F5F Stay Refreshed Hardware Desktop Is high memory consumption still present even after a fresh Windows 11 setup?

Is high memory consumption still present even after a fresh Windows 11 setup?

Is high memory consumption still present even after a fresh Windows 11 setup?

Pages (2): Previous 1 2
L
lord_pug1234
Member
51
08-23-2022, 10:23 PM
#11
I can't respond that way. I will note though that some tools that show memory also include "virtual" limits that aren't actually used and can even surpass the system's real memory. For instance (at least in Linux), I've observed an iGPU on a machine with just 4 GB or 8 GB of RAM reporting 48 GB going to it. This wasn't actual usage, but rather a list of what the iGPU could access if that amount existed.

The concern about needing contiguous memory might be more significant than it appears. I don't understand why memory is being allocated, but a driver update or change (like a patch) can add a new driver before the iGPU loads. This may cause fragmentation, making the remaining memory unavailable.

I'm unsure how to diagnose the exact reason for the iGPU's high memory usage on Windows. A possible debug step is to create a new user account—perhaps non-admin plus one admin—and log in to those accounts. Check if the memory consumption drops; if it does, the problem likely stems from an auto-start driver when the high-memory account logs in. If usage stays high, it may point to a driver or other issue occurring before login. If that's the case, try testing without network access to see if network access affects the situation.
L
lord_pug1234
08-23-2022, 10:23 PM #11

I can't respond that way. I will note though that some tools that show memory also include "virtual" limits that aren't actually used and can even surpass the system's real memory. For instance (at least in Linux), I've observed an iGPU on a machine with just 4 GB or 8 GB of RAM reporting 48 GB going to it. This wasn't actual usage, but rather a list of what the iGPU could access if that amount existed.

The concern about needing contiguous memory might be more significant than it appears. I don't understand why memory is being allocated, but a driver update or change (like a patch) can add a new driver before the iGPU loads. This may cause fragmentation, making the remaining memory unavailable.

I'm unsure how to diagnose the exact reason for the iGPU's high memory usage on Windows. A possible debug step is to create a new user account—perhaps non-admin plus one admin—and log in to those accounts. Check if the memory consumption drops; if it does, the problem likely stems from an auto-start driver when the high-memory account logs in. If usage stays high, it may point to a driver or other issue occurring before login. If that's the case, try testing without network access to see if network access affects the situation.

B
bajenskie
Junior Member
2
08-24-2022, 02:09 AM
#12
Did you notice in the images that you only have one stick of 5200 MT/s RAM? That might not be an issue for basic tasks, but it would limit your computer's performance beyond office work and YouTube watching. You should have at least two sticks to fully utilize its speed. The 5200 MT/s RAM is also quite slow for a Ryzen system, which depend heavily on faster RAM with lower timings like 6000 MT/s with CL30. I recommend replacing it with a new set of RAM.
B
bajenskie
08-24-2022, 02:09 AM #12

Did you notice in the images that you only have one stick of 5200 MT/s RAM? That might not be an issue for basic tasks, but it would limit your computer's performance beyond office work and YouTube watching. You should have at least two sticks to fully utilize its speed. The 5200 MT/s RAM is also quite slow for a Ryzen system, which depend heavily on faster RAM with lower timings like 6000 MT/s with CL30. I recommend replacing it with a new set of RAM.

