All new Registrations are manually reviewed and approved, so a short delay after registration may occur before your account becomes active.
Regarding amps required for a server
Hi,
I am planning to do colocation for a server, but I am trying to figure out the hardware I need exactly, the budget and my colocation requirements.
In all colocation offers I see, they are billed by amps, 0.5 amps,1amp, 2 amps, etc.
Currently the way I estimate my requirements is by taking the CPU typical TDP and divide by voltage (220V for Europe).
So for example for an i9-9900 95W/220V ->0.43 So 0.5amps is enough
for an i9-10850k 125W/220V ->0.56 So 0.5amps is not enough, I need 1 amps
Is the way I do it correct? is there anything I am not taking into account about consumption of other components (the server itself and the drives)?
Thanks in advance for your help
Comments
ram, drives, fans, etc all use power
you also need to typically buy more on top since you can only use 80%
Yeah, I was thinking about this too... So any idea how much more should I add to my estimates for these?
So basically, to allow some margin for drives, fan, ram the only servers you can fit on 0.5 amps are 6 cores maximum, with TDP no more than 65w for the CPU to leave margin for the others?
Also HDD will use more than an SSD; keep that in mind.
@afn I use this to get a general idea of wattage used.
https://outervision.com/power-supply-calculator
These days TDP has little to do with real power consumption. Just check the latest Intel cpu: they can take twice more power than their TDP design value.
There is no realiable way to estimate or calculate the power you need. The only proper way is to measure it. You can buy such a gear for a few bucks (check amazon for "power meter"), it is used between wall socket and power cord. Then you put your server under load (i.e. kernel kompile on all cores) and you get max power you need...
CPU watts is what uses the most power. Unless you have like 10+ non SSDs
Every 120w is about 1AMP
Do you have a target location (country) for your box?
@DataIdeas-Josh Thanks! great link
@Jarry, sadly this is not really possible in my case since I am actually trying to build a server (or maybe 2) such that it fits in 0.5amp and another one that has to fit in 1 amp. So basically I need to have a general idea of the hardware before ordering, can't wait till I get the hardware to measure...
So far the general idea I have for my setup 2 DDR3/DDR4 ram, 2 SATA drives, 1 SSD, fan,
with:
I am guessing servers like the hetzner ex line
https://www.hetzner.com/dedicated-rootserver/matrix-ex
Should fit fine in 0.5 amps?
@oplink,
Thanks for the info, also for your question, yes NL or maybe FR
For a nominal 120V (US) supply, yes, whereas it's normally 110V (EU) and in developed countries (:-p ) 240V. Notwithstanding supply voltage drops/spikes.
Also prefer SSDs over spindles where possible and 2.5" over 3.5" and be sure to get and look at the power consumption numbers of your disks; there can be quite significant differences. And forget about intel DT processors. Go for server processors with 45 - (max) 65W and it might work out.
Server alone can't draw more than the power supply rating. Though some are deliberately well over-specified.
And then there is Europe with its 230V.
It doesn't sound like you actually have a special server or requirements that require Colo. Why even colocate instead of renting a dedicated server and limit your risk?
Maybe because they want to colo.
Some people don't like the idea of having their stuff on other peoples hardware....
@jsg thanks a lot for the hints.
Assuming the server costs around 800-1000€ Saving 30€ per month , means 2-3 years to get my money back, after it, you get a very cheap monthly price for a couple extra years remaining in the server's life. + The money spent on hardware is not lost money, you can always reuse/resell if you need an upgrade. Even if I assume my drives will die during the first 3-5 years, it is still almost the same price wise, but at least you get something slightly better for the money (+ you get the fun of it, the experience and the stress that comes with it xD, the full pack)
Also to me, faster cpu = less computation time => time saved=> implicitly more money.
Not directly, but in some sense there are some other hidden earnings if I have a better cpu. (and no, not using it for crypto mining)
@afn if interested I offer colo here in Texas.
Thanks for the offer, I appreciate it I have been actually keeping an eye on your website since a while in case I need services in the US. But sadly I'm interested in Europe only (mainly NL, or maybe FR)
The Supermicro 6039P-TXRT has 10A power supply.
We just ask for one 20A circuit for every two servers.
No need to measure anything or change any configuration.
The power supply is not supposed to use more than its labeled value.
The server actually has two power supplies, but works on just one.
So it's calculated as 10A instead of 20A.
Apparently the vendor puts in two power supplies, so that I can move the server onto another power circuit if the current one needs repairs.
What a waste!
Just checkpoint the experiment and turn off the server when you need to repair power circuits.
