Shared GPU Memory is one feature offered by some modern graphics cards, especially those based on NVIDIA Turing or AMD RDNA architecture. This feature allows the graphics card to use a portion of the system memory (RAM) in addition or reserve to the video memory (VRAM) attached to the graphics card itself.
Thus, the graphics card can have more space to store graphics data, such as textures, shaders, and frame buffers, which can improve graphics performance and quality in some scenarios.
However, this feature also has some drawbacks and limitations that you need to know before using it. In this article, we’ll explain what shared GPU memory is, how it works, when you should use it, and what are its advantages and disadvantages.
How Does Shared GPU Memory Work?
To understand how shared GPU memory works, we need to know some basic concepts about memory and graphics cards. In simple terms, memory is the place where data is stored and accessed by the processor, both the CPU and the GPU. There are different memory, such as DRAM, SRAM, GDDR, HBM, and others, which have different characteristics and functions.
System memory, or RAM, is the most common type of memory used by the CPU to store and retrieve data. RAM typically uses DRAM technology, which means dynamic memory with random access. This means that data can be accessed randomly from any location in memory without having to follow a specific sequence. However, this also means that data will be lost if no power flows into the memory.
Video memory, or VRAM, is a type of memory specifically used by GPUs to store and retrieve graphics data. VRAM typically uses GDDR or HBM technology, which are variants of DRAM optimized for high speed and bandwidth. This means that data can be moved quickly between the GPU and VRAM, which is important for producing images with high resolution and frame rates. However, this also means that VRAM is more expensive and harder to produce than RAM.
The graphics card is the hardware responsible for processing graphics data and sending it to the monitor. Graphics cards usually have GPU, VRAM, and some other components attached to a circuit board. The GPU is the brain of the graphics card, which performs complex mathematical calculations to produce images. VRAM is the place where the GPU stores the graphics data needed to produce the image, such as textures, shaders, and frame buffers.
Now, let’s see how shared GPU memory works. By default, the GPU can only use the VRAM attached to the graphics card as a memory source. The amount of VRAM available depends on the model and specifications of the graphics card. For example, the NVIDIA GeForce RTX 3080 graphics card has 10 GB of VRAM, while the AMD Radeon RX 6800 graphics card has 16 GB of VRAM.
However, there are cases where the available VRAM is not enough to store all the graphics data required by the GPU. For example, if you run a game or graphics application that is very demanding, or use very high resolutions or graphics settings, you may experience a phenomenon called VRAM bottleneck. This means that VRAM becomes full and cannot hold more data, which can lead to decreased performance and graphics quality, such as stuttering, popping, or artifacts.
To solve this problem, some modern graphics cards offer a shared GPU memory feature, which allows the GPU to use a portion of RAM in addition or reserve for VRAM. Thus, the GPU can have more space to store graphics data, which can improve performance and graphics quality in some scenarios.
The way shared GPU memory works is as follows:
- First, the GPU will try to store all the graphics data it needs in VRAM, as usual.
- Second, if the VRAM becomes full and cannot hold more data, the GPU will pick up less important or rarely used graphics data, and move it to RAM. Graphics data that is transferred to RAM is called shared GPU memory.
- Third, if the GPU needs graphics data that has been moved to RAM, it will retrieve it from RAM and move it to VRAM. This process is called memory swapping.
In this way, the GPU can use RAM as an alternative memory source, which can increase the amount of graphics data that the GPU can store and access.