C
ChibiFalso
Junior Member
19
09-07-2022, 03:45 AM
#13
It's actually quite amusing and a bit awkward. I'm glad you pointed it out, even though it's completely unrelated to what I was discussing.
I was working in the old north-bridge design where all high-speed components shared the same bus.
I haven't really examined motherboard architectures in over ten years on x86 systems.
After hearing your comments, I decided to investigate further because your feedback caught my attention.
It turns out today's designs have changed significantly—there are dedicated buses for every high-speed peripheral, including RAM—which on consumer boards feature two separate channels. Right now I'm using just one channel, which means I'm missing out on a lot of bandwidth potential (from around 41.6 GB/s to about 83.2 GB/s with DDR5-5200 in dual-channel).
There will definitely be a performance boost, and I plan to upgrade next week—thank you for bringing this to my attention.
I also checked the AMD side to see if the CPU could actually take advantage of such high bandwidth.
It's worth mentioning that the actual performance gain might not be as strong as expected. Modern AMD CPUs handle their Infinity Fabric (internal interconnect) at a 1:2 ratio with DDR5 memory (clock speed around 1300MHz for DDR5-5200), which limits stability and adds some extra latency that somewhat reduces the bandwidth benefits. Also, your idea of switching to DDR5-6000 wouldn't necessarily make a big difference since we'd still be operating at the same 1:2 FCLK ratio.
C
ChibiFalso
09-07-2022, 03:45 AM #13

It's actually quite amusing and a bit awkward. I'm glad you pointed it out, even though it's completely unrelated to what I was discussing.
I was working in the old north-bridge design where all high-speed components shared the same bus.
I haven't really examined motherboard architectures in over ten years on x86 systems.
After hearing your comments, I decided to investigate further because your feedback caught my attention.
It turns out today's designs have changed significantly—there are dedicated buses for every high-speed peripheral, including RAM—which on consumer boards feature two separate channels. Right now I'm using just one channel, which means I'm missing out on a lot of bandwidth potential (from around 41.6 GB/s to about 83.2 GB/s with DDR5-5200 in dual-channel).
There will definitely be a performance boost, and I plan to upgrade next week—thank you for bringing this to my attention.
I also checked the AMD side to see if the CPU could actually take advantage of such high bandwidth.
It's worth mentioning that the actual performance gain might not be as strong as expected. Modern AMD CPUs handle their Infinity Fabric (internal interconnect) at a 1:2 ratio with DDR5 memory (clock speed around 1300MHz for DDR5-5200), which limits stability and adds some extra latency that somewhat reduces the bandwidth benefits. Also, your idea of switching to DDR5-6000 wouldn't necessarily make a big difference since we'd still be operating at the same 1:2 FCLK ratio.

P
PsychoPugx
Member
210
09-08-2022, 05:30 AM
#14
I plan to address this by offering a 2:1:1 ratio on AMD systems for optimal performance with 7000 and 9000 CPUs, which greatly lowers latency—crucial for gaming on AMD platforms. The most precise timing you can achieve is the best possible. If you're keen on going deeper, especially regarding gaming, you need the lowest first word latency achievable, which depends on RAM speed and cycle count. Aiming for 10 nanoseconds or less is ideal; it’s not significantly pricier than JEDEC standards but offers a noticeable boost. I’m sure this will have a real effect on data rates and timing, even though having at least two sticks is more important.

More details about real-world effects and scaling can be found in the TPU article.
P
PsychoPugx
09-08-2022, 05:30 AM #14

I plan to address this by offering a 2:1:1 ratio on AMD systems for optimal performance with 7000 and 9000 CPUs, which greatly lowers latency—crucial for gaming on AMD platforms. The most precise timing you can achieve is the best possible. If you're keen on going deeper, especially regarding gaming, you need the lowest first word latency achievable, which depends on RAM speed and cycle count. Aiming for 10 nanoseconds or less is ideal; it’s not significantly pricier than JEDEC standards but offers a noticeable boost. I’m sure this will have a real effect on data rates and timing, even though having at least two sticks is more important.

More details about real-world effects and scaling can be found in the TPU article.

D
DingbatPlayzMC
Senior Member
425
09-12-2022, 02:04 PM
#15
I chose a 32GBx2 6000 MT 30CL kit with a certified EXPO.
I’m hoping I won’t face any stability problems.
D
DingbatPlayzMC
09-12-2022, 02:04 PM #15

I chose a 32GBx2 6000 MT 30CL kit with a certified EXPO.
I’m hoping I won’t face any stability problems.

Pages (2): Previous 1 2