No, we have not gone crazy and it is that it would be possible to make graphics cards without their own memory. After all, these are the most used on the market and we are not just saying it, since the one integrated into your PC’s processor complies with said specification. However, we are referring to releasing them as expansion cards. Well, in this article you will learn the reason why nobody makes them and they are not for sale.
The reason why the different components have their own memory is to avoid contentions and fights for access to it with the CPU and because they perform much better having their own memory. This is the case with graphics cards. However, today we bring you a concept that may sound strange to many, but it is not so crazy, given that it has occurred in many systems in history, why not on PC? Well, we explain it to you.
Graphics cards without memory?
Well, even if it seems an impossibility to you, it would be possible to do them, however, they would not have the same performance as the units designed for gaming because they would use the system’s RAM for it, which they would access through the PCI Express port. That is to say, they could not be better than an integrated graphics card, but the good thing is that we could see them in the form of an M.2 expansion card thanks to it. Why don’t they see each other? Well, due to the fact that by not reaching more power than an iGPU, well, they are still redundant and stupid.
However, it is possible to do it and it is all due to a peculiarity and that is that GPUs are used to working with two different memories:
- One is the local memory that is accessed through the memory controller built into the GPU and is usually found in video memory, which is what we call VRAM.
- The other is system RAM, which is accessed by the GPU through its DMA units.
Now, many of you will wonder if this is not counterproductive due to the latency due to the distance between the two parties. Well, we have to think that GPUs have the ability to mask latency to a certain degree. Not surprisingly, even with the video memory itself, we can find access times between 5 and 10 times compared to the usual RAM and that are not higher than when the card itself accesses its own memory.
We will see it in the future, but not in the way you think
With the advent of disaggregated chips, we are going to see how the GPUs integrated in the processors are going to separate from the main chip. We have a preview of this with the desktop Ryzen 7000, where we have a small graphics card integrated into the chip that is the IOD, but we will see the separation for the first time in Meteor Lake, where although it is not possible to speak of a graphics card, expansion, it will be the first time we have a GPU separate from the main unit without its own memory.
However, we will not see these units in M.2 units and that is that we have left one element for the end to finish our argument and that is that the power delivered to said expansion cards is not greater than 10 W, while within the framework of the CPU can receive much higher power. So in the end, we would not only be limited to the capabilities of an iGPU by memory bandwidth, but also in terms of clock speed due to low power consumption.
And that is why there are no graphics cards without memory, since the performance would be low enough to be useful in any case and would therefore be a waste of time and resources. What’s more, the input range is already being absorbed by the integrated graphics and gradually disappearing.
