Power Consumption Thread

Notice: Page may contain affiliate links for which we may earn a small commission through services like Amazon Affiliates or Skimlinks.

Layla

Game Engine Developer
Jun 21, 2016
215
177
43
40
I'm using a per-outlet monitored PDU to measure AC power use (@ 208V) of devices I have in service, and I'm surprised by some of the results so far. I thought I'd start a thread about this (to which I can add results as I go). I'll start with switches, since this is posted in networking, but I also already have surprising findings about IPMI standby power, etc. Note: DC Output Watts, where posted, is reported by the device itself.

Switch Comparison:
DeviceSingle AC Input @ 208VDual AC Input @ 208VDC Output WattsComments
Arista 7060CX154.9WTBD138.8W10/32 100G Ports Active
2/2 10G Ports Active
(30% fan speed)
Celestica DX010124.5W136WTBD00/32 100G Ports Active (Idle)
Arista 7050QX-3297.0WTBDTBD00/32 40G Ports Active (Idle)
Gnodal GS001882.2WTBDTBD10/18 40G Ports Active
Quanta LB4M68.9WTBDTBD0 Ports Active (Idle)
Aruba S2500-48P67.7WN/ATBD0 Ports Active (Idle)
Brocade 7250-4844.1WN/ATBD0 Ports Active (Idle)
Brocade 6450-4834.2WN/ATBD0 Ports Active (Idle)

Things I thought would be a bit more power hungry but are not:
DeviceAC Input @ 208VDC Output WattsComments
Gnodal GS001882.2WTBD10x 40G Ports Active
Brocade 6450-4834.2WTBD0 Ports Active (Idle)

Things I thought would be more power efficient than they are:
DeviceAC @ 208VDC Output WComments
Arista 7060CX154.9W138.8W10x 100G Ports Active (30% fan speed)
Aruba S2500-48P67.7WTBD0 Ports Active (Idle)
Lenovo X3850 X6 IPMI26.5WTBD
Eaton ePDU G35.4WTBDIdle, connected to Ethernet Daisy Chain

Things that are about as power hungry as I expected:
DeviceAC Input @ 208VDC Output WattsComments
SuperMicro X8DTT-H6IBXF 4-Node Chassis202.8WTBD1/4 Nodes Powered Up
SuperMicro X9-DRT-HIBFF 4-Node Chassis137.1WTBD1/4 Nodes Powered Up

Things I haven't measured yet:
Celestica DX010HP DL385 Gen10
Mellanox Voltaire 4036HP DL580 Gen9SuperMicro 4028GR-TRT
Quanta LB4MHP DL580 Gen8SuperMicro X10-DRT-P 4-Node Chassis
Quanta LB6MHP DL580 Gen7
Brocade 7250-48HP DL560 Gen8
Brocade 6450-48HP S6500 Gen8
Mikrotik CSS326HP SL230s Gen8
Mikrotik CRS326HP SL250s Gen8

I'll update this post as I test more of these things. I'm on a quest to optimize power usage in my network, so I'm likely going to be displacing some of the less power efficient gear; but, the measurements are very useful! If I'd just had to guess, I would have ripped the GS0018 out assuming it was not power efficient (because it is older), when actually, it appears to draw much less power at low port counts than underutilizing a much newer Arista 32-port switch. I've tried shutting down the ports on the Arista, but unfortunately, that doesn't really seem to impact power use at all. If anyone has any ideas on how to reduce the power consumption on an Arista switch when fewer ports are in-use, I'd be very interested to know that.
 
Last edited:

Freebsd1976

Active Member
Feb 23, 2018
387
73
28
I'm using a per-outlet monitored PDU to measure AC power use (@ 208V) of devices I have in service, and I'm surprised by some of the results so far. I thought I'd start a thread about this (to which I can add results as I go). I'll start with switches, since this is posted in networking, but I also already have surprising findings about IPMI standby power, etc. Note: DC Output Watts, where posted, is reported by the device itself.

Switch Comparison:
DeviceAC Input @ 208VDC Output WattsComments
Arista 7060CX154.9W138.8W10x 100G Ports Active (30% fan speed)
Arista 7050QX-32TBDTBDTBD
Gnodal GS001882.2WTBD10x 40G Ports Active
Quanta LB4M68.9WTBDIdle, 0 Ports Active
Aruba S2500-48P67.7WTBDIdle, 0 Ports Active
Brocade 7250-4844.1WTBDIdle, 0 Ports Active
Brocade 6450-4834.2WTBDIdle, 0 Ports Active

Things I thought would be a bit more power hungry but are not:
DeviceAC Input @ 208VDC Output WattsComments
Gnodal GS001882.2WTBD10x 40G Ports Active
Brocade 6450-4834.2WTBDIdle, 0 Ports Active

