What Are “Dedicated GPU Memory” and “Shared GPU Memory” in Task Manager? — Why the Numbers Differ So Much on Integrated GPUs, and What Happens When Memory Runs Short

When you open the GPU section of Task Manager, you see two numbers side by side: “Dedicated GPU memory” and “Shared GPU memory”. On a laptop with an integrated GPU, dedicated memory may be tiny while shared memory is shown as large, about half of the PC’s memory.

Dedicated GPU memory is memory used only by the GPU; shared GPU memory is ordinary memory that both the CPU and the GPU can use. The large shared GPU memory number is the upper limit the GPU is allowed to use. That much memory is not always taken by the GPU.

This article is for general users who are curious about what Windows 11’s Task Manager shows. It is based on official documentation from Microsoft, Intel and AMD checked on November 3, 2026. I checked the settings screen for choosing a GPU per app in a Windows 11 Enterprise evaluation (26H2) virtual machine. I also checked the Task Manager and dxdiag output, the shared GPU memory limit, and the GPU selection screen on a physical Windows 11 Pro (25H2) PC with both an integrated and a discrete GPU (128 GB of memory). I did not check the per-process columns on the [Details] tab. The names shown on screen may differ slightly by Windows version and language.

How to read the GPU section of Task Manager

GPU information appears on Task Manager’s [Performance] tab under items such as “GPU 0”. On a PC with two GPUs they are listed as “GPU 0” and “GPU 1”; the numbers are assigned by Task Manager.¹ On the left, the GPU’s name appears under “GPU 0”, and below it the usage (and temperature), as in “25% (40 °C)”.

In this article I write “dedicated GPU memory” and “shared GPU memory” in running text, but on screen the fields are labelled “Dedicated GPU memory”, “Shared GPU memory” and “GPU memory”.

At the top right are graphs for each part of the GPU. A GPU has a part that handles 3D drawing, a part that handles data transfer, a part that handles video playback and so on, and each runs separately. Microsoft calls these GPU engines. The graphs are named “3D”, “Copy”, “Video Decode” and so on, and which engines exist depends on the GPU.¹

flowchart TB
        accTitle: Layout of the GPU section in Task Manager
        accDescr: Shows that the GPU item on the Performance tab displays a usage graph per GPU engine, the overall GPU usage, and the usage of dedicated and shared GPU memory.
        A["GPU 0 on the Performance tab"] --> B["A graph per GPU engine"]
        A --> C["Overall GPU usage"]
        A --> D["Dedicated GPU memory"]
        A --> E["Shared GPU memory"]
        B --> B1["3D, Copy, Video Decode and others"]

Figure 1: The GPU section shows usage per engine and the usage of two kinds of memory.

The overall usage shown under “GPU 0” on the left, such as “25% (40 °C)”, is not the average of the engines. It is the usage of the busiest engine. Video playback may not use the 3D engine at all. In that case the overall figure is the value of the busiest engine, such as the one handling video. Microsoft’s GPU engineers explain that they chose this approach because an average would make a game that is fully using the GPU look low.¹

When I played a 4K video on a physical Windows 11 Pro (25H2) PC, the overall figure was 24–25%, Video Decode was 24–25%, 3D was 18–22%, and the other engines were 0%. The overall figure matched Video Decode, the busiest engine, not the average of the engines (about 11–12%).

flowchart TB
        accTitle: How overall GPU usage is determined
        accDescr: Shows that the highest of the per-engine usages, such as 3D, Copy and Video Decode, is displayed as the overall GPU usage.
        A["3D engine usage"] --> M["Pick the highest value"]
        B["Copy engine usage"] --> M
        C["Video Decode engine usage"] --> M
        M --> R["Overall GPU usage"]

Figure 2: Overall GPU usage is represented by the value of the busiest engine.

To see GPU information in Task Manager, the GPU driver must support WDDM 2.0 or later. You can check the driver’s WDDM version by starting dxdiag from the Start menu and looking at the “Driver Model” value in the driver section of the [Display] tab.¹ On a physical PC (25H2) with two monitors connected, the tabs were numbered, as in [Display 1] and [Display 2]. dxdiag shows display tabs only for GPUs that have a monitor connected. I could not see the driver model of the integrated GPU, which had no monitor connected, in dxdiag.

