Will 1x 130W processor use more electricity than 2x 80W?

[SPCRich]

So the title is a little obvious, but here's the gist. I'm currently running a Supermicro 6027-DTRF dual node chassis as a home kubernetes cluster. I'm always trying to optimize my home equipment to use less power (I pay ~45c/kWh during peak), so I'd like to try and dial this down a bit, if possible. Each node came with 2x 2630V2 processors, for a total of 4 processors, and 8 sticks of DDR3. Specs on the processors:

2630v2: 2.6Ghz (3.1Ghz Turbo) @ 80W TDP

There are some good deals on E5-2643 v2's on Ebay right now, so I'm thinking of picking up 2 of them, and swapping the 4 processors for 2. Specs on the E5-2643 v2:

2643 v2: 3.5Ghz (3.8Ghz Turbo) @ 130W TDP.

95% of the time, my dual 2630's in each node are sitting around 5-10% usage. I should be fine going from 2x hex cores (24 threads) down to 1x hex-core (12 threads), and still keep that usage under 25 or so. The only time it ever really peaks is when I transcode video, then it goes to about 80%, but 3.5Ghz might help that somewhat. So my question is really this: Even though the 2643 V2 is higher TDP, the fact that it will idle most of the time (albiet a little higher idle than dual 2630 v2s) should result in less power consumption than running dual 2630v2's at 80w TDP, correct?

Edit: Yes I'm aware that TDP != power consumption. I'm asking if going to a single cpu that idles slightly higher will consume more electricity than 2x cpus that are idling slightly lower.

CPU usage on one node for last 24 hours:

 

[msg7086]

CPU at the same generation usually have similar power consumption curves. So yea, single E5v2 will use less power than dual E5v2 even considering the usage.

However E5v2 are pretty power efficient by themselves, so you won't see that much difference. My suggestion is, only do so if it truly benefits your usage. Power saving is just a tiny side effects. Those CPUs themselves could cost much more than the power difference.

Another option. I'm not sure how you are using those servers, but if they are only used for transcoding, try to set it up so that some of the servers suspend to memory and listen to wake-on-lan for remote wake up. That'll help a lot on the power consumption because suspended state uses almost no power.

 

[Evan]

Yes, at idle and low load each cpu will be similar, it’s only when they are working a bit harder will you see each cpu starts to consume a lot more different based on number of cores and speed those cores are running at.

So 1 cpu instead on 2 will certainly save power if you don’t need that compute capacity (or memory or pci slots etc)

 

[SPCRich]

Another option. I'm not sure how you are using those servers, but if they are only used for transcoding, try to set it up so that some of the servers suspend to memory and listen to wake-on-lan for remote wake up. That'll help a lot on the power consumption because suspended state uses almost no power.

I'm using both for Kubernetes... Running Plex + about 30 other services on them, so I can't really power one down until WOL, since they're balancing load now, and if i do maintenance on one I can fail pods over to the other one and still keep them running.

Thanks for the confirmation, I think I'm gonna do it. The 2x processors are ~200$, which will pay for themselves in about 1.5-2 years (based on what my power usage went up when i run both nodes, I figure each CPU is drawing around 20-30 W idle based on the fact that with both nodes idling my rack power consumption goes up about 200-ish watts., figure idle of about 20-30W/CPU, and like i said for 1/4 the day I'm paying 45cents/kWh, the rest it's either 12 or 22cents.

If it doesn't decrease significantly I can always hock them back on ebay again.

Thanks!

 

[msg7086]

For ~45c it'll definitely pay off pretty quick. For 10c/kwh it's about $1/watt per year. Yours are about $2-3/watt so 20w = $40-$60 difference.

Oh, also want to remind you that 2643v2 is slower than dual 2630v2.