The consumption of graphics cards will go higher and higher and will exceed 700 watts in 2025

There is no doubt that the power consumption of graphics cards has seen a significant increase in the last two decades. To illustrate it, nothing better than a specific example, and for this we are going to go back to 2010, the date on which AMD launched the Radeon HD 6000. The Radeon HD 6970 was the top of the range of that generation, and its maximum consumption was close to the 200 watts in games.

In 2020 the Radeon RX 6900 XT arrived, one of the most powerful graphics cards of this generation, and when measuring its consumption in games it was seen that it was around 320 watts. As we can see, the latter is much more powerful, uses a much higher architecture and is manufactured on TSMC’s 7nm node, while the former used the veteran 40nm node.

Node jumps and new architectures are no longer enough to keep consumption at lower levels compared to previous generations, and thanks to AMD we know that this reality will go further with the next graphic generations. To illustrate this, the Sunnyvale company has published a graph in which we can see the curious evolution that the high-performance graphics card sector has experienced over the years.

Between 2005 and 2015 we lived through a fairly stable decade, in which process leaps and new architectures made it possible to create increasingly powerful graphics cards while maintaining stable consumption. As of 2015, there was an important turning point, and consumption began to grow enormously, a trend that will continue to grow, as we have said, and that in 2025 we could find graphics cards that will consume more than 700 watts.

A next generation graphics card will consume the same as a current full PC

At least according to the AMD graphics, and the truth is that this is really impressive, but we must give it the proper context to better understand what is happening and why this trend towards higher consumption makes sense and is “acceptable”. In this sense, we must bear in mind that:

• The GPU is an increasingly complex component whose performance has grown exponentially. The higher the performance, the higher the consumption, and this can perfectly imply greater efficiency as long as the rule of higher performance per watt consumed compared to the previous generation is fulfilled.
• GPU designs are no longer limited to the traditional concept, they now incorporate specialized cores that add greater complexity and trigger consumption, and we can also find other elements that improve performance, such as infinite cache in the case of AMD.
• Increasingly faster graphics memory is also used, something that improves performance and ends up shooting up consumption, as is evident. In this sense, the case of NVIDIA stands out, which mounts GDDR6X memories.

This trend towards greater consumption makes sense because of those three points that we have seen, and it is acceptable because it is accompanied, as we have said, by a greater performance per watt compared to the previous generation, and because we are at a stage in which the Ray tracing and AI applied to games have become two key pillars that accompany the classic raster. Games are getting more demandingusers want more power, and to meet those needs optimally it is necessary to make certain sacrifices.

Sam Naffziger, Senior Vice President, Corporate Partner and Product Technology Architect at AMD, delved into this topic by talking about the slow jumps to more advanced manufacturing nodesof increase of the occupied area at the silicon levela direct consequence of that greater complexity of the GPU that I have already explained to you, and also confirmed that they are analyzing the approach that NVIDIA is following, and that they aim to “do it much better” than it.

We will see how the situation evolves, but it is clear that we are approaching the limits of traditional computing in terms of efficiency, and that the transition to new designs at the hardware level, and to new tools at the software level, will be key to moving forward and improve in this aspect. One of the most important advances that we can expect in the medium term will be the transition to a modular design in the GPU sectora path that AMD has taken before NVIDIA.