In the diagram a solid line marks a relation that always holds and a dashed line marks a conditional one (the conditions are given per relation on the detail page). The full list of relations (12 in total, with evidence and certainty) and the definitions of the main concepts are collected on the knowledge map detail page (in Japanese). Data: JSON-LD / Turtle

Dedicated GPU memory is memory used only by the GPU

Dedicated GPU memory is memory reserved for use by the GPU alone. In Windows, the memory used by the GPU is managed by a component called the video memory manager (VidMm), and Task Manager’s numbers are collected from it.¹

On a discrete graphics card (discrete GPU), dedicated GPU memory is the memory on the card, the so-called VRAM. Its amount is fixed by the card.¹²

Integrated GPUs are different. An integrated GPU is a GPU built into the processor. Some integrated GPUs reserve part of ordinary memory for the GPU’s exclusive use when the PC starts or the driver initializes. The reserved part is removed from Windows’ memory management and used only by the GPU. However, Microsoft considers using shared memory more flexible and efficient than this reservation, and many integrated GPUs reserve almost nothing.¹

flowchart TB
        accTitle: What dedicated GPU memory consists of
        accDescr: Shows that on a discrete GPU the VRAM on the graphics card is the dedicated GPU memory, and on an integrated GPU the portion of ordinary memory reserved at startup or similar is the dedicated GPU memory.
        A["Dedicated GPU memory"] --> B["Discrete GPU"]
        A --> C["Integrated GPU"]
        B --> B1["VRAM on the card"]
        C --> C1["A portion reserved from ordinary memory"]
        C1 -.-> C2["Many integrated GPUs reserve nothing"]

Figure 3: What dedicated GPU memory consists of depends on the type of GPU.

The dedicated GPU memory usage on the Performance tab is the total that all processes are using right now. It is not the amount that apps asked to allocate.¹

Shared GPU memory is ordinary memory both the CPU and GPU can use

Shared GPU memory is ordinary memory (RAM) that both the CPU and the GPU can use. Its use is not fixed. The GPU uses it when the GPU needs it, and it goes back to the CPU side when it is no longer needed. Both discrete and integrated GPUs can use this shared memory.¹

The large number shown on the right is the upper limit of ordinary memory the GPU is allowed to use. The GPU has not reserved that amount. The usage on the left is what the GPU is using right now.¹

Windows decides this limit. When GPU information was added to Task Manager in 2017, the explanation was that the ordinary memory the GPU could use at once was capped at half of physical memory. On a 16 GB PC, the GPU could use up to 8 GB, so that the remaining memory keeps the system running.¹

Microsoft’s current documentation for driver developers gives the following formula. The upper limit of ordinary memory the GPU can use (including the portion reserved for dedicated use) is the smaller of “80% of memory” and “the larger of memory minus 16 GB and half of memory”. Memory here is the amount visible to Windows, excluding what the BIOS reserves.³

flowchart TB
        accTitle: How the limit on ordinary memory the GPU can use is determined
        accDescr: Shows that from the amount of memory visible to Windows, the larger of half the memory and the memory minus 16 GB is chosen, and the value capped so it does not exceed 80% of memory becomes the limit.
        A["Memory visible to Windows"] --> B["The larger of half, and minus 16 GB"]
        B --> C["Cap it at 80%"]
        C --> D["Limit on ordinary memory the GPU can use"]
        D -.-> E["What remains after the dedicated reservation is the shared limit"]

Figure 4: The shared GPU memory limit is calculated from the amount of memory.

Applying this formula gives the following limits.

Up to 32 GB it is half, as in the 2017 explanation. Above that it is more than half. The shared GPU memory limit is what remains after subtracting the dedicated reservation, and depending on the driver’s settings it can be smaller still.³

On a physical PC (25H2) with 128 GB of memory (127.69 GB visible to Windows), Task Manager’s shared GPU memory limit was 102 GB and dxdiag’s “Shared Memory” was 104605 MB. Both match the 80% side of the formula.

