Processor downgrade not resulting in large power savings?

[altano]

Hey all,

Northern CA electricity prices are killing me so I downgraded my home server. I went from:

* 2 x E5-2670 v1 (115w TDP)
* 8 x PC3L-10600R ECC DIMMs

To:

* 1 x E5-2650L v1 (70 watt TDP)
* 4 x PC3L-10600R ECC DIMMs

And the idle power draw was only reduced from 180 watts to 130. How could the reduction be so little, even at idle?

Other parts using up energy:
* SuperMicro X9DRD-IF-O
* 3 x Dell PERC H310
* 11 fans, mostly 120mm (don't ask, it made sense at one point)
* 11 3.5" SATA HDs, mostly WD Reds
* 2 Samsung SSDs

I still have to collect more data over time but I'm really disappointed so far. Did I forget to flip the "stop using so much power" switch somewhere? =\

 

[pgh5278]

Hello altano , was there a specific reason You thought you would save more than this?
50 / 180 = 38% reduction...not bad...
the 115W and 70W is the maximum power draw of the 2 CPU, they do not draw this when idling along running a few disks etc,.
Depending on Fans , 11 of them can draw plenty of W..

To make further cuts, look at increasing HD size, decreasing qty of disks, HBA and Fans.
Depends on the cost benefits.
This forum has some good articles and tables on system power / CPU power use etc..

 

[altano]

I guess in retrospect this was stupid, but I was hoping the idle power consumption would scale as a ratio of TDP. If I maxed out the CPUs all the time and they consumed 2x115 watts vs. 1x70 watts, the power reduction of the CPUs (not the whole system) would be 70%. But I guess power consumption doesn't work like that, not even close? Looking for confirmation here.

I bought a 4TB drive to replace 3 ancient 2TB drives in the system so hopefully that will help too. Shrug. Maybe I'll move to a Xeon-D platform to reduce it much more (as discussed in this thread: Reducing power consumption of my homelab)

 

[pyro_]

min power on the E5 chips is going to be similar most of the time regardless of tpd numbers, where the difference will be is when there is a load on the CPU this is where your energy saving will potentially be, are the fans PNW fans or running at a strait speed that could make a difference. Also try to cut down on the number of HD that are in the machine

 

[i386]

I would recommend this thread: The truth about CPU power consumption
There is more than just cpu tdp that affects (system) power consumption.

 

[i386]

I would recommend this thread: The truth about CPU power consumption
There is more than just cpu tdp that affects (system) power consumption.

 

[Evan]

30-40watts in the 3 x SAS cards.
Yeah replacing disks will make a difference.

Xeon-D idle ~25 watts
1 x sas hba ~10-15 watts
4 x 8tb disks and a couple SSD idle ~25 watts

A new system should idle around ~60 watts at a guess.
So see if it makes sense to update based on that ?

This is why I love to try to make super low power servers for home, the energy costs add up, sure so does purchase price but it's nice to have new cool running low power gear.

 

[TLN]

30-40watts in the 3 x SAS cards.
Yeah replacing disks will make a difference.

Xeon-D idle ~25 watts
1 x sas hba ~10-15 watts
4 x 8tb disks and a couple SSD idle ~25 watts

A new system should idle around ~60 watts at a guess.
So see if it makes sense to update based on that ?

This is why I love to try to make super low power servers for home, the energy costs add up, sure so does purchase price but it's nice to have new cool running low power gear.

Yeah, but some CPU @ 60W will be staying idle, say at 30% of CPU usage and that's basically the same. Except that you can use all that horsepower when needed.

 

[rubylaser]

Have you tried spinning down disks? This makes a huge difference on my fileserver. Also, can you remove, or at least unplug, some of the fans to see how much they are drawing?

 

[Evan]

Yeah, but some CPU @ 60W will be staying idle, say at 30% of CPU usage and that's basically the same. Except that you can use all that horsepower when needed.

2 x E5-2670 your right but 1 x e5-2650 I would take the Xeon-D
The Xeon-D will always have the power advantage but a an upfront cost that I guess won't recover the cost, hence why I gave an indication of what may be possible.

 

[Patrick]

This is a fairly common conception.

TDP is thermal design power which is not a power consumption spec.

