Ryzen 7 Mini-PC makes a power-efficient VM host | Hacker News Reader

Ryzen 7 Mini-PC makes a power-efficient VM host

michael.stapelberg.ch

22 pointssecure12 comments

Power efficiency & power management really need to be a focus. I have an old Gigabyte GA-AB350-Gaming-3 I got when I eventually upgraded to a Ryzen 1700 (now 5800), and I: a) cannot get this system to idle below 80W after stripping non-essentials, b) suspend or sleep in any way (the system goes down but I literally have to pull the plug, wait, and replug it in to turn back on). I've tried undervolting, spent hours trying to tune power profiles &making sure maximum PCIe aspm link savings are active; the system is just a brute, and unmanageable. It feels so cursed & has been such a leaden disappointment I've tried to carry for so long; it has truly shattered my faith in AMD in general.

I'd rather not get a G core, but hearing that this system idles at 10W is incredible. That's what my 8600t HP tiny PC's idle at. That would be stellar.

I'd love to see more reviews and write ups include power consumption, and also ideally suspend capabilities. Ideally wwol would also be verified working.

>cannot get this system to idle below 80W

You can't do anything about it. It's a limitation of your Ryzen's chiplets based design where the ineficient IO die sucks a lot of power leading to poor idle efficiency.

Intel chips, Ryzen laptop chips and the new G chips like the one in the article don't have such issues because they're a monolithic die.

That doesn’t seem right. The tdp for the 1700 is 65w… no way an io die is consuming most of that. Here’s a comparison of a system with a ryzen 1700 idling vs intel contemporaries which don’t have a separate i/o die: https://www.bit-tech.net/reviews/tech/amd-ryzen-7-1700-revie...

Note that it’s only 7W greater at stock clocks.

Yes it does.

You're looking at chips from 2017 when Intel had stinkers and under load, but what I said is true, the IO die has high idle power draw compared to modern monolithic designs, which gets hidden away under load.

Just Google if you don't believe me, plenty of older desktop Ryzen owners complain about higher idle power draw compared to Intel.

The ryzen 1700 is also from 2017. Intel’s cpus still dominated at the time for anything single-threaded, they were just bad at multithreaded workloads. I’m not saying i/o die power consumption isn’t higher, it is for sure. And I agree, it doesn’t go down at idle because it can’t turn off or do power gating.

What I’m saying is that a computer with a cpu that is 65W TDP (from a time when amd’s TDP was close to being accurate as ~ max power consumption under load), the i/o die (which is part of that 65w TDP; which is for load) cannot possibly be the main reason his computer is idling at 80W. Especially when I linked an instance of a system also with a ryzen 1700 that was idling for 57W and with a similar configuration as an intel contemporary only being 7W greater at idle.

The 65W tdp you keep bringing up is under load and data is from 2017 when I tell was still on 14nm, but we're talking about idle power draws here and Ryzen looses to Intel in most cases in most modern data in the <10nm era.

Here:

https://news.ycombinator.com/item?id=32809852

https://youtu.be/JHWxAdKK4Xg?si=OFx6puKRSc1TYSX8

For both the 1700 and 5800X, the i/o die uses ~12W at idle, and 20W max at load (assuming he’a doing something that keeps the i/o die at max power consumption when everything else is idle).

This leaves us with 60W-68W unaccounted for at idle. Even in the worst case for i/o power usage that’s 75% unaccounted for.

I keep talking about TDP and load power because even in the case where the cpu isn’t using lower power states correctly for whatever reason, the i/o die cannot possibly be majority of the 80W power usage.

Source for power usage of i/o die:

1700 (same i/o die as 1300x/1500x): https://www.anandtech.com/show/11658/the-amd-ryzen-3-1300x-r...

5800x: https://www.anandtech.com/show/16214/amd-zen-3-ryzen-deep-di...

What i do not get is,why not add condensators and computation only on filled condensator and memory fetch completed?

I kind of doubt the 10W number this article states. Other reviews of this CPU I've seen say that the CPU alone idled around 20W. It may be that ASRock might be doing some aggressive tuning in the BIOS to get the CPU down to 10W, which would explain that doing any work at all causes it to jump to 50W. Could also be a function of this review using Linux while the other reviews are using Windows.

Either way, I suspect with a few VMs and the host all at idle, power consumption would average 50W for the CPU and 75 to 100W for the whole system.

2 days ago

https://news.ycombinator.com/item?id=40864364

Seems strange that they compare this to a raspi in the conclusion rather than one of the fully integrated mini PCs with a soldered down laptop CPU, eg from Minisforum.

Besides, the barebones unit is over €200, the 8700G is €300, and RAM is €150-250, for a total of €650-750. Meanwhile an 8GB Raspi 5 is less than €90 all-in and a 7 node cluster of these would use about as much power in total as the 8700G does on its own.

I've had a dozen Pis over the years, and every single time I'd prefer an old laptop or server (amd64). It turns out that things like casing, power management, platform expandability, standardisation, sensible storage devices, things not made by broadcom, etc. are all desirable features.

Pis are toys, fun but frustrating if you want reliability.

Reply on news.ycombinator.com