¨0.06 KWh to emerge gcc-7.3

Irre · Guru Joined: 09 Nov 2013 Posts: 434 Location: Stockholm

I emerged gcc on two different architectures, and came to the conclusion that energy consumption is the same.

First on Raspberry PI first generation:

genlop -t gcc

Wed Feb 7 01:54:05 2018 >>> sys-devel/gcc-7.3.0
merge time: 1 day, 17 hours, 24 minutes and 26 seconds.

Estimated power: 2 W
Energy consumption to build sys-devel/gcc-7.3.0: (24+17+24/60)*2 Wh = 0.063 KWh

Then on Intel i7-2670QM CPU 2.20GHz:

genlop -t gcc

Wed Feb 7 10:11:07 2018 >>> sys-devel/gcc-7.3.0
merge time: 1 hour, 6 minutes and 57 seconds.

Estimated power: 50 W
Energy consumption to build sys-devel/gcc-7.3.0: (1+7/60)*50 Wh = 0.055 KWh

Wallsandfences · Guru Joined: 29 Mar 2010 Posts: 378

Irre · Guru Joined: 09 Nov 2013 Posts: 434 Location: Stockholm

you are right! (24+17+24/60)*2 Wh = 0.083 KWh

NeddySeagoon · Posted: Wed Feb 07, 2018 1:10 pm Post subject:

Irre,

I think both your power estimates are low but the result is expected.

Silicon performance has long been limited by the energy density its possible to extract to keep junction temperatures within bounds, your two processors will run at the same W/m^2 of die.
It follows that a smaller die must take longer to perform the same amount of work.
This is true regardless of the feature size on the semiconductor.

To increase the energy density, either the die must be run hotter, that demands novel semiconductors. They are around, in blue LEDs.
or novel cooling to dissipate the heat better. That's around too, its just very expensive. Water cooling is not the sort of thing I had in mind but its an example.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

Irre · Guru Joined: 09 Nov 2013 Posts: 434 Location: Stockholm

Compiling with /var/tmp in tmpfs was much faster:

Wed Feb 7 20:00:41 2018 >>> sys-devel/gcc-7.3.0
merge time: 43 minutes and 2 seconds.

Estimated power: 50 W
Energy consumption to build sys-devel/gcc-7.3.0: (43/60)*50 Wh = 0.036 KWh

John R. Graham · Posted: Wed Feb 07, 2018 6:25 pm Post subject:

There are some architectural differences that are a factor as well. The highly refined & much-iterated x86 CISC still beats any RISC, clock for clock and (usually) watt for watt.

- John
_________________
I can confirm that I have received between 0 and 499 National Security Letters.

eccerr0r · Posted: Wed Feb 07, 2018 6:59 pm Post subject:

<sarcasm>
Gentoo is costing us $0.01 every time they make us compile gcc-7.3, (webkit, chromium, firefox, rust, ...) and $0.02 for openoffice... there's my two cents...
</sarcasm>
:)
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

NeddySeagoon · Posted: Wed Feb 07, 2018 9:21 pm Post subject:

John R. Graham,

The highly refined & much-iterated x86 CISC is actually a RISC core and has been for many years.
The CISC part is emulated by the microcode on the RISC core.
_________________
Regards,

NeddySeagoon

Computer users fall into two groups:-
those that do backups
those that have never had a hard drive fail.

Naib · Posted: Wed Feb 07, 2018 9:27 pm Post subject:

John R. Graham · Posted: Wed Feb 07, 2018 9:30 pm Post subject:

It's not that simple (but I know you know that). For instance, branch prediction and speculative execution, which are both part of the "highly refined" part, are not done (or not all done) in microcode, but in hardware. Now you can go ahead & tell me that these techniques are equally applicable to RISC architectures.

- John
_________________
I can confirm that I have received between 0 and 499 National Security Letters.

Naib · Posted: Wed Feb 07, 2018 9:59 pm Post subject:

eccerr0r · Posted: Wed Feb 07, 2018 10:38 pm Post subject:

I don't know why speculative execution and branch prediction give CISC so much flak wnen the Alpha by (formerly) DEC had these too...and it's RISC. These performance enhancing features are merely omitted for simplicity for many processors because solely of complexity - not due to instruction set, but to hardware to ensure correct operation, though it's even more difficult on CISC and legacy code.

I suspect the demonization is solely due to the outstanding spectre/meltdown attacks. It boils down to a tradeoff, do you want GCC to finish compiling in 1 hour or 1 day, especially if you're trying to code/debug GCC...

Or in our case, run Gentoo... because a Gentoo box spends a lot of time with the gcc process in the run queue than many other Linux machines.
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

Naib · Posted: Wed Feb 07, 2018 10:45 pm Post subject:

eccerr0r · Posted: Thu Feb 08, 2018 10:47 pm Post subject:

Speaking of ramdisk, is the wattage also including disk power consumption for both machines? If including the same disk/storage consumption for both, I suspect the faster machine gets even more attractive to use.

However the faster machines do have dead weight like video cards and keyboards that don't help towards compilation, so that also may need to be accounted for.
_________________
Intel Core i7 2700K/Radeon R7 250/24GB DDR3/256GB SSD
What am I supposed watching?

Irre · Guru Joined: 09 Nov 2013 Posts: 434 Location: Stockholm