Below dedicated and shared, the GPU section also has a “GPU memory” row. I could not find any Microsoft documentation that explains the meaning of this row specifically. The driver documentation calculates total video memory as the sum of dedicated video memory, dedicated system memory and shared system memory.³ The physical PC (25H2) also matched this. The discrete GPU showed “GPU memory 2/114 GB”, the sum of 12.0 GB dedicated and 102 GB shared. The integrated GPU showed “0.2/102 GB”, only the shared portion. dxdiag’s “Total Memory 116599 MB” is also the sum of “Display Memory (VRAM) 11994 MB” and “Shared Memory 104605 MB”.

An app can still allocate more video memory than this limit. Windows’ video memory is virtualized, and the limit on how much can be allocated is the system-wide commit limit, which combines memory and the page file. VidMm keeps the ordinary memory the GPU uses at once from exceeding the limit by pinning and releasing memory pages.¹ How to read Windows’ memory numbers is explained in “What Does Windows’ “Memory Usage” Actually Mean?”.

Why integrated GPUs show small dedicated and large shared numbers

On a PC with an integrated GPU, it is normal for dedicated GPU memory to be small and shared GPU memory to be large. Integrated GPUs run mainly on shared memory. The shared GPU memory limit is half of memory on PCs with up to 32 GB, and larger on PCs with more (on a 128 GB PC, 102 GB, which is 80% of memory).

On the integrated GPU (Intel UHD Graphics 770) of a physical Windows 11 Pro (25H2) PC, neither a “Dedicated GPU memory” field nor its graph appeared. Only “Shared GPU memory 0.2/102 GB” and “GPU memory 0.2/102 GB” were shown. Some PCs have no dedicated field at all.

Intel explains that Intel integrated GPUs have no dedicated memory and use shared memory allocated automatically. What you can adjust in the BIOS is only the upper limit of how much can be used.⁴ AMD likewise explains that PCs with integrated graphics are built on a unified memory architecture (UMA) and share part of ordinary memory with the GPU.⁵

According to Microsoft, a small amount of memory reserved by the driver may also be shown as “Hardware Reserved Memory”.¹

flowchart TB
        accTitle: How integrated and discrete GPUs use memory
        accDescr: Shows that a discrete GPU mainly uses the VRAM on the card and uses ordinary memory for what does not fit and similar, while an integrated GPU has a small or no dedicated reservation and uses shared ordinary memory.
        A["Discrete GPU"] --> A1["Dedicated: VRAM on the card"]
        A --> A2["Shared: ordinary memory"]
        B["Integrated GPU"] --> B1["Dedicated: small or none"]
        B --> B2["Shared: mainly ordinary memory"]

Figure 5: Integrated GPUs use shared ordinary memory, so the dedicated number comes out small.

Don’t add up the per-process numbers

GPU memory usage can also be viewed per process on the [Details] tab. Right-click a column header, open [Select columns], and add the dedicated GPU memory and shared GPU memory columns.¹

If you add up the figures for each process here, the total is larger than the overall GPU usage on the Performance tab. This is because memory shared by several processes is counted again in each process’s figure. Microsoft gives the example of memory holding the contents of a window created by an app, which is referenced by three processes: the app itself, csrss.exe, and dwm.exe, which composes the desktop. The memory is allocated only once, but it is included in the figure of each of the three processes. That is why the figures for dwm.exe and csrss.exe look larger than they really are.¹

flowchart TB
        accTitle: How shared video memory is counted
        accDescr: Shows that video memory holding an app window's contents is allocated only once but counted in the usage of each of the app, csrss.exe and dwm.exe.
        M["Memory holding the window's contents"] --> A["Counted in the app's figure"]
        M --> C["Counted in csrss.exe's figure"]
        M --> D["Counted in dwm.exe's figure"]
        A --> S["Adding them up overstates it"]
        C --> S
        D --> S

Figure 6: Shared memory is counted again in the figure of each process that uses it.

