Low power server

[Dunuin]

Any recommendations for some used server mainboards with best bang-for-the-buck in terms of low-power while still supporting things like IPMI and ECC? I don't care about the size. Would put those boards in some existing 4U chassis anyway that are now running E3 v3 Xeons.
Performance of those E3 v3 Xeons is still enough. Just would be nice to be able to use more than 32GB of RAM while saving some electricity.

 

[zenowl77]

Any recommendations for some used server mainboards with best bang-for-the-buck in terms of low-power while still supporting things like IPMI and ECC? I don't care about the size. Would put those boards in some existing 4U chassis anyway that are now running E3 v3 Xeons.
Performance of those E3 v3 Xeons is still enough. Just would be nice to be able to use more than 32GB of RAM while saving some electricity.

that could be difficult, pretty sure the memory controller is in the cpu for intel so whatever board you get unless you upgrade the cpu you're stuck with the 32gb maximum. (unless we are talking a dual/quad socket board then you would be limited to 32gb per cpu but power usage would be a lot higher)

 

[Dunuin]

that could be difficult, pretty sure the memory controller is in the cpu for intel so whatever board you get unless you upgrade the cpu you're stuck with the 32gb maximum. (unless we are talking a dual/quad socket board then you would be limited to 32gb per cpu but power usage would be a lot higher)

Yes, I thought about keeping the quiet 4U cases, ATX PSUs and SATA SSDs and replacing the mainboard + CPU + RAM. Right now two Supermicro X10SLL-F + Xeon E3 1330v3 + 4x 8GB DDR3 ECC I bought like 4 or 5 years ago for 130€ each. Without any guests they idle at 20W but with as much guests the 32GB RAM can handle (CPU at maybe 15-30% load) it shoots up to 35-43W. Would be nice to run two of similar sized nodes + qdevice 24/7 in a cluster but then watthours add up.

 

[celemine1gig]

It is not a power-efficiency champion, according to some comments, but it might be an improvement (https://www.gigabyte.com/de/Enterprise/Server-Motherboard/MC12-LE0-rev-1x):
https://www.ebay.de/itm/132936385331

I don't think you'll win much, considering the mere idle power consumption, but over the runtime, the architectures' efficiency might count. Always depends. Anything with BMC is hard to find in a power-efficient form.

 

[zenowl77]

an asus board might be a better option, especially with the EPU feature, many report a fairly significant drop in watt usage using it as it decreases voltages slightly (usually by 0.02-0.03 i believe) and this makes for usually a 7-11w drop on both idle and load usage, some users report even more but i would guess it really depends on your system watt usage and how much it would normally consume i have seen one report showcasing a load watt drop from 150w to 83w with the epu enabled so i would guess it really has a big impact on lighter loads watt usage too, and it seems it can drop system power up to 30% overall with all the features of the EPU, which is obviously a huge plus for us using proxmox having the systems running 24/7

 

[celemine1gig]

If you cannot deliver any real technical background, as to why that ASUS feature would be so groundbreaking, concerning power-efficiency-optimization, then I'm afraid you just repeated Asus marketing babble.

Sorry if that sounds harsh, but if you use phrases like "usually a 7-11w drop on both idle and load" without any reference values, then this is basically nonsensical. So if I had 7-11W idle consumption, the Asus feature would put me to 0W?
Furthermore "showcasing a load watt drop from 150w to 83w" (if those values should indeed be real) would hint at rather horrible settings before and maybe slightly optimized settings after, by the "EPU".
Don't get me wrong: My intention is not to offend you.
However, if marketing speak and/or random postings on the internet sound too good to be true (e.g. "drop system power up to 30% overall"), then they are most likely just that.

You are probably right in one regard:
From what I know, Gigabyte is anything but known for its low power designs.
But hey, if it's better then what you currently have, then it is still better. Regardless if there might be even more efficient options. It probably also depends if you are willing to pay 50,-EUR for a slight improvement, or rather 300 for an even better one. Nearly always depends on a lot of personal factors.

 

[alexskysilk]

From what I know, Gigabyte is anything but known for its low power designs.

virtually none of the Taiwanese OEM/ODMs do any "design work." They use reference designs provided by the chipset vendor. the less reputable ones may use substandard subassembly components which would be the only reason for differences in power/heat/longevity. Consequently, power draw will be most consequently a function of the chipset and power supply efficiency.

 

[zenowl77]

If you cannot deliver any real technical background, as to why that ASUS feature would be so groundbreaking, concerning power-efficiency-optimization, then I'm afraid you just repeated Asus marketing babble.

