

Did I see correctly in your pictures that you only have 1 stick of 5200 MT/s RAM? That is not necessarily a problem for basic functions but you would be gimping your computer for any use other than office work and watching youtube. You need to have at least two sticks of RAM to take full advantage of its bandwidth speeds. 5200 MT/'s RAM is also awfully slow for a Ryzen system as they are very reliant on faster RAM with lower timings like 6000 MT/s with CL30. I would suggest a new kit of RAM.

 

[Sapling09]

Did I see correctly in your pictures that you only have 1 stick of 5200 MT/s RAM? That is not necessarily a problem for basic functions but you would be gimping your computer for any use other than office work and watching youtube. You need to have at least two sticks of RAM to take full advantage of its bandwidth speeds. 5200 MT/'s RAM is also awfully slow for a Ryzen system as they are very reliant on faster RAM with lower timings like 6000 MT/s with CL30. I would suggest a new kit of RAM.

This is actually pretty funny and kind of embarrassing. I am glad you called me out on this, despite this being a completely unrelated issue.

I was stuck in the good old north-bridge architecture where all the high-speed devices shared the same bus.
I haven't looked properly at mother board architectures for over a decade on x86 architectures.

So after you mentioned this I decided to actually look it up because what you said surprised me.

Turns out modern architectures are quite different - there are dedicated buses to every high-speed peripheral, including the ram---Which on consumer boards have two dedicated channels. Right now I'm only using one of these channels, so I am indeed missing out on potential bandwidth (going from theoretical ~41.6 GB/s to ~83.2 GB/s with DDR5-5200 in dual channel).

There will definitely be a performance improvement, and I will order another stick next week - thanks for bringing this up.

So I also looked at the AMD architecture to see if the CPU can actually even utilize this much bandwidth.

Well, it's worth noting that the performance impact might not be as dramatic as suggested. Modern AMD CPUs run their Infinity Fabric (the internal interconnect) at a 1:2 ratio with DDR5 memory (FCLK at 1300MHz for DDR5-5200) due to stability limitations. This introduces some additional latency that partially offsets the bandwidth gains. Also, your suggestion about upgrading to DDR5-6000 wouldn't necessarily yield significant benefits since we'd still be running at the same 1:2 FCLK ratio.

 

[helper800]

This is actually pretty funny and kind of embarrassing. I am glad you called me out on this, despite this being a completely unrelated issue.

I was stuck in the good old north-bridge architecture where all the high-speed devices shared the same bus.
I haven't looked properly at mother board architectures for over a decade on x86 architectures.

So after you mentioned this I decided to actually look it up because what you said surprised me.

Turns out modern architectures are quite different - there are dedicated buses to every high-speed peripheral, including the ram---Which on consumer boards have two dedicated channels. Right now I'm only using one of these channels, so I am indeed missing out on potential bandwidth (going from theoretical ~41.6 GB/s to ~83.2 GB/s with DDR5-5200 in dual channel).

There will definitely be a performance improvement, and I will order another stick next week - thanks for bringing this up.

So I also looked at the AMD architecture to see if the CPU can actually even utilize this much bandwidth.

Well, it's worth noting that the performance impact might not be as dramatic as suggested. Modern AMD CPUs run their Infinity Fabric (the internal interconnect) at a 1:2 ratio with DDR5 memory (FCLK at 1300MHz for DDR5-5200) due to stability limitations. This introduces some additional latency that partially offsets the bandwidth gains. Also, your suggestion about upgrading to DDR5-6000 wouldn't necessarily yield significant benefits since we'd still be running at the same 1:2 FCLK ratio.

I will go ahead and counter with this. You want to have a 2:1:1 ratio on AMD systems for best performance on 7000 and 9000 CPUs as this significantly reduces latency which is highly important for gaming on AMD. The tightest timing you can get the better. If you really want to get down into the weeds, specifically with gaming, you really want the best first word latency you can get which is a function of the speed of the RAM and how many cycles it takes the first word. 10 nano seconds and faster is the "sweet spot" because its not much more expensive than the JEDEC standards while being reasonably faster. Believe me it does make a meaningful impact to have these higher data rates and tighter timing, though neither of those are as important as having at least two stick, lol. Significantly more information is in this article from TPU about the actual real world performance impacts and scaling.

 

[Sapling09]

I decided to get a 32GBx2 6000 MT 30CL kit with a certified EXPO.
Hopefully I won't run into stability issues