Uhm, that box is not available anymore ("status: discontinued") and highly likely uses (or wastes) way too much power, is clearly far beyond what @OP wants, and has > 1 kW power supplies, but supports no graphics (no built in GPU).
Did you even read the OP?
Side note: dual PSU is not a waste but meant to increase reliability/availability by keeping the server running in case one PSU or one power line fails. In fact I strongly suggest to insist on dual PSUs when renting a dedi (and they are standard except for the lowest end providers).
To build on top of dual PSUs. It can be used for A/B power. In case you have a total power loss say from the grid. You have secondary power from battery/generator.
That's what I said (minus the 'can'. With me it is 'should').
Nope, that's not how it's supposed to work. Both power feeds to the rack should be UPS'd. The point of a dual PSU is to continue running when either a power feed fails (or is in maint.) or a PSU fails.
No, I agree that both feeds should be UPS'd but some places don't. 🤷
And I was agreeing with what you said.
that is wrong. The OPP will trigger normally at ~130% of the rated power on the 12v rail, so depending on the efficiency of the PSU they will draw 135-150% of which the PSU is rated from the wall before they turn off.
Y'all people just asking for something to burn if pushing PSUs that far though.
With all that power managment of modern CPUs and GPUs the usage is so fucking spicky that you can trip OCP or OPP even if your build is designed to only use 80% of your power budget. But beside that even if you push the PSU that hard it doesn't matter they will turn off before they would be overloaded. Of course some of the 10$ China ones won't but you can't get them legally anywhere anyways, so i hope nobody uses such crap.
Yes, I did.
I meant that, if OP wants 1A server, just take a PSU that is 1A.
This model has no built-in GPU, but it has 11 PCIe slots (only four are x16 lanes).
If you put 4x Tesla K80 in there, total power consumption is more than 1.5 kW, and you'll need both PSU.
We added 3x ConnectX-5.
They draw up to 27W each, so I guess the PSU only outputs 500W or so.
I wouldn't mop over this model being discontinued.
The next one, I'll ask for ConnectX-6 200Gbps and EPYC processor.
I could plug in two different power circuits, but then I would need twice as many power circuits.
What a waste!
It's simpler to just checkpoint the experiment and turn off the server.
The only time(s) I used the second PSU is when I need to untangle the cords in order to move fibers/ other machines.
In that case, I could plug in a second cord and unplug the first one.
The PSU is labeled "980W" and "input 10A", so I suppose the inefficiency is already accounted for.
In Power state 2. In Turbo (P1) a 9900K will use up to 170W for 25 seconds (across all cores), and up to 190W for 10 seconds on a single core.
Intel TDP is about HALF of real usage in 9th/10th/11th Gen CPUs, as this was the only way to keep a marketing number somewhat near the cheaper, better, AMD CPUs.
A i7-10700 (8c/16t) has a bench of 17400 at 65W TDP (P1 95W, P1+ 110W), 14nm Intel
A Ryzen 5 5600X (6c/12t) has a bench of 22100 (30% more at 2 cores less) at 65W TDP (P1 65W, Turbo 90W), 7nm TSMC
Intel also has no ECC support and no PCIe 4.0.
The amp sticker has to account for loss and overload. The Amp stated on the sticker are a finite limit, and there is a legal requirement for them to be correct in EU.
Also each PSU, at least HP/Dell/IBM, comes with a 10A+ glass fuse that limits it anyway.
This.
The PSU output/efficiency and server internals are irrelevant, in my statement. If the PSU draws more than its' rated INPUT amperage, then it is faulty/out-of-spec/f'kin dangerous. Let's hope it blows a fuse!
(OVH fire, cough)
Notably at 210-240V (EU, Asia, anything aside US and Japan anyway) you should buy a 230V ONLY PSU (NOT A 80-280V universal) - Efficiency in Switch Mode Power Supplies (SMPS) is higher the more limited your input voltage is.
Modern server PSUs run at 80-270V (80V is for Japans 100V system, 270V is for Eastern Europe where 250V+ is not rare) - Efficiency around 90-93% (VERY high, Gold+), Fixed voltage goes up to 95-98% (Platinum Plus).
Also a universal 1000W PSU will come with a 10A fuse (120V=1200W, 220V=2200W) which is too much for EU, while the EU version will come with a 5A one (and a MOV) instead.
Lastly, if you ever need it - All of this PSUs run perfectly fine on 90-230V DC - Chain 6 car batteries and your server will boot just fine. Polarity irrelevant (but should change it at times, to switch the full bridge rectifier to the other 2 diodes, DC will load only 2 always).
LOL. Just don't use a cheap Chinese car charger on them! (I've seen over 15V output from those time-bombs.)