BUILD i3-4130T vs. E3-1231 v3 power usage tested

[Aaron Ryan]

I'm a power efficiency nazi, and I like my servers running cool and low power.

I built a Mini-NAS for my home, and I got a AsRock E3C226D2I w/ 16GB ECC RAM,
Seasonic SSR-360GP 360W power supply, Fractal Design Node 304 case, and 4x 4TB WD Red drives.

I also has a choice between a i3-4130T 35watt TDP or E3-1231v3 80watt TDP.

I thought since the i3-4130T has only 2 cores with 2 hypertreading that it would be the power saver.
Well it turns out that at Idle the E3-1231v3 and i3-4130T pull the same power (24watts).

I think the added Intel HD Graphics 4400 on the i3-4130T makes up for 2 missing cores.

Of course, under load, the E3-1231v3 pulls 91watts and the i3-4130T is only 42watts. (this included m/b components and memory booted from a USB Flash drive)

The E3-1231v3 computation power is awesome (passmark = 9567)
compared to the i3-4130T (passmark = 4134)

I'm keeping the E3-1231v3!!

I thought I'd share incase someone else is trying to choose too.

 

[joeschmuck]

When you say "under load", what is this specifically?

 

[Aaron Ryan]

When you say "under load", what is this specifically?

I use the following perl code which launches 8 threads:
(perl -e 'while (--$ARGV[0] and fork) {}; while () {}' 8)

Quick and simple, not the best thorough test, but a good way to easily max out your CPU's real quick.

 

[joeschmuck]

Okay, but not real world. What are your values during a scrub. For me a scrub tends to pull the most power from the system, sure the hard drives are more active so the drives and MB will consume more power but the majority of the power consumption will come from the CPU.

I ask this specifically because in my system I only hit 20% CPU utilization when I perform a scrub. The only other time I've been able to exceed that is when I had created my own custom jail template (Ubuntu Jail) running BOINC for SETI which used a bit more without the scrub. I didn't check CPU utilization during a scrub and this jail running, I'm sure it would have been maybe as much as 20% more, maybe.

So I run your test on my CPU and it did run at 100% and ate up 152 watts of power. Mine rests at 72 watts so one day I'd like to get something that eats only 24 watts like yours. Are your drives sleeping? 24 watts, I'm a bit jealous.

 

[Ericloewe]

Well it turns out that at Idle the E3-1231v3 and i3-4130T pull the same power (24watts).

Not the least bit surprising. In fact, it's exactly what's supposed to happen.

I think the added Intel HD Graphics 4400 on the i3-4130T makes up for 2 missing cores.

Idle power, probably. Gating isn't perfect and I don't think Haswell can fully shut down cores.

E3-1231v3 pulls 91watts

Still running quite a bit below TDP, then. At 84W, 7W would be for the rest of the system (essentially impossible).

What I've yet to see properly quantified is power consumed for a given workload:
Is the non-limited CPU going to use less power because it finished more quickly and can now sleep? Is the thermally-limited CPU at an advantage since temperatures should be a bit lower, reducing leakage current? Are the higher frequencies less efficient than the lower ones the T/L model is constrained to?

 

[Jailer]

Okay, but not real world. What are your values during a scrub. For me a scrub tends to pull the most power from the system, sure the hard drives are more active so the drives and MB will consume more power but the majority of the power consumption will come from the CPU.

I ask this specifically because in my system I only hit 20% CPU utilization when I perform a scrub. The only other time I've been able to exceed that is when I had created my own custom jail template (Ubuntu Jail) running BOINC for SETI which used a bit more without the scrub. I didn't check CPU utilization during a scrub and this jail running, I'm sure it would have been maybe as much as 20% more, maybe.

So I run your test on my CPU and it did run at 100% and ate up 152 watts of power. Mine rests at 72 watts so one day I'd like to get something that eats only 24 watts like yours. Are your drives sleeping? 24 watts, I'm a bit jealous.

If you want to fully test your system run handbrake in Virtualbox with the OS of your choice allocating all your cores to the VM and encode a video while preforming a scrub. That should put the load on your system you're looking for.