Things I thought would be more power efficient than they are:
DeviceAC @ 208VDC Output WComments
Arista 7060CX154.9W138.8W10x 100G Ports Active (30% fan speed)
Arista 7050QX-32TBDTBDTBD
Aruba S2500-48P67.7WTBDIdle, 0 ports active.
Lenovo X3850 X6 IPMI26.5WTBD

Things that are about as power hungry as I expected:
DeviceAC Input @ 208VDC Output WattsComments
SuperMicro X8DTT-H6IBXF 4-Node Chassis202.8WTBD1/4 Nodes Powered Up
SuperMicro X9-DRT-HIBFF 4-Node Chassis137.1WTBD1/4 Nodes Powered Up

Things I haven't measured yet:
Mellanox Voltaire 4036HP DL580 Gen9SuperMicro 4028GR-TRT
Quanta LB4MHP DL580 Gen8SuperMicro X10-DRT-P 4-Node Chassis
Quanta LB6MHP DL580 Gen7HP DL385 Gen10
Brocade 7250-48HP DL560 Gen8
Brocade 6450-48HP S6500 Gen8
Mikrotik CSS326HP SL230s Gen8
Mikrotik CRS326HP SL250s Gen8

I'll update this post as I test more of these things. I'm on a quest to optimize power usage in my network, so I'm likely going to be displacing some of the less power efficient gear; but, the measurements are very useful! If I'd just had to guess, I would have ripped the GS0018 out assuming it was not power efficient (because it is older), when actually, it appears to draw much less power at low port counts than underutilizing a much newer Arista 32-port switch. I've tried shutting down the ports on the Arista, but unfortunately, that doesn't really seem to impact power use at all. If anyone has any ideas on how to reduce the power consumption on an Arista switch when fewer ports are in-use, I'd be very interested to know that. My other two Arista switches are currently each full with 32-port loads (and not on monitored-per-outlet PDUs), so doing an Apples-to-Apples 10-port utilization power-use comparison will have to wait for now.
change 100g switch to msn2100 or msn2010
 

klui

Well-Known Member
Feb 3, 2019
824
453
63
The only switch I know of that saved power, but I never hooked up to a meter or did I have a PDU that provided that data at that time is Nortel's 55x0. In power save mode the gig interfaces fell back to fast interfaces. I tried to shut down ports on a VDX7640T and got the same results as you on the Arista--no power savings. The Celestica 100G use roughly the same power when idle with no ports connected as the 7060cx. But I don't have a 7050QX to compare against Celestica's D4040, though my guess is they're around the same.
 
  • Like
Reactions: Layla

Layla

Game Engine Developer
Jun 21, 2016
215
177
43
40
change 100g switch to msn2100 or msn2010
Sure, an 18 port switch will likely use less power; but, I'm not going to buy a different switch for the period of time before I'm using all ports :) I was hoping for some kind of software/management change which would tell the switch to save some power. I haven't found any, so far.
 

Layla

Game Engine Developer
Jun 21, 2016
215
177
43
40
The only switch I know of that saved power, but I never hooked up to a meter or did I have a PDU that provided that data at that time is Nortel's 55x0. In power save mode the gig interfaces fell back to fast interfaces. I tried to shut down ports on a VDX7640T and got the same results as you on the Arista--no power savings. The Celestica 100G use roughly the same power when idle with no ports connected as the 7060cx. But I don't have a 7050QX to compare against Celestica's D4040, though my guess is they're around the same.
I just actually confirmed that - straight out of the gate, the Celestica DX010 100G switch draws 124.5W. I'm not sure how long it takes to boot (I literally just unboxed this one, it's new to my network), but I'll give it a few minutes to see if the power usage drops into a lower steady state or not (or stays at this 124.5W steady state it is currently at). I also don't have a D4040, so I won't be able to get any numbers for that one, but I just tested one of my 7050QX-32 (reverse airflow), which idles at about 95W, so a good deal lower than the 100G switches.
 

LodeRunner

Active Member
Apr 27, 2019
540
227
43
My 7050QX-32S-F reports the following:
Code:
Power                        Input  Output  Output
Supply Model       Capacity Current Current  Power Status          Uptime
------ ----------- -------- ------- ------- ------ ------ ---------------
1      PWR-500AC-F     500W   0.41A   2.95A  34.7W Ok     8 days, 0:41:55
2      PWR-500AC-F     500W   0.46A   3.41A  40.9W Ok     8 days, 0:41:56
Total  --             1000W      --      --  75.6W --                  --
Voltage is 120 as it's fed from my UPS. So 97.2 VA on the input side.
 
