[Hardware] Will current NVIDIA and AMD graphics cards be able to power games in 5 years? We explain the techniques used

[Draeno]

 

As time goes by, more and more powerful graphics cards are needed and it is the component that always has a grosser evolution generation after generation. But is there any way that you can predict what future, present and future graphics cards will be like? Well yes, through the games that are in development and their computing and bandwidth needs and in this article we will tell you the keys to all of this, which are the operations per pixel and the overdraw

We must start from the fact that not all games have the same technical requirements, but from time to time technically cutting-edge games appear whose development will determine how the most powerful graphics cards should be once they are released. The objective? Whoever makes the title work with the best graphics on the market in the best way will end up being the winner.

How to predict the graphics cards of the future?
To do so, we must take into account two important points, the first of which is the necessary computing power, which is measured through operations per pixel. Although the number of polygons per second must also be taken into account, this is discarded nowadays as it is executed in the same units and due to the fact that the number of pixels per frame will always be higher than the number of polygons.

The second element is the overdraw, which indicates how many times the image buffer will be accessed and, therefore, will have an impact on the necessary bandwidth. In any case, we must clarify that although there are techniques to reduce the cost of both elements, what we are going to focus on is the raw power in both aspects.

Pixel Overdraw
Overdraw is understood as the fact that the value of a pixel in the image buffer is changed several times during the construction of a frame. This is due to the fact that when building a three-dimensional scene, the polygons that form it are not drawn in a specific order and without the relevant optimizations, it is possible to draw several objects on the same pixel. Other times, instead, it is due to post-processing calculations or interaction between them. For example, an object reflecting a light source or casting a shadow on another.

The fact is that from the moment the framebuffer or image buffer is outside the GPU and in the VRAM for each pixel that we modify we need to access said memory. The color calculation is done in the GPU cores and the modification is carried out by the ROPS, units that copy the information from the L1 data cache to the VRAM in parallel. Normally, the bandwidth is related to the number of these units.

In other words, in each frame the same pixel will be redraw today dozens of times. This should not be confused with the necessary computing power. Obviously, the number of operations per pixel in Doom will be much lower than in Cyberpunk 2077. We also don't have to forget about the increases in screen resolution that have occurred over time that have multiplied the number of pixels that a card draws. graphics in your VRAM.

 

https://hardzone.es/noticias/tarjetas-graficas/potencia-tarjetas-graficas-overdraw-opp/