Check how much the GPU as a whole is using on the Performance tab. In Windows 10 there was also a known issue where the dedicated GPU memory column on the Details tab appeared to grow by about 100 MB each time you sent an app to the background and brought it back. Microsoft explains that even then, the Performance tab shows the correct value.⁶

What happens when GPU memory runs short

There is a budget, set by Windows, for how much video memory an app can use. This budget changes with apps running in the background and with switching to other apps. According to Microsoft’s documentation, when an app exceeds the budget, the app is intermittently suspended or fails to allocate memory.⁷

On a discrete GPU, Windows may move data that doesn’t fit in VRAM to ordinary memory. Microsoft, however, positions this as a last resort.⁷

flowchart TB
        accTitle: What happens when the video memory budget is exceeded
        accDescr: Shows that when an app exceeds the video memory budget, the app is intermittently suspended or fails to allocate memory, and that on a discrete GPU data may be moved to ordinary memory as a last resort.
        A["The app exceeds its budget"] --> B["The app is intermittently suspended"]
        A --> C["Memory allocation fails"]
        A -.-> D["On a discrete GPU, data may move to ordinary memory"]
        D -.-> E["Microsoft treats this as a last resort"]

Figure 7: When the budget is exceeded, the app stalls or fails to allocate.

To users, this can show up as a game or similar app stalling now and then. AMD notes that in games not optimized for integrated graphics, textures may be missing or low resolution, or a warning may appear that the GPU does not meet the requirements.⁵

To check in Task Manager, see whether dedicated GPU memory usage is close to its limit while the app in question is running. If on a discrete GPU dedicated memory is used almost to its limit and shared usage is growing, the part that does not fit in VRAM may be placed in ordinary memory.

Can dedicated GPU memory be increased?

Windows’ settings have no item that increases dedicated GPU memory. On a discrete GPU it is fixed by the amount of VRAM on the card.

On an integrated GPU, some PCs let you change a setting in the BIOS.

• Intel: An item such as “DVMT (Dynamic Video Memory Technology)” in the BIOS may let you choose the upper limit of usable memory, such as 128 MB, 256 MB or maximum. There is no setting that assigns a fixed amount; it only sets the limit.⁴
• AMD: Its desktop documentation recommends leaving “UMA Frame Buffer Size” in the BIOS at the default “Auto”. It says setting it manually may resolve problems in games not optimized for integrated graphics, and gives the example of 1 GB or 2 GB on PCs with 8 GB or more of memory. It warns that setting 2 GB on a PC with 4 GB of memory may reduce overall performance.⁵
• Intel Core Ultra Series 2 and later: “Shared GPU Memory Override” in Intel Graphics Software lets you change the share of memory the integrated GPU can use. The default is 57%, and at least 10 GB of memory is required. Intel warns that changing it can affect system performance.⁸

flowchart TB
        accTitle: Choosing a way to increase GPU memory
        accDescr: Shows that a discrete GPU is fixed by the VRAM on the card, and an integrated GPU can sometimes be changed with settings the PC provides. Intel uses DVMT in the BIOS, AMD uses UMA Frame Buffer Size in the BIOS, newer Intel products use Shared GPU Memory Override in Intel Graphics Software, and AMD recommends Auto.
        A["Type of GPU"] --> B["Discrete GPU"]
        A --> C["Integrated GPU"]
        B --> B1["Fixed by the VRAM on the card"]
        C --> D["Change it with settings the PC provides"]
        D --> D1["Intel: DVMT in the BIOS"]
        D --> D2["AMD: UMA setting in the BIOS"]
        D --> D3["Newer Intel: dedicated software"]
        D2 -.-> N["AMD recommends Auto"]

Figure 8: How you can change the memory depends on the type of GPU and the PC model.

Some PCs have no such BIOS item. The item’s name and location also differ by BIOS, so check your PC’s manual. If you add memory, the limit on ordinary memory the GPU can use also grows according to the formula. However, if the driver sets a smaller limit, it stops at that value.

Choosing which GPU an app uses

On a laptop with both an integrated and a discrete GPU, you can choose which GPU each app uses. On the Windows 11 (26H2) screen, the steps are as follows.⁹

1. Open [Settings] → [System] → [Display] → [Graphics].
2. In the list of custom settings for apps, select the row of the app you want to change to open it. Add an app that is not in the list with [Add desktop app].
3. Under [GPU preference], choose [Let Windows decide (Power saving)], [Power saving] or [High performance]. Your choice is saved as is.
4. Restart the app.

