Power consumption G530 vs i7-3770T

[Umfriend]

Hi all,

Got a Q: My server currently runs on a Celeron G530, TDP of 65W. I am considering replacing it with an i7-3770T, TDP of 45W. Typically, the CPU will bu under full loan (Rosetta@Home), the GPU will, I guess, not be used at all (headless server).

I am wondering, should I indeed expect lower power consumption? It seems strange going from a 2-core 32mm Sandy Bridge to an 4-core/8thread 22mm Ivy Bridge could actually save power under full load. Sure, the design may be somewhat more efficient, but it would just seem off to me.

Any ideas?

[Jammrock]

The G-series CPUs from Intel generally have smaller on-die cache and lack many of the features, like speedstep, to run the processor more efficiently. If Rosetta@Home can take advantage of the built-in GPU it will, even if the server is headless. May be a good way to boost your numbers.

The power savings are real. The 22nm die process is much more efficient. Haswell goes ever further, but mostly for idle power. If your server is not idle there is not real advantage to going Haswell core.

The faster CPU+GPU should really boost your Rosetta@Home score.

http://www.cpubenchmark.net/cpu.php?cpu=Intel+Celeron+G530+%40+2.40GHz

http://www.cpubenchmark.net/cpu.php?cpu=Intel+Core+i7-3770T+%40+2.50GHz

[Umfriend]

Thanks,

Rosetta@Home does not use the GPU. Not looking for Haswell as I'd like to keep my current MB so as to not waste money (OK, schizofrenic I guess). I know about the score increase, it'll be huge. But do you really think the fully loaded i3770T will have lower power consumption as the G530?

[Jammrock]

Yes. The TDP's from Intel are pretty accurate.

[Umfriend]

That's the thing though. I have read around and never ever found anyone actually suggesting the G530 can use 65W at full load. Highest ever came up was 30-40W. Or should I expect, e.g., 25-30W for an i7-3770T?

[dZeus]

That's the thing though. I have read around and never ever found anyone actually suggesting the G530 can use 65W at full load. Highest ever came up was 30-40W. Or should I expect, e.g., 25-30W for an i7-3770T?

afaik, Intel's TDP is the max. power usage for all CPUs in a certain bin. This does not mean that each CPU will act like the worst performers within the bin. There might also be additional things going on, such as a production of CPUs which could be rated at lower TDP vs. lower demand for these CPUs, letting the 'good' CPUs end up in bins with higher TDP than accurately reflects their capabilities.


but this is just my understanding. Maybe someone who is more familiar with how Intel's TDP rating works can explain all the details...

[Umfriend]

Yes, please.

[Jammrock]

Processors are produced on a "wafer". A big circle of silicon with hundreds of raw processors on it. Processors in various regions of the wafer end up with different performance characteristics. Most of the processors in a generation are identical, they just came come from different parts of the wafer so they can perform differently.

Dual-core procs are often quad-core with one or two bad cores. Intel simply disables some of the cores and features so they can resell it as a lower end processor.

When one processor type of say, a lower performance spec, is a hot seller they will use higher end stock to augment the lower end line. So sometimes you buy a lower end proc with higher end performance characteristics. As dZeus said, the TDP is the maximum watts of power that the line will dissipate based on the characteristics tested in that area of the wafer.

Intel recruited heavily from my tech program. Learned lots of fun things from their recruiters

[Umfriend]

So Intels TDP may be accurate as a maximum, just not as a maximum for any single CPU? My G530 might be a CPU that uses less power than, say, an average i3-2100T that has 2 cores @ 2.5Ghz, 4 threads, 3MB of cache (instead of 2.4Ghz, 2 cores/2 threads, 2MB cache) with a TDP of a mere 35W? Seems like there's only one way to find out: measure. Wonder where I could borrow a power consumption meter and an i3-3770T.

[Jammrock]

Pretty much. The TDP's on the Core line are with the GPU and CPU running full tilt. If the GPU is idle your max TDP will go down significantly. I still think the i3-3770T will be faster and cooler than your G530, but as you said the only way to know is to measure.

There is a device called a kill-a-watt, and several like it, that you can use to measure power draw at the outlet. This will measure whole system power draw mind you. You could get a baseline on the G530 and then measure the 3770T to compare. Granted, you still need a 3770T.

[dZeus]

So Intels TDP may be accurate as a maximum, just not as a maximum for any single CPU? My G530 might be a CPU that uses less power than, say, an average i3-2100T that has 2 cores @ 2.5Ghz, 4 threads, 3MB of cache (instead of 2.4Ghz, 2 cores/2 threads, 2MB cache) with a TDP of a mere 35W? Seems like there's only one way to find out: measure. Wonder where I could borrow a power consumption meter and an i3-3770T.

In case you're curious, I've ordered an Intel Core i3-4130T with 35W TDP (building a system for a relative), and can do some power measurements when everything has arrived and when I've assembled the thing. I have a simple wall-plug power consumption meter that gives semi-accurate readings...

[Umfriend]

I am curious.

[dZeus]

I am curious.

ok all components have arrived, I'll post a power reading this weekend.

meanwhile, while reading up on low power models of Intel CPUs, the consensus among the hardware enthusiasts seems to be that the low power desktop models have not specifically been selected for having less power leakage (or being better suited to run at lower voltages), but are just regular desktop CPUs that have been limited to run at lower frequency.

Most people claim you can achieve the same results with limiting the frequency of a regular desktop CPU in software. See for example here for some opinions on this.

[Jammrock]

It's called underclocking. Amazing how a decade ago we tried to overclock to suck out every bit of power, and now we underclock because we have too much

I've read a lot of similar reviews. Underclocking and undervolting can reduce power consumption a lot, at the cost of performance, of course. It's a trick that a lot of silent computer builders use. Undervolting the proc can be very dangerous though.

http://www.makeuseof.com/tag/how-undervolting-decreases-heat-increases-battery-life/

[Umfriend]

It just goes to show how silly I really am. I'm looking for a solution that will use less power at full load just so it can run at full load so that I get shitloads of credits. The right way to look at this of course is to buy the most efficient solution and then, if power consumption is to high, turn it off from time to time.