The truth about CPU power consumption
- Thread starter sub.mesa
- Start date
Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.
I have learned something here,
Designing CPUs that perform tasks efficiently without overheating is a major consideration of nearly all CPU manufacturers to date.
Designing CPUs that perform tasks efficiently without overheating is a major consideration of nearly all CPU manufacturers to date.
Too many variables. you can calculate, but the only sure thing is measuring at the wall output with a kill a watt meter (assuming those things are accurate). Its pretty difficult to get anything to idle with less than 50-60W total in a WS or server i think. plan at least 100w.
No, it's actually really pretty easy to get a server to idle less than 60W if you use anything made in the last 2 or 3 years.
Dell T20 with Xeon and 4 HDD - idle 14W, peak consumption 80W while all drives spinning up, idle with drives spinning 40W.
Fujitsu TX1310 with Xeon and 6 HDD - idle 12W, peak consumption 85W spinning up, idle with drives spinning 35W.
Heck, even the dual Opteron Dell R515 with 10 SAS drives and 128GB RAM I have in work barely uses more than 125W with all the drives running.
How do you like the T20 in general? They are really inexpensive here in DK, so I've ordered one just yesterday. Would appreciate any experience you might have.
The fact it only has 4x SATA ports on the board sort of sucks. Cooling of fast drives isn't great, as the whole machine is cooled by the rear 92mm fan. Slower drives like the WD Red, or Helium drives are fine, but something like a WD RE or Toshiba enterprise drive will run really hot.
Otherwise it's fine. It runs REALLY quiet, almost silent. Quieter than my Synology DS916+.
I also have a Fujitsu TX1310-M1, which is a similar machine, but it has 6 SATA ports, nicer HDD bays, and 2x120mm fans so cooling is much better. It's more expensive, of course. A little noisier than the T20 as the fans make a bit of motor noise. I use that for CCTV running Windows 10 and BlueIris, and it runs great.
Seems like the only true way to know idle consumption is by hooking up a kill a watt meter?
I would say that CPU power consumption is actually just a minority of the total for most people unless they're running 24/7 computing task.
I would say that CPU power consumption is actually just a minority of the total for most people unless they're running 24/7 computing task.
With testing and at scale, a digital "kill a watt" is the way to go for power optimization.
By knowing the consumption in real time, its possible to tweak the knobs to optimize compute/watt or other relevant metric.
Here's more info of this in action: Data Centers | Google Sustainability
and a related example Google DeepMind-style datacenter optimization AI model (on the cheap)
By knowing the consumption in real time, its possible to tweak the knobs to optimize compute/watt or other relevant metric.
Here's more info of this in action: Data Centers | Google Sustainability
and a related example Google DeepMind-style datacenter optimization AI model (on the cheap)
Last edited:
i dont agree with you...even when the thread is 15yr old, still you can read and ask things
Hello, currently I'm running an E3-1220 v2 in an intel motherboard. Without any disks and in idle the platform pulls about 33watt from the wall in a supermicro 2U chassis and with desktop 80mm fans.
I really want to change to a Supemicro board because this intel one gives me too much issues with virtualization on Proxmox. Also changing to a CPU with hyper threading might be interesting.
Now I can't seem to find any good deals on x9 boards for E3 cpu's in my area. But there are several x9 DP boards for the E5-2600v2 platform.
Can anyone tell me if the power usage at idle is much different between these? If you would use a single CPU like a low end E5-2650V2 or E5-2620V2, would there be much difference?
Thx for any feedback!
K.
I really want to change to a Supemicro board because this intel one gives me too much issues with virtualization on Proxmox. Also changing to a CPU with hyper threading might be interesting.
Now I can't seem to find any good deals on x9 boards for E3 cpu's in my area. But there are several x9 DP boards for the E5-2600v2 platform.
Can anyone tell me if the power usage at idle is much different between these? If you would use a single CPU like a low end E5-2650V2 or E5-2620V2, would there be much difference?
