The truth about CPU power consumption

Patriot

Moderator
Apr 18, 2011
1,304
689
113
The Xeon 'L' parts (as well as the Core CPU 'S'/'T' parts) only have a lower TDP because they've had their top-end limited. At idle they're really no different than the mainstream CPUs. And at load they're less effective (i.e., slower). Also, they're often more expensive.

Intel's Low-TDP variants are intended for applications where thermal performance is limited, such as small/tight cases with little airflow. In a regular case they're wasted potential.
Hold that thought for a moment, because I really do want to get to the bottom of this once and for all. When I compare e3-1220L v3 (ARK | Intel Xeon Processor E3-1220L v3 (4M Cache, 1.10 GHz) ) with e3-1220 V3 (ARK | Intel Xeon Processor E3-1220 v3 (8M Cache, 3.10 GHz) they look completely different. The e3-1220Lv3 (13 watts TDP) has half the cores and 1/3 the base frequency of the e3-1220v3 (80 watts TDP). Yet you're saying that at idle they're really no different? At least on the face of it, that just doesn't sound right.
There is truth to the thought that Intels chips all have very good idle and they are not far apart when idling. However, the L chips are also heavily binned and have lower vcore by default than non-L chips. Yes you could probably undervolt and underclock your i5 and get to a similar place...

Frankly 1/6 *80w gets you 13w ... so these V3 chips scale pretty darn well.

As I have a 1.6ghz 1220v3 as well as quite a few other v3 chips ... I could do some testing in the coming weeks.
Suggestions as to accurate power meters?
 

Patriot

Moderator
Apr 18, 2011
1,304
689
113
Hold that thought for a moment, because I really do want to get to the bottom of this once and for all. When I compare e3-1220L v3 (ARK | Intel Xeon Processor E3-1220L v3 (4M Cache, 1.10 GHz) ) with e3-1220 V3 (ARK | Intel Xeon Processor E3-1220 v3 (8M Cache, 3.10 GHz) they look completely different. The e3-1220Lv3 (13 watts TDP) has half the cores and 1/3 the base frequency of the e3-1220v3 (80 watts TDP). Yet you're saying that at idle they're really no different? At least on the face of it, that just doesn't sound right.

@MacLemon: v2 and v3 are still "launch phase." v1 is EOL, though I'm not sure why that alone would matter.
Huh, didn't realize the E3-1220Lv3 specifically was so heavily gimped. I'm surprised they didn't give it a different number to more easily distinguish it from the regular E3-1220v3 (yay Intel marketing!). That particular one looks like an exception to what I previously stated.

I was thinking more along the lines of the E3-1230(L)v3 or the E3-1275(L)v3, where the frequencies are closer and have the same number of cores.
I was just as surprised by them naming the C2550/2750 products as "Atom" as the performance is an order of magnitude above any previous Atom product. On that note, they're probably better than the 1230L...
No, it's not even close to the compute performance of the 1230L. The C2750 (TDP 20 watts) has a passmark of 3929, and the E3-1230L (TDP 25) has a passmark of 7244. Also, the 1230L has VT-D, and the Avoton doesn't.
 

Patriot

Moderator
Apr 18, 2011
1,304
689
113
I was under the impression the Haswell power states were only in standby/sleep.
I don't know... but frankly I am not in the mood for digging through intel white papers right now...
I would be surprised if it was not more complicated.

It may be that there are only a few P states... but how they interact with the C-states and T-states ... creates the complicated mess that it Intel's dynamic frequency and power usage.

Does anyone have powermeter suggestions for doing some in-depth analysis?
I don't think a killawatt is going to be as accurate as we desire for low end systems... I also may have damaged mine trying to measure a 1480w load once...

@HellDiverUK FSP 1u unit model number?

I think it is time to have some fun with power draw numbers.
 
Last edited:

NeverDie

Active Member
Jan 28, 2015
299
26
28
USA
I don't know... but frankly I am not in the mood for digging through intel white papers right now...
I would be surprised if it was not more complicated.

It may be that there are only a few P states... but how they interact with the C-states and T-states ... creates the complicated mess that it Intel's dynamic frequency and power usage.

Does anyone have powermeter suggestions for doing some in-depth analysis?
I don't think a killawatt is going to be as accurate as we desire for low end systems... I also may have damaged mine trying to measure a 1480w load once...

@HellDiverUK FSP 1u unit model number?

I think it is time to have some fun with power draw numbers.
@Patriot: How's it going with the powerdraw numbers?
 

mrkrad

Well-Known Member
Oct 13, 2012
1,244
52
48
So how stable are those power savings modes versus PCI-E versus crashing? Back in the X5600/5500 series we had to peg power save to DISABLE/MAX POWER USAGE to achieve stability and maximum pci-express performance with esxi?