Processors in a given generation tend to idle close to one another, especially in newer generations. While in older generations (e.g. the Xeon 5400 series) this was not necessarily the case, today Intel CPUs are very good at power management. Even V4 chips are a fairly big upgrade over V1 / V2 in terms of how fast they can throttle up/ down cores.

Another key factor is a base system load. This can be drives, fans, PSUs, and add-on cards, but there are also various onboard components. For example Supermicro SYS-5029A-2TN4 Review: A small Intel Atom C3338 NAS

There is an example of a 9W TDP CPU and the system, at idle, had a power draw of around 2.5x that amount.

You can get some gains by moving to Xeon D (especially if you have an older 10Gbase-T NIC), or going from dual V1's to single V4.

The other side to it is that maximum power consumption should be way lower for you with that CPU over your old dual processor combo.

 

[K D]

30-40watts in the 3 x SAS cards.
Yeah replacing disks will make a difference.

Xeon-D idle ~25 watts
1 x sas hba ~10-15 watts
4 x 8tb disks and a couple SSD idle ~25 watts

A new system should idle around ~60 watts at a guess.
So see if it makes sense to update based on that ?

This is why I love to try to make super low power servers for home, the energy costs add up, sure so does purchase price but it's nice to have new cool running low power gear.

This about sums it up.

Xeon D 1528, 826 chassis with 920SQ psu and TQ backplane, 2x Seagate 3tb 7200rpm drives, 2x DC 3500 400 gb, 2x Noctua Redux 80mm fans - I have 2 nodes identically configured that consumers ~ 54 watt idle

Another xeon e3 1275 system on an x11ssh-ctf board with everything else the same also consumes around the same power idle.

I've been able to hit less than that only with no drives/fans/hbas etc connected. I believe that ~50 -60 watt is a good target.

 

[PigLover]

Going to fewer/larger HDDs will help some, but not as much as you think.

A 3TB WD Red will consume ~4w "idle/spinning" and ~5-6w during active transfers, and it will drop to ~2w if you force spindown. Not a very big range at all. With 11 drives you are probably seeing ~40-50w from HDDs. If you go do 5x 8tb drives you can get this down to ~20w (which is about the same as you'd get with the current drives and spindown - but remember spindown/spinup will likely shorten drive life).

 

[altano]

Have you tried spinning down disks? This makes a huge difference on my fileserver.

I wish I could but Storage Spaces doesn't support it :/

 

[altano]

Does a platform like the X10SDV-7TP4F save a lot of power with the integrated Broadcom controller or would it be equivalent to adding the HBAs as pcie cards?

 

[OBasel]

I'm not totally sure but I think it's relatively minor. IMO it's just a PCIe card onboard, maybe some shared power stuff. But HBA's are like 10-15w right? If you save 3w or 20-30% is that really much?

 

[altano]

I'm not totally sure but I think it's relatively minor. IMO it's just a PCIe card onboard, maybe some shared power stuff. But HBA's are like 10-15w right? If you save 3w or 20-30% is that really much?

Well I guess I was wondering about the approach in general. If I get the X10SDV-7TP4F and I don't use some of the SFP ports, for example, will I be wasting a lot of power, or is the board good about being able to turn this stuff off in the bios?

 

[Rand__]

Not sure about that specific board but on my 1528 i was able to disable the 10gbe secondary nics by jumper leaving only the primary 1gbe nics.
O/c on x10DRi-T4 with 4x 10GBe Nics I was only able to disable the 2 secondary leaving 2 primary 10gb nics enabled

 

[Evan]

Does a platform like the X10SDV-7TP4F save a lot of power with the integrated Broadcom controller or would it be equivalent to adding the HBAs as pcie cards?

Question I was also interested in but I suspect we are taking about 1 or 2 watts different only.
The other issue is that's a LSI 2116 and if you want to use SSD your better off with a 3008 or similar 12gb HBA.

 

[Evan]

According to @Patrick testing here at STH then the 2116 is idleing 8 watts higher than non 2116 versions, so probably a little saving. (And that was a 16 core D)

Maybe he has some insight to the performance of the 2116 with ssd, certainly it's a cheaper way to get a SAS adapter and low power.