This is still the header! Main site

Power Supplies are Underrated

2024/06/15

Power supplies are one of the more boring parts of desktop computers. They provide a range of voltages for the parts that actually matter. The main thing they can do is not being extremely loud.

Or at least, that's what I thought before conducting some experiments. They had somewhat unexpected results.

Background

My main desktop computer started its life as a discounted Steam Box in 2016. Does anyone remember Steam Boxes? (Yes, there was a discount for a reason.)

I eventually decided to move the parts over into a way larger case so that I can add hard drives and have better cooling. (It's kind of fun how tiny the micro-ITX motherboard looks like in it.)

What did not change though is the power supply. It was a model designed for the smaller case so it didn't even have proper attachment points for this one to bolt it down. (Turning the machine upside down at any point would have had questionable consequences.)

I have recently decided to add a GPU to it (more about that in a later post!). I was somewhat concerned, though, that this tiny power supply would not be enough for a relatively more beefy video card, so I used the opportunity to also obtain a power supply somewhat more appropriate for the build.

Measurements

I started by just taking samples of the existing situation, the 350 watt power supply, and the existing GTX 950 GPU.

Here are some benchmark results too! I was using FurMark to give the GPU something to do; the rendered graphics look pretty cool too.

Then, for fun, I removed the existing GPU first and repeated similar measurements.

As you can see I totally forgot to add an experiment where we are only running the CPU all the way. Nevertheless, it's interesting how, when launched, it consumed 90 watts for the "both" case., but dropped down quickly to 77 again; it's probably limited by heat dissipation.

Nothing surprising so far. Obviously, an integrated GPU will consume less power than a discrete one. It's also interesting to note that the latter was taking 10 watts doing (approximately) nothing, too.

Now, we swap the power supplies! The new one ends up being slightly over-provisioned, with a total capacity of 750 watts. Definitely not breaking a sweat here.

Interestingly, we are saving 4-5 watts of power compared to the old power supply, despite this one being larger???

Finally, we added a new GPU, a 3060.

Here are the results:

Are the FPS counts Good Science? Most definitely not. As it happens, the screenshots are from different output resolutions, so they are completely useless. (On the other hand, new GPUs are definitely cooler than... less new GPUs.)

Why was this even interesting?

The main takeaway is that you can save some power by just swapping out power supplies!

(You can also help avoiding crushing the rest of your computer, for extra points.)

While the difference in power consumption is definitely less impressive than the 10 watts I remembered when I decided to write this article, at least we're doing Good Enough Science to catch this in the process.

Also, 5 watts is around an entire Raspberry Pi!

I also tested out a new and interesting way of taking notes by recording sound and sticking the results into Whisper later. (You might guess what piece of hardware the model was running on this time.) 


88
00:05:20,540 --> 00:05:23,320
And now we are going to start the CPUs.

89
00:05:31,470 --> 00:05:32,150
Numbers.

90
00:05:35,290 --> 00:05:35,970
255.

91
00:05:36,050 --> 00:05:37,650
240 something.

92
00:05:41,320 --> 00:05:43,460
Still not especially too bad.