Sorry if that sounds harsh, but if you use phrases like "usually a 7-11w drop on both idle and load" without any reference values, then this is basically nonsensical. So if I had 7-11W idle consumption, the Asus feature would put me to 0W?
Furthermore "showcasing a load watt drop from 150w to 83w" (if those values should indeed be real) would hint at rather horrible settings before and maybe slightly optimized settings after, by the "EPU".
Don't get me wrong: My intention is not to offend you.
However, if marketing speak and/or random postings on the internet sound too good to be true (e.g. "drop system power up to 30% overall"), then they are most likely just that.

You are probably right in one regard:
From what I know, Gigabyte is anything but known for its low power designs.
But hey, if it's better then what you currently have, then it is still better. Regardless if there might be even more efficient options. It probably also depends if you are willing to pay 50,-EUR for a slight improvement, or rather 300 for an even better one. Nearly always depends on a lot of personal factors.

it simply lowers voltages and favors lower clock speeds in a better more efficient way as far as i know / have seen, like the 83w instead of 150, it just likely didnt instantly up the cpu to full speed when it wasnt needed, etc, and it has some pretty good results, it is something you could likely do yourself with enough tweaking on most boards, its not exactly some revolutionary change in the way the components are powered or anything, but it is nice to have the bios itself designed with a feature to do it automatically at a firmware level which is better and more reliable overall, it does produce good results with minimal effort.

yeah personally i do not exactly like gigabyte, i have had every single one i have owned give out over time and end up with the exact same issue where the bios just stops booting and you are left with a blinking cursor in the corner, so i wouldn't consider them solid or reliable long term. but can absolutely be an improvement over cheaper boards.

 

[celemine1gig]

Undervolting and server-type usage is not really what most people like to mix. For a gaming machine, where you might lose all your save games in the worst case, it probably is OK (at least for most people). However, when it comes to valuable personal data, I personally would think more than twice about it.

And especially so, if I didn't know exactly what the "optimization" did and I knew I would have to trust Asus to do it right.
And you can input nearly any brand name in the same spot, where I named "Asus" above.

As alexskysilk hinted at already:
There are sadly very few real manufacturers left, who seem to know what they are doing, to an extent, that is respectable and trustworthy.
I see more and more people who are just lost by the sheer complexity of todays' systems/architectures. And I have seen that with simple end-customers, but also with people who should know.
So, if you don't work in the industry, you have no real chance of realizing, if a manufacturer knows what he is doing or not.

 

[zenowl77]

excessive undervaluing is definitely bad and can lead to data loss and errors in calculations, instability, etc, but with extremely slight undervolts it is usually safe especially with newer hardware if you also slightly decrease the clock speed and with hardware such as gpus where its common to slightly overvolt stock voltages. which i would not do if you have turbo boost or overclocks present. you definitely want to be very careful and run tests for stability and all its a complicated process which is also why the EPU is nice i have never had issues with it, (not to say they don't exist.) and the clock stepping it does is similar to speedstep just seems better.

i would also absolutely disable turbo boost if you seek power saving, reliability/stability, lower heat, etc, that alone by itself with no tweaks can make a huge difference. (personally i always disable it on everything)

great examples below of stock TDP (PL1) VS Turbo (PL2) watt usage on a few different intel CPUs.
https://www.hardwaretimes.com/intel-11th-gen-rocket-lake-s-cpu-power-consumption-explained/


there definitely has also been a decline in recent years of quality in a lot of manufacturers including asus, they use to be the best in my opinion for any consumer high end hardware but they have also had a decline, definitely debatable on which companies are still good.

 

[bluepr0]

The Minisforum ms-01 looks quite cool, 2x SPF+ ports!

 

[fabitaly67]

The Minisforum ms-01 looks quite cool, 2x SPF+ ports!

After long evaluations of powerful and low power miniserver for homelab I just bought/chosed the Minisforum MS-01 Barebone (with I7-12700 Intel CPU) and ordered apart 64Gb DDR5 RAM + 3.84TB U.2 Nvme Drive. Total cost about 1200€ for 30Watts power consumption expected.
For your information and according my investigation and searchs this is actually the most powerful (with low power),expandable and small PC you can find in the market with 2x 10gbe SFP ports and a PCIe slot.
I will use as second node of my PROXMOX Cluster. Main node is also with very powerful and low power hardware (70 to 90W) based on Intel I9-14900T CPU.
As third node (for Cluster votes compliant or a Qdevice) I chosed an Lenovo m720q with I7-8700T 16Gb RAM 12 Watts consumption
Total consumption foreseen of this 3 nodes Cluster (30+95+12) 137 Watts