Core i3/i5 undervolting: how long can we go?

[quest_for_silence]

Well, I'm looking around for a new rig, and the Core i3/i5 are among the best candidates.

As in subject, how long can you undervolt them? With which motherboard/software?

Any hint will be appreciated: thanks in advance for sharing.

Regards,
Luca

[lodestar]

It seems to be possible, see http://www.tomshardware.co.uk/25w-perfo ... 816-4.html. The way a Core i3/Core i5 works is that it will ramp up the CPU voltage under load, but by less if it starts from a lower base. So there is a benefit, in this case a claimed reduction of overall system power consumption to 76w from 80w, or around 5%.

[MtnHermit]

I tried undervolting my i3-530 on a Gigabyte H55 MB. I could only get to 1.0V before it refused to POST AND their was NO power savings.

I subsequently learned all the Core iX CPU's have a million transistor power manager on chip to dynamically control power and clock. Basically, Intel has taken all the fun out of under/clocking/volting.

[BillyBuerger]

I tried undervolting my i3-530 on a Gigabyte H55 MB. I could only get to 1.0V before it refused to POST AND their was NO power savings.

I too have an i3-530 on a gigabyte H55 MB. Using the BIOS options, I was able to undervolt by 0.1875 which put it at 1.0V under load. It seemed to be fine under stress but I did have a crash or two randomly even when not loaded. So I bumped it back up to about 1.025V or so. Yes, idle power did not change one bit by undervolting and pretty much stayed at 28W AC or just under 20W DC (estimated). Which was sad but also a testament to how efficient these chips are. But under load it dropped from about 71W to 58W AC. I would say that's worth something. I tried changing the IGP voltage but that didn't seem to do much either.

[smilingcrow]

So I bumped it back up to about 1.025V or so. But under load it dropped from about 71W to 58W AC. I would say that's worth something.

I’m just building a system with a similar combo of CPU/Mobo and that sounds a decent saving. What application did you use to load the CPU when measuring power consumption?
I usually use Orthos but when I initially ran it only the first 2 cores in Task Manager were loaded and I’m not clear which are the real cores and which the SMT cores. Would someone please confirm that for me?
I’ll try running two copies of Orthos and play around with the affinity and see how that affects power consumption and temps.

[quest_for_silence]

Which are currently the very best options, for their inherent undervolting capabilities, among the P55/H55 motherboards?

Regards,
Luca

[BillyBuerger]

I’m just building a system with a similar combo of CPU/Mobo and that sounds a decent saving. What application did you use to load the CPU when measuring power consumption?
I usually use Orthos but when I initially ran it only the first 2 cores in Task Manager were loaded and I’m not clear which are the real cores and which the SMT cores. Would someone please confirm that for me?
I’ll try running two copies of Orthos and play around with the affinity and see how that affects power consumption and temps.

I use CPU Burn-in. Not sure how it measures up for maximum load. But I found it a couple years back and have continued to use it. To get to a full 100% CPU load, I ran 4 instances of it.

I don't know a lot about HT but I'm pretty sure you don't have real and "fake" cores. Each real core acts as two cores. I would expect that if you're using 2 cores, it probably balances it so that you're using one from each physical core. Then if something starts using the 3rd and 4th core, it might just slow down the other cores since it than has to share physical cores.

For instance, if I run 4x CPU burn-in and then try to run FurMark, FurMark moves very slow since it does need some CPU power. But if I just drop down to 3x CPU burn-in, then FurMark runs just fine even though it's sharing that 4th core with one of the CPU burn-in instances.

[ilovejedd]

I use Intel Burn Test. It's usually what gives me the highest temp and power consumption values.