Has this front improved much?
 

Patriot

Moderator
Apr 18, 2011
1,304
689
113
@Patriot: How's it going with the powerdraw numbers?
Still need to get a lower power psu to get good load lines... seasonic makes like a 300w gold... thats the best I have found so far.
Also have not dropped the $250+ to get a power analyzer... The killawatts are not accurate under 60w.

But I do have a e3-1220l v3 1.6ghz 16w chip I want to test and a E5-2630L v3 ...
 

NeverDie

Active Member
Jan 28, 2015
299
26
28
USA
Still need to get a lower power psu to get good load lines... seasonic makes like a 300w gold... thats the best I have found so far.
Also have not dropped the $250+ to get a power analyzer... The killawatts are not accurate under 60w.

But I do have a e3-1220l v3 1.6ghz 16w chip I want to test and a E5-2630L v3 ...
Wouldn't measuring voltage and current with a multimeter get you the answer?
 

Patriot

Moderator
Apr 18, 2011
1,304
689
113
2 values continually fluctuating under load that have to be calculated? no thanks. I need a RMS power meter that sits between wall and psu.
 

NeverDie

Active Member
Jan 28, 2015
299
26
28
USA
I notice some PSU's (e.g. the Corsair RM450 and others) provide power data and even graph it:

Not sure about accuracy, but if you view it as a combo it might be thrifty.

Brand new an RM450 is $90.
 
Last edited:

NeverDie

Active Member
Jan 28, 2015
299
26
28
USA
Corsair link has some interesting issues. hmmm I will look into it.
I'll be interested in what you find out. I ordered a refurbished RM450 that should arrive next week sometime. I didn't buy it for the Corsair link, per se, but I noticed it's part of the package.
 

NeverDie

Active Member
Jan 28, 2015
299
26
28
USA
Link support.... you probably still have to buy the $30-60 link module.

Also... Not terribly accurate given this review.
JonnyGURU - Corsair HX750i 750W
That's a pity. Great review though. I wonder if Corsair's "flagship" AX760i would be any better?
Corsair Power Supply Units — the state of the art PC Power Supplies for reliable delivery of clean power under imperfect conditions
The table lists the AXi series as having "advanced" Corsair Link support, versus merely "Intermediate" support for the HXi series, which is what JohnnyGURU reviewed. At the bottom the chart claims the AXi series is designed for "system monitoring and control," but isn't making the same claim as the design target for the HXi series.

I realize those are rather shallow reasons for optimism, but who knows? Maybe. I would think the DSP that's in their AXi (but apparently not in their HXi) would need accurate measurements if it's to be of any value.
 
Last edited:

T_Minus

Build. Break. Fix. Repeat
Feb 15, 2015
6,974
1,561
113
CA
2 values continually fluctuating under load that have to be calculated? no thanks. I need a RMS power meter that sits between wall and psu.
Which unit were you looking at? I want to get something more accurate than the Kill-A-Watt too.
 

Dajinn

Active Member
Jun 2, 2015
512
78
28
29
So are Xeon L5640 CPUs a waste for home use? Is that a take away I can have from this discussion? lol
 

Patrick

Administrator
Staff member
Dec 21, 2010
11,883
4,845
113
So are Xeon L5640 CPUs a waste for home use? Is that a take away I can have from this discussion? lol
Westmere-EP low power means a lot more than it does now in terms of idle power consumption. I do like the L5640's for home use still.

Intel's power management has gotten ever increasingly more complex hence the move to put more of those components on die. Newer chips do have very similar idles.
 

vrod

Active Member
Jan 18, 2015
233
33
28
27
So it's a while since anyone posted anything in this topic, but have people made improvements? Are there anyone else out there than me with "custom case" designs?

I'm currently netbooting all my servers through PXE and anything else than my central storage box are running in my own custom "shelf" cases. Basically a 2U shelf with a plastic spacer between the shelf and the board.

Board is attached with zip ties and beside is the either 2U EPS Psu or TFX psu. The plastic plates have some extra spacing on the side so the board can't get in any metallic contact. Then just a couple of low power, self regulating arctic cooling fans who keeps a decent airflow depending on temp.

Works really well and the depth of the shelves enables me to put one in the front of the rack and one in the rear (front one just rund airflow in reverse to match rear). IO panel is facing outwards so easy enough to attach vga, network or something else.

With a S2600cp4 board coupled with some 2670's, I can get down to 40w idle. It's really beneficial for me as well because I can disable power hungry stuff like the ahci controllers and such.

The density also makes it a good solution for cost savings and I'm currently trying to draft a 1U solution which will enable me to have 4 2670's in 1U. :D

Best of all is that it's a lot cheaper than normal rack cases!