Thx for any feedback!
K.
Yes of course, did I miss something?
I'm specifically wondering if the X9 LGA 1155 boards use more then the X9 2011 boards in terms of chipsets etc. And if an E3-1220v2 cpu uses more power at idle then a E5-2650V2 or E5-2620V2.
They have different TDP's but as mentioned in the first post that doesn't mean anything in terms of power consumption.
To end the back and forth: an E3 setup of that generation (Ivy Bridge) will draw LESS power at idle than an E5 setup regardless of which E3 and/or E5 you use. Besides the CPUs themselves, the C6xx chipset in the E5 boards draws more power than C2xx chipsets in the E3 ones. It won't be a lot more, but it will be more.
Depending on how you are virtualizing, your problem may be with the E3 platform itself and not that its in an Intel motherboard. PCI passthrough, for example, was difficult on that generation of E3. So, regardless of power consumption, the E5 may be a better fit for your needs.
As @spyrule notes, this is a sticky thread. So, for any other/future questions you'll probably want to open a new thread.
Cheers!
I wondered if anyone validated the accuracy of the power consumption listed in CORE Temp? In IPMI? and with a UPS? For the socket 2011 v2 processor I have in CORE Temp dynamically shows power consumption. For the Socket 1366 it shows the TDP (this has no idle mode) As has been stated previously in this thread CPU watts are not the only consideration. The UPS should be the best measurement of the overall current draw?
- I used APC Power Chute (Windows on separate machine), Core temp and IPMI to measure current draw
- I found my idle Dual E5-2650L v2 system with 32GB memory, one hard drive, 3 Mellanox network cards, USB 3.0 card draws 36 watts(UPS). Power supply on standby showed UPS showed 0 watts
- To load the CPUs I used Vegas Video under windows. When rendering video all 20 threads are 100% loaded this draws 60 watts according to the UPS and 57.1 watts according to Core Temp. IPMI 160watts Peak/96average
- When all drives powered the power went to 69.9 watts (I rebooted with Truenas, so no Core temp) IPMI reports 200watts peak/115 watts average
- When initiating a large copy to a Truenas drive the UPS reports 77.8 watts, IPMI reports 200watts peak/152 watts average. I was surprised how little power the drives added. They are 6TB Hitachi spinners. my 10Gb/s network was not working so transfer was over 1Gb/s network.
Last edited:
This thread is funny, why is it held at the top of the list??? Id consider a low power and powerful home server pc to be sub 10w, and many people do these days. Not sure why this is sticky in 2024.
OFC. Do you not think its a little out dated though? Ive followed servehome for a long time, this post doesn't cut it these days.
chill baby, a sticky post is like an FAQ, keep it to the top so FAQ's are there, right at your finger tips, its a help to people with common questions who CBA to use the search functionality. 
Close enough ?
Have a read of the thread, all 3 pages, tell me how relevant it is 2024. Low power servers these days are sub 10w. Its actually servethhome that made me get rid of my noisy HP DL380gen9 a few years ago and spent so much money buying many teeny tiny micro factor machines, which rock btw.
my current server is an M920q, 8TB storage, i5, 64gb ram, 1gb + 2.5gbe...... 6w at idle running HA and a few LXD's. Need more storage though
Again, i have lots of respect for servethome, watched all the YT vids and its ace.... so no disrespect is meant here.
Close enough ?
Have a read of the thread, all 3 pages, tell me how relevant it is 2024. Low power servers these days are sub 10w. Its actually servethhome that made me get rid of my noisy HP DL380gen9 a few years ago and spent so much money buying many teeny tiny micro factor machines, which rock btw.
my current server is an M920q, 8TB storage, i5, 64gb ram, 1gb + 2.5gbe...... 6w at idle running HA and a few LXD's. Need more storage though
Again, i have lots of respect for servethome, watched all the YT vids and its ace.... so no disrespect is meant here.
Last edited: