Skip to main content

The Math Behind GPU Power Consumption And PSUs

The End Of A Long Journey

The usual more-is-better rationale of PSU output can’t be upheld here. As is so often the case, the important consideration is good balance. Spikes in power consumption demonstrated by modern graphics cards (and illustrated by us) need to be seen for what they are: a challenge to the quality of a PSU, not its maximum wattage. If you want to be protected from bad surprises and potential long-term damage, then good quality is what you need. And the decreased efficiency of oversized PSUs at idle shouldn’t be ignored either.

So, what should an ideal PSU for today’s powerful graphics cards look like, and how should it be measured?

The average power draw under load over a certain period of time, as well as the power consumption of the rest of the system, are the only measurements that are really important in PSU selection, not the brief spikes. You've hit the ideal power range if the sum of all components amounts to approximately 75 percent of the maximum capacity during a stress test.

Component quality makes for an interesting story, too. It needs to be stressed that a great brand name just isn’t everything. What’s important is that the capacitors in question are suitable for the task at hand. So, the type and features of the capacitors used on the secondary side are what really need to be evaluated.

The statement that a PSU uses capacitors made exclusively in Japan might make for great marketing, but it doesn’t necessarily make for a great PSU. This is where the spikes we discussed come into play. They're what ultimately determine how consistent the supply will be and how long the components will last.

You can learn a lot by measuring a lot. This is my personal mantra that will stick with me after this year-long journey. After a lot of direct contact with graphics card and PSU manufacturers, talks with their development departments and even visits to their factories, I have personally achieved a much clearer picture of the many problems and how they relate to each other. This has been a slow, steady and certainly worthwhile process.

In addition, the cooperation with the measurement equipment manufacturers has done its part to find the origins of several problems. This is why I’d like to publicly thank all participants, yet again. It’s not a sign of weakness to profit from other people’s knowledge, but being too proud to ask certainly is.

Along those lines, it would certainly be helpful if PSU reviews didn’t just use Chroma protocols and ohm loads, but created the same spikes that are delivered by the actual technology that PSUs need to contend with. That’s going to be the point when puzzling and hard-to-measure problems such as lost Power Good signal flags from the motherboard are finally a thing of the past. After all, one thing's for sure: many hard-to-explain PSU failures and system power losses have simple explanations, namely the extremely fast load fluctuations with their high peaks, PSUs that didn’t take them into account in their design or the selection of suitable capacitors and, finally, the wrong PSU choice.

In summary, after a whole year, I’ve finally arrived where I meant to be after a few weeks. In hindsight, this was never realistic. However, the long and arduous journey has provided both the cause and the motivation for many insights and interactions that I was able to collect on the way, and I’d never have had them otherwise.