Which has the most energy efficient GPUs? NVIDIA or AMD?

One of the biggest concerns for designers of new GPUs and therefore graphics cards is energy consumption, since the PCI Express interface brings with it limitations in terms of the power limit that they can transmit. This is aggravated if we take into account that the energy consumption to achieve the highest VRAM bandwidth, measured by picojoules per transmitted bit, is increasingly large and therefore leaves the processor that dominates the graphics hardware with less and less margin. on the energy side.

That is why the challenge of the three main manufacturers of graphics hardware: Intel, AMD and NVIDIA; It is no longer the fact of always designing new GPUs that are those that draw more pixels on the screen, calculate more polygons or more TFLOPS, but rather that they seek to get the maximum possible performance with less and less available energy.

What are the frames for joule?

If we talk about the delta in energy efficiency between two graphics cards, then we need to compare the same game running under the same conditions in both in terms of graphics settings and resolution. In such a way that both take exactly the same image and we can measure the performance of each one according to the speed at which the frames are taken, obviously the one that renders the most frames the winner.

From two concepts such as frames per second and watts, which are nothing other than joules per second, we can get the joules per frame. Obviously, the graphics card that consumes the least joules when generating a frame will be the one that will be more energy efficient under said game.

How are the frames per joule of a graphics card measured? The way to do it is very simple, we only have to divide the consumption of the graphics card by the number of frames per second that it obtains and thus in a simple way the energy performance of each GPU in particular is obtained.

The best way to measure energy efficiency

But isn’t it much better to compare the technical specifications of two graphics cards on paper? Not really, since the figures given are theoretical maximums that can only be given under ideal conditions that, in 99.9%, are not met for one thing or another. Let’s not forget that when generating a frame, the CPU is also important, an element that must be the same in all comparisons, as well as the speed of the RAM.

This is why it is more efficient to compare two GPUs with each other in a real-world scenario. In addition, this allows us to find the bottlenecks within the same hardware and to know if it is limited by bandwidth of the VRAM or by the calculation capacity of the graphic processor itself.

Frames by joule and new architectures

At present, all PC graphics processor works at variable speed, this means that depending on the workload that it has, it is put at a higher or lower speed. We must not forget that video games are increasingly complex visually and this means an increase in the amount of data to be processed. That is why if a game is relatively old, our GPU is not going to step on the accelerator to render it, after all the standard frequency in frames per second is 60 Hz.

So when buying a new graphics card it is much better to opt for newer architectures, not only because of the fact that new games make use of their most cutting-edge technologies included in them, but because of the fact that energy efficiency is much older. If we were to fix the resolution, graphics settings and frame rate, we would see that an RX 6000 from MMD would be more efficient than an RX 5000 of the same brand and even an RX Vega. And the same can be said of NVIDIA where the GeForce Ampere are more efficient than the Turing and these than the ones with Pascal architecture.

Each sheep with its partner, each GPU with its resolution

The frames per joule always give us very clear information about why there are high-end GPUs and what is the reason why both NVIDIA, Intel and AMD in their marketing recommend certain graphics cards for some resolutions and others.

By simple observation we will see that a mid-range or lower graphics card will not be able to move a large frame rate at higher resolutions than it is intended for. If we extract the frames by joule we would see that they are less efficient as we increase the resolution of the output of the games. The reason is that the power consumption is maintained, but the number of frames decreases . Hence, for example when generating the same frame at 4K an RTX 3080 is more efficient than an RTX 3070.

Nor can we forget the use of the most advanced technologies. Newer GPUs often add to them via different types of accelerators, which perform the function for a much lower energy cost. When these small units are not in the graphics processor then the energy efficiency will end up being much lower.

Comparing different GPUs in frames by joule

In the two tables above we can see the evolution of the frames by joule on the different generations of graphics cards from both NVIDIA and AMD, which allows us to know how they have evolved in terms of energy performance, but also how higher resolutions affect performance.

The results obtained give us a very specific clue and that is that despite the advances in forms of new architectures, the performance measured in the frame per joule in each GPU has not been duplicated in all this time, which would demonstrate the need for implementation of artificial intelligence algorithms such as NVIDIA DLSS, Intel XeSS or AMD FSR in order to increase the frame rate in certain resolutions, rendering at a lower internally, and thus increase the number of frames per joule.

To finish we have not included the energy efficiency of the Intel and AMD GPUs, the reason is that being part of the same chip as the CPU and sharing the power supply system is much more difficult to break down.