Basics GPUs: Power Consumption, Power Supply Conflicts & Other Myths | 2014 and 2020

Who measures a lot measures crap. This rule applies as long as one does not actively deal with the circumstances and requirements and is not prepared to invest a lot – mostly private – time in its exploration. I started working more intensively on the power consumption of graphics cards in 2014 and to find a really good solution. In this way I also came across many interesting insights and things that I would not have suspected.

At that time I wrote this article in the first-person form 6 years ago, because it describes the very personal, but quite rocky path behind what one sees in reviews as a finished result. Actually, in 2014 I only wanted to improve the measurement of the power consumption of the graphics cards tested by us, so I thought about how to get away from silly socket estimates with at least a manageable effort and achieve reliable results. What then developed from it… But let's start with the "Urschleim".

Simple current pliers, multimeters, first data loggers – always something was wrong and it was really to despair in places if each measurement yielded something different. So there was something else – just what? This is exactly where I want to start on the next page, so that Unfortunately I can't spare the reader a certain amount of theory. But it is precisely this knowledge that is so important in order to understand why one often simply cannot move forward with simple means. But don't worry, it won't be professorial or furry – and where it goes, I've simplified a lot of things.

First measurement setup with oscillographs in 2014, i.e. 6 years ago

Together with our partners from the industry, the measuring station was systematically rebuilt and expanded. Setups were tested and discarded, firmware changed and started again near zero. In short, for a development that had to be carried out during ongoing operations, this was quite a lot of work besides work. But it was really worth it. And if I'm to be honest, secrecy may be good for your own business, but transparency is even more important.

Especially when it comes to being able to justify the published readings in the articles. That's why I reveal a little "the secret" of how exactly the exact numbers of our measurements actually come about. Today's measuring station has of course changed, because in addition to the triggered oscillographs, I also rely on a sepated riser card for direct power consumption measurement on all rails with shunt and evaluation software, as well as a large, MCU-based Messtation, where every single supply voltage line can be logged in small time intervals and you can even measure everything in the direct interaction of GPU, CPU and motherboard on a total of 12 different rails at the same time. Rest ingenuation is not.