Graphics cards tested in 2021: power consumption of modern GPUs re-measured

The power consumption is an important factor when buying and therefore also when testing a graphics card. ComputerBase has changed the test procedure in order to react to current developments in the GPU and monitor area and thus to be able to make better recommendations. The new results are very interesting in any case.

Table of contents

1. 1 New measurement of the power consumption of modern GPUs
   1. New test scenarios, still measured with PCAT
   2. New 2D load: consumption with moving desktop content
   3. New 3D game test procedures: three resolutions and FPS limit
   4. All changes at a glance
2. 2 current measurements on the Windows desktop, on YouTube and in games
   1. Power consumption during the “window test”
   2. Power consumption when playing games with and without FPS -Limiter
3. 3 The energy efficiency and the conclusion
   1. Performance per watt when playing with and without the FPS limiter
   2. Conclusion

The focus is on the gaming benchmarks in graphics card tests and purchase recommendations; no topic arouses more interest from the readers. But of course the judgment about a graphics card is based on more than just looking at the FPS. Volume and power consumption are also two important factors in a GPU consultation, and this article is about the latter.

In recent years there have been a number of developments that have had an impact on the energy behavior of modern graphics cards: monitors with higher frame rates, videos with new codecs that are accelerated in hardware and new energy-saving functions have made new measurement methods necessary.

New test scenarios, still measured with PCAT

Accordingly, the editors have made changes to the tests, which are now used in graphics card tests and which are presented in this article. The hardware with which the power consumption is tested remains the same – Nvidia's PCAT system is still the best solution with quite detailed results for little effort. But the scenarios in which PCAT is used changes in detail.

The measurement of the idle consumption with one and two monitors remains untouched. The same applies to the video playback on YouTube, which still consists of two test series with a 2,160p60 video and SDR and HDR.

New 2D load: consumption with moving desktop content

Nevertheless, there is a new series of measurements concerning 2D mode. Because the pure idle mode and videos are not the only tasks that a computer has to do away from gaming. After all, people sometimes work on the PC and surf the Internet and the GPU load is higher than when the mouse is not moved a millimeter.

Neither of these can be simulated in an absolutely realistic way, but it can be represented to some extent. To do this, the Windows 10 control panel window is always dragged to the same size in the test and the window is then alternately moved from left to right in the same rhythm over and over again. Apart from the window movement, the GPU load is generated by showing the halved full-screen windows of Windows 11. The resolution for this test is 2,560 × 1,440, the refresh rate is 144 Hz. This means that the GPU has to transmit more pixels than 3,840 × 2,160 at 60 Hz, which requires a higher clock rate and thus demands the clock behavior of the GPU at partial load is.

The resolution and refresh rate play an important role in this test, but not in complete idle mode. Modern GPUs do not care whether it is Full HD at 60 Hz, Full HD at 360 Hz, WQHD at 144 Hz or UItra HD at 60 Hz. All graphics cards can then reduce the clock rate as much as possible. Whether this will also apply to the upcoming Ultra HD and 144 Hz displays remains to be seen.

New 3D games Test procedure: three resolutions and FPS limit

The biggest changes affect the games area. So far, ComputerBase has always measured the power consumption in graphics card tests in three different games exclusively with Ultra HD resolution and with its average and the performance ratings then the performance per watt ratio for the three resolutions Full HD, WQHD and Ultra HD calculated. While the method produced reliable measured values ​​and results, it was far from the optimum for various reasons.

So actually all graphics cards in high resolutions like Ultra HD run into the power limit. As a result, the three different games never really showed a difference. In addition, if a graphics card in WQHD and, above all, Full HD requires less energy, this was simply ignored in the test – although this is definitely the case with some graphics cards, as the new test procedure will show. On top of that, the performance-per-watt rating has always been created from the performance rating for the respective resolution, but only with the power consumption from the Ultra HD measurement. And last but not least, the rating represents a good average of all games, but only that. In the individual games, the graphics cards actually behave differently and therefore only partially match the watt measurements.

All three resolutions get their own measurement

The new test procedure addresses all of these points effectively. The power consumption of each graphics card is measured in the resolutions 1,920 × 1,080, 2,560 × 1,440 and 3,840 × 2,160 and the respective performance-per-watt rating is then created with this value. And this eliminates both weaknesses in the old measurements in one go.

With a maximum of 144 FPS there are completely new challenges

< p class = "p text-width">In addition, there is also a completely new series of tests in the games section. The main measurement is carried out independently of the graphics card with maximum GPU load and accordingly with an unbraked frame rate. However, this is a scenario that does not apply to every player. Because some limit the FPS in games for various reasons, be it with an FPS limiter or VSync.

In such a scenario, the GPU has to show how well it can cope with a high load, but not with maximum load. Then it depends more on how effectively a graphics card can reduce the GPU clock and voltage in order to achieve the required frame rate. Fast models are required significantly more in this measurement, while slow models may not be affected at all – since the required frame rate is not achieved anyway. In addition to the pure measurement result, there is another diagram that deals with energy efficiency (performance per watt) with a limited frame rate.

In order to meet the trend of increasing refresh rates, ComputerBase limits the frame rate to 144 Hz regardless of the resolution. If a graphics card can achieve more frames per second, it is slowed down accordingly and then has to For a good test result, reduce the clock and voltage as much as possible. If an architecture is not so effective at this, the measurement result will be correspondingly poor.

Doom Eternal is perfect for measurements

The game measurements are all performed with Doom Eternal for a variety of reasons. One reason for this is that the technical side of the shooter will probably not change after the ray tracing update and the results will be correspondingly robust. Another is that Doom Eternal has no problem with demanding fast graphics cards even in 1,920 × 1,080. High-end models also have a high GPU load, albeit not necessarily a maximum. In addition, the game offers high frame rates and is therefore well suited for the series of measurements with a limited frame rate. It should also be noted that both AMD and Nvidia GPUs get along well with the Doom Eternal and achieve a comparably high performance.

Doom Eternal is consistently tested with the Ultra Nightmare preset and thus the maximum graphic details – but ray tracing and DLSS remain switched off. The frame rates are taken from the normal GPU benchmarks of the game, which are current for the respective test course. For this purpose, the benchmarks can be taken from the test of the GeForce RTX 3070 Ti, the missing tests with the Radeon RX 6600 XT from the corresponding article on the AMD graphics card. If the test course is renewed in the future, the new measured values ​​apply accordingly.

Even if the new test procedure addresses and fixes many weak points, a new one is inevitably added. Since the power consumption is only measured in one game, the values ​​only apply to Doom Eternal. Experience has shown that the measured values ​​in other games are more or less the same, but this only applies to the tests with an unlimited frame rate – the results, which are limited to 144 FPS, are very different in other games. However, this is a disadvantage compared to the numerous improvements in the new test procedure and is therefore accepted.

With Windows 11 in the future

The operating system is also new, because instead of Windows 10, Windows 11 is already used for the power consumption measurements. The latest Microsoft product is not yet officially developed, but already in a usable condition and also the future. The next regular new graphics card test system, which will be revised later this year, also uses Windows 11. The drivers include the GeForce 471.68 and the Adrenalin 21.30.11.01 and thus the AMD driver for Windows 11 delivered via Windows Update for use.

All changes at a glance

On the next page: Current measurements on Windows desktop, at Y ouTube and in games