A Look At The Linux vs. Windows Power Use For A Ryzen 7 + Radeon RX Vega Desktop

[phoronix]

Phoronix: A Look At The Linux vs. Windows Power Use For A Ryzen 7 + Radeon RX Vega Desktop

Recently I have been posting a number of Linux laptop battery benchmarks including how the power consumption compares to Windows 10. If you are curious how these numbers play out on the desktop side and when using AMD hardware, here are some results for your viewing pleasure with a Ryzen 7 2700X and Radeon RX Vega 64 desktop system...

http://www.phoronix.com/scan.php?page=news_item&px=Windows-Linux-AMD-Power

[GrayShade]

Does the second graph have the wrong title?

I've been contemplating getting a desktop or maybe a small server kind of box for a while now. But I'm somewhat put off by the high power usage. Why exactly can these idle at ~10-15 W, while larger CPUs are 80 W or so (I realize that there's a dedicated GPU and maybe screen in these tests)? How high can I go while keeping under a say 20-30 W envelope?

[eydee]

Power users are using more power. Nothing unexpected, nothing to see here.

[GrayShade]

PCs use power, not users.

[starshipeleven]

PCs use power, not users.

Users use chemical power, PCs use electrical power.

[starshipeleven]

Kabylake/Skylake CPU eats about 3W idling (C3 package, C7 cores), if you disable C-States then it eats ~10W doing halt states (noop). 18 core Xeon eats away ~15W with 0.01 load. Motherboard/GPU typically draws much more power (10-50W) when PC is idle.

I tried to get my desktop power usage as low as possible when idle while back and absolute minimum that I managed was 35W (powertop)/55W (measured from power socket). Interestingly enough, even when suspended/"off" it drew 25/11W (you can't actually turn modern PC off unless you plug power cord).

That's a tad high for "off". What PSU are you using? I've got better results with picoPSU + power bricks rated for 80-160 watts, if you use a 400w (or more) PSU it's efficiency in the low 50w range is going to suck. It's just physics.

Apart from that yeah, most of the power comes from the mobo, and this means that you should disable in UEFI all the hardware you don't need/use (Sata ports you don't use for example), and hope that the chipset itself isn't a pig like most stuff paired with Atom boards. I mean seriously, I got like 25w at idle with an Atom on most industrial boards I tried. Whatafak.

Decent mini-itx boards with such picoPSU setups can still idle as low as 10w https://mattgadient.com/2016/07/30/b...10-watts-idle/

Here is the blog of a guy that did some research and hacking around to find and cut down power on a board https://ssj3gohan.tweakblogs.net/blo...-computer.html

Years later he did some testing on boards without hacking (yes it's not in english but it's easy enough to read numbers and graphs) https://ssj3gohan.tweakblogs.net/blo...aart-2014.html and most of the tested boards with a PicoPSU were able to idle in the 5-10w range.

[GrayShade]

I don't see efficiency curve for <50W load

Because it's too poor and they don't want to show it.

My NAS (Intel J1900, 2 HDD and change) idles at around 25 W, for what it's worth.

[starshipeleven]

I suppose so. It's few years old Seasonic S12II (unfortunately I don't see efficiency curve for <50W load).

Neat, also johnnyguru testing in low load (50w) confirm that this thing manages to remain in the "80 plus" efficiency range so it's not bad at all. It would decrease, but it starts from nominal efficiency so it should not be the culprit here (assuming your PSU isn't massively degraded with age). http://www.jonnyguru.com/modules.php...tory2&reid=185

This means your board is a pig and wastes power when "shutdown" or your meter is inaccurate. It's actually quite common for laptops, oddly. Most laptops I know must be kept attached to the power cord or they discharge the battery over time by like 5% per day (while the battery does not discharge over time if removed).

Idle load is OK, considering that you have a dedicated GPU.

[starshipeleven]

I never wrote that, quote starshipeleven, not me.

I never said that either. I think he just failed to quote your post, and all the text in that post is his own.

[AndyChow]

This is a great example of why 450 watt power supplies are usually more than enough. Or 550 Watt, 650 watt. The 800-1000 watt power supplies are ridiculous with current tech.