Believe it or not, it is possible to create graphics cards without dedicated memory. While integrated graphics processing units (GPUs) are commonly found in PCs without their own memory, we are referring to the possibility of releasing them as expansion cards. In this article, we will explore why this concept is not currently available on the market and the reasons behind it.
Typically, individual components like graphics cards have their own dedicated memory to avoid conflicts with the CPU for access to it and to enhance performance. However, we are introducing a concept that may seem peculiar to some but has been implemented in various systems throughout history. So, why not on a PC? Let’s find out.
Graphics cards without memory?
Although it may seem impossible, it is technically feasible to create graphics cards without their own memory. However, their performance would not be comparable to those designed for gaming, as they would use the system’s RAM accessed through the PCI Express port. In other words, they would not surpass the performance of an integrated graphics card. Nevertheless, the good news is that they could be developed in the form of an M.2 expansion card. But then, why are they not available on the market? Simply put, since they do not provide more power than an iGPU, they are considered redundant and impractical.
It is indeed possible to make graphics cards without their own memory, thanks to a peculiar characteristic of GPUs. These components are capable of working with two different types of memory: local memory and system RAM.
Local memory is accessed through the memory controller integrated into the GPU, and is usually found in VRAM. On the other hand, system RAM is accessed by the GPU through its DMA units. By using system RAM, a graphics card without its own memory could potentially be created, and would access the RAM through the PCI Express port.
However, it is important to note that such a graphics card would not perform as well as a dedicated GPU with its own VRAM. In fact, it would likely be inferior to an integrated graphics card. This is why such graphics cards are not commonly seen on the market, as they would be redundant and offer poor performance.
Some of you may be wondering whether using system RAM instead of VRAM would not be counterproductive due to increased latency caused by the distance between the two components. However, it’s important to note that GPUs have the ability to mask latency to some extent. In fact, even with VRAM, access times can be up to five to ten times slower than typical RAM, yet the card can still access its own memory without issue.
We will see it in the future, but not in the way you think
Disaggregated chips are bringing a new wave of change in the graphics industry, as integrated GPUs in processors are separating from the main chip. An example of this can be seen in desktop Ryzen 7000, which has a small graphics card integrated into the IOD chip. The separation of GPUs will be visible for the first time in Meteor Lake, but we cannot refer to it as a graphics card expansion since it will not have its own memory.
However, we cannot expect to see these units in M.2 form factors due to power limitations. Expansion cards can deliver a maximum power of 10 W, while a CPU can receive much higher power. As a result, graphics cards without memory would not only be limited by memory bandwidth but also in terms of clock speed due to low power consumption.
Therefore, there are no graphics cards without memory in the market because their performance would be so low that they would not be useful, wasting both time and resources. Moreover, integrated graphics have already taken over the entry-level segment and are gradually eliminating the need for additional hardware.