  • Like
Reactions: Layla

klui

Well-Known Member
Feb 3, 2019
824
453
63
Here are my numbers for network devices I have stacked on my bench and not in storage. All figures had only one PSU, if dual, connected. I'll not replicate @Layla's SKUs.

Things of note:
  • EX4200-48T is much louder than the 48P/PX w/ 930W PSUs. Maybe due to PSU fan. The 48P/PXes are quieter than the ICX6610-48P.
  • The X24S started at 133W and fans slowly ramped down until it reached 121W.
  • SX6012 had fan mod; when fan was at full speed, it drew 37W and was as loud as the ICX6610 at power up.
  • EX4200 10G modules didn't recognize Mellanox 25G DACs and only linked at 1G.
  • Looks like older switches do save power if ports are disabled, but not very much. I didn't do a thorough job for the Juniper switches because I hadn't defined a range as I did for the 2200. At most saved would be a couple of watts.
Device - idle no network ports connectedPower meter usage - single AC 120V
HP 3400cl76W, disabled all ports 73W
Nortel 5510-48T86W, disabled all ports 83W
Nortel 5520-48T-PWR89W, disabled all ports 85W
Celestica D404095W
Arista DCS-7280TR-48C6, dual PSU (in lab at work)152W
Juniper EX2200-48T-4G60W, disabled all ports 57W
Juniper EX4200-48T w/ 10G module, dual 320W PSU121W
Juniper EX4200-48P w/ 10G module, single 930W PSU108W
Juniper EX4200-48PX, single 930W PSU104W
Brocade ICX6610-48P, dual PSU124W
Ruckus ICX7150-C12P, fanless18W
Juniper SRX240PoE51W
Juniper SRX550 w/ 16-port PoE, 2X 10G, single PSU111W
Mellanox SX6012, dual PSU28W
Extreme Brocade VDX7640T, dual PSU195W
Supermicro X24S, dual PSU121W
Extreme X450e-48p78W, disabled all ports 74W
 
Last edited:
  • Like
Reactions: tinfoil3d and Layla

NablaSquaredG

Layer 1 Magician
Aug 17, 2020
1,319
800
113
I wouldn't trust the power supply numbers, real consumption is usually much higher (especially with Supermicro PSUs)
 
  • Like
Reactions: Layla

Rand__

Well-Known Member
Mar 6, 2014
6,626
1,767
113
I wouldn't trust the power supply numbers, real consumption is usually much higher (especially with Supermicro PSUs)
Just checked, my SN2100 says 60W on the power page and my PSM module says its using 74W... so 25% more...
 

tsteine

Active Member
May 15, 2019
167
83
28
How loud are the SN2700's fans?
Well, they're definitely quite loud. It's not something you would want to have sitting in the same room as you for any extended period of time.
Par for the course as far as data center switches go in terms of fan noise
 
Last edited:

Layla

Game Engine Developer
Jun 21, 2016
215
177
43
40
@Layla hi, you've mentioned

Arista 7060CX154.9WTBD138.8W10/32 100G Ports Active
2/2 10G Ports Active
(30% fan speed)

How long have it been running with 10/32 qsfp28 active and just 30% fans? Is it DAC or optics?
I know it's offtopic and I'll remove this later.
Thanks for the great info.
It's mostly CWDM (singlemode) optics, two DAC I think. Have been running it almost 6 months now. Fan speed is set to auto, but I've rarely seen it go above 30% - really only if I increase ambient Temps a lot.
 
  • Wow
Reactions: tinfoil3d

Layla

Game Engine Developer
Jun 21, 2016
215
177
43
40
Love it. So 30% should be something like, very normal noise, not quiet but you can have it sitting a meter away from you right.
I have it a meter away, but in a quiet rack. At 30%, it is very quiet for a datacenter switch, but still not silent. So if you aren't very noise sensitive, I'd say yes. If you are, putting it in an xrackpro2 like I did might help.
 
  • Like
Reactions: tinfoil3d

klui

Well-Known Member
Feb 3, 2019
824
453
63
They have a hollow/tinny whine like the DX010. Not quite as loud as the ICX6110. But you can patch the fan configuration although I haven't done it for SONiC. The problem with using SONiC on the D4040 is the switch was never officially supported. @okrasit shared his deb modules and while it generally worked the LEDs didn't and it was for an old version -- I recall a 2018 snapshot. Newer versions won't load the modules because they were compiled for an older kernel. He indicated he was going to update it but hasn't logged on here for over a year. Maybe he has moved to better platforms; I hope he's doing well.
 
  • Like
Reactions: AveryFreeman