Why does Windows take RAM from you to use it on the graphics card?

You may have noticed, if you look in the Windows 10 task manager, that the operating system is taking part of the RAM to give more VRAM to the graphics card , even if it is dedicated and has its own memory. ¿ Why is this happening and how can you avoid it ? We will tell you everything below.

When your computer does not have a dedicated graphics card with its own VRAM memory, it is normal for the system to assign part of the RAM to the graphics, since it is necessary for it to work. However, when you have a dedicated graphic with its own memory, why does Windows keep allocating part of the RAM memory of the computer?

RAM memory shared with the graphics

To see what we are telling you, you just have to open the Windows Task Manager, access the “Performance” tab and select GPU on the left.

If you look at the lower area, where the information on the graphics card appears, “Dedicated GPU memory” appears, which is, indeed, and taking this example, the VRAM of the graphics card, 8 GB. But just below it appears « Shared GPU memory» , where it appears that 300 MB of 16 GB allocated are being used, and in fact just to its left appears the term «GPU memory», where you can see that 4 are being used 24 GB, which is nothing but the physical 8 GB of the graphics card added to the additional 16 GB allocated by Windows.

This means that Windows is using part of the computer’s RAM to allocate it to the graphics card, but why is it doing this?

Shared memory is in virtual reality

To begin with, we will tell you to worry, it is not that Windows is “stealing” RAM memory at all (well, actually, but very little). The reality is that that shared graphics memory, Windows uses it for the processor iGPU because we do not have the graphics disabled in the BIOS (so if you want this to stop happening you just have to go to the BIOS and disable the iGPU , although not We recommend it just in case).

In any case, deactivating the iGPU is not something we recommend because if the dedicated graphic fails at any given moment, we will be quite helpless because we will not be able to use the integrated one in the meantime and we will necessarily need another dedicated graphic.

If you want to do it anyway, you must enter the BIOS and go to the advanced settings section. From here it depends on the manufacturer of the motherboard, but normally we will have to enter the chipset configuration where we will find something related to “Internal Graphics”, where we can deactivate it.

This small portion of RAM that Windows “steals” from us is actually a few MB, not the 16 GB it has “assigned” because they are not assigned as such, it simply has them available but not consumed to assign to the graph in the case if necessary. And, a curious fact, if in the BIOS we selected the iGPU as the primary graphics but having a dedicated one connected, Windows would use the VRAM of the dedicated graphics for the iGPU . In any case, this is not the common thing, but I use DDR, so we always have to be clear that the VRAM that is assigned in this way has nothing to do with the dedicated memory of a graphics card

GPU VRAM or VRAM with RAM

In this case it is true that we will gain performance in sequential writing and reading, something where the GDDR sins of a very poor performance, but it is that its function is not based on these terms, but on the copy, where it multiplies by several times the performance of the DDR4 in Quad Channel, not to mention the difference with Dual Channel. Although the part reserved for the iGPU of the system memory is currently miniscule, the truth is that the amounts of this must increase in a short time due to the increase in the average size of the RAM and especially the requirements of the new iGPUs.

So what is expected is that we will see 2 GB of allocation or perhaps up to 4 GB for computers that have 32 GB or more, figures that will soon be standard on any average PC if the advance with NAND Flash meets its deadlines. In any case, the point will be reached where the iGPUs will have fully coherent access to the RAM with the CPU, this will avoid having to borrow part of the RAM and the amounts used at all times will vary according to the workload.

GPGPU and RAM usage by graphics

GPUs today are not only used for rendering graphics, but also for computing tasks, where in many cases it needs to communicate in the same memory space as the CPU. The key is that for two hardware units to be synchronized they have to use the same memory and that is why all GPUs today can work in two memory spaces:

  • A non-CPU coherent memory pool that corresponds to your local memory or VRAM.
  • A memory well consistent with the CPU, which corresponds to the system RAM.

Many programs today use the capabilities of the GPU not to render graphics but as support for the CPU, for this they make it interact at the same level of memory. Which is possible with the GPU’s DMA drives having access to RAM through the PCI Express port.

The way it works is simple, the CPU invokes the GPU through a Compute Shader program, which is solved by the GPU that writes the result back to the system RAM instead of its VRAM. Many algorithms have parts that are sped up not by one, but by several orders of magnitude when run on a GPU versus a CPU.

 

by Abdullah Sam
I’m a teacher, researcher and writer. I write about study subjects to improve the learning of college and university students. I write top Quality study notes Mostly, Tech, Games, Education, And Solutions/Tips and Tricks. I am a person who helps students to acquire knowledge, competence or virtue.

Leave a Comment