How the options are worded varies by PC. On a physical Windows 11 Pro (25H2) PC with two GPUs, the options were “Let Windows decide”, “Power saving (Intel(R) UHD Graphics 770)” and “High performance (NVIDIA GeForce RTX 4070 Ti)”, with the GPU each one refers to shown in parentheses.

flowchart TB
        accTitle: Steps for choosing which GPU an app uses
        accDescr: Shows opening Graphics under System and Display in Settings, selecting the app in the custom settings for apps, choosing Let Windows decide, Power saving or High performance under GPU preference, and restarting the app.
        A["Open Graphics in Settings"] --> B["Select the app in the custom settings for apps"]
        B --> C["Choose a GPU preference"]
        C --> C1["Let Windows decide"]
        C --> C2["Power saving"]
        C --> C3["High performance"]
        C1 --> D["Restart the app"]
        C2 --> D
        C3 --> D

Figure 9: You choose a GPU per app in Graphics in Settings.

This choice is meant for PCs with more than one GPU. After changing the setting, open the [Processes] tab in Task Manager while the app is running, and check in the GPU engine column of that app’s row which engine of which GPU is being used. Another setting related to how the GPU draws is explained in “What Is Windows’ Hardware-Accelerated GPU Scheduling?”.

Rules of thumb when you look at the numbers

Summary

• Dedicated GPU memory is memory used only by the GPU. On a discrete GPU it is the VRAM on the card; on an integrated GPU it is a small reserved portion of ordinary memory.
• Shared GPU memory is ordinary memory that both the CPU and the GPU can use. The large number is a limit, not the amount taken.
• The limit on ordinary memory the GPU can use is half of memory on PCs with up to 32 GB. On PCs with more it is above half, up to 80%.
• On an integrated GPU, a small dedicated and a large shared number is the normal state.
• When the video memory budget is exceeded, the app is intermittently suspended or fails to allocate memory.
• Windows has no item that increases dedicated GPU memory. On an integrated GPU it can sometimes be changed in the BIOS or elsewhere, but normally you leave it on the automatic setting.

References

Related Articles

Recent articles sharing the same tags. Deepen your understanding with closely related topics.

Related Topics

These are topic pages close to this theme. Starting from the article, you can go on to related services and other articles.

Services Connected to This Topic

This article leads to the following service pages. Start from the entry point closest to your situation.

Frequently Asked Questions

Questions commonly asked at the time of consultation about the topic of this article.

Does a large shared GPU memory number mean that much memory is taken by the GPU?

No. The number on the right of shared GPU memory is the upper limit of ordinary memory the GPU is allowed to use. The part not in use can be used as normal by the CPU and other apps. How much the GPU is using right now is shown by the usage figure on the left.

Why is shared GPU memory half of my PC's memory?

Because Windows sets an upper limit on how much ordinary memory the GPU can use. With the formula in Microsoft's documentation, it is half on PCs with up to 32 GB of memory and more than half on PCs with more. Depending on the driver, it can also be smaller.

Is there a way to increase dedicated GPU memory?

Windows' settings have no item that increases dedicated GPU memory. On a discrete GPU, it is fixed by the amount of memory on the graphics card. On an integrated GPU, some PCs let you change the limit or allocation in the BIOS, but both Intel and AMD usually recommend the automatic setting.

My integrated-GPU PC shows very little or zero dedicated GPU memory. Is it broken?

It is not broken. Most integrated GPUs have almost no memory of their own and use ordinary memory shared with the CPU as needed. That is why dedicated GPU memory is small and the shared GPU memory limit is shown as large. On some PCs the dedicated GPU memory field does not appear at all.

Why is the sum of per-process dedicated GPU memory on the Details tab larger than the overall usage?

Because memory shared by several processes is counted again in each process's figure. Check how much the GPU as a whole is using on the Performance tab.

Author Profile

Profile page for the article author.

Go Komura

Representative of KomuraSoft LLC

Focused on Windows software development, technical consulting, and bug investigation, with particular strength in projects where existing assets remain and in investigating failures whose cause is hard to see.