[HARDWARE] Cost of crunching (was: Mac Questions)

Thord Nilson thordn at stacken.kth.se
Sun Mar 17 23:36:26 EST 2002


On Sun, 17 Mar 2002, Peter Cordes wrote:

> On Wed, Mar 13, 2002 at 03:50:21PM -0700, John L. Bass wrote:
> > It doesn't work to just use processor power, there are a lot of other components
> > in the "system" ... even with high integration PC's which clock the front side
> > buss (and everything on it) at 100/133mhz. The LCIII will not tick unless the
> > processor buss and get memory cycles ... and the memory controller and DRAM's
> > use power too.
> 
>  You can calculate the incremental cost of having the CPU working hard vs.
> idling, assuming the computer would be left on all the time anyway.  (If
> that's not the case, then the incremental cost of crunching is a lot higher.)
>  
>  On x86 CPUs, I know there is a halt instruction that stops the clock and
> everything on the CPU until the next interrupt.  Most OSes for that platform
> use it in their idle loop.  Win9x is a notable exception, but even without
> it, running an idle loop switches a lot less transistors (and thus uses less
> power) than d.net.  This being the hardware list, I don't think I need to
> elaborate on why d.net uses more power than pretty much anything else you
> could run.
> 

I have measured the power taken from some systems when running RC5
versus idling on Linux. I have used a FLUKE 41 power meter to get
accurate results.

In WIN95 there is no significant difference if the machine is 
idle or crunching. (It even takes LESS power with certain programs)
Also the RC5, DES, OGR CSC takes different power. (on some machines DES
takes more than RC5 on onther less)

Assuming:
All machines run on Linux.
The cost for 10 KWh is 1 USD.
The machines are on anyway, only extra power is from crunching vs idling.
Total # of keys to check = 2^64.
USD = Delta/10000/3600/Perf * 2^64  (Perf is in keys/sec)

                    Power in W                 Performance  M-USD to
System              Running RC5  idle   Delta   Mkeys/s     complete***

486-DX4-OD 100 MHz     112.8     102.8   10      0.093       55.1  (621)
Alpha 21064 100 MHz**   88.8      85      3.8    0.086       22.6  (528)
Pentium-MMX 200 MHz     41        32      9      0.428       10.8   (49)
AMD-K6 233 MHz		72.5      35     37.5    0.400       48.0   (93)
AMD-K62 500 MHz         84        57     27      0.864       22.2   (69)
2xCeleron 450 MHz      116.7      55     61.7    2.544       12.4   (23)
Athlon TB 1100 MHz     115        98.3   16.7    3.388        2.5   (17)

*   Celeron 300A overclocked to 450 MHz.

**  Running OGR actually takes less power than idling.

*** Number in () is if machine is on because of crunching.
    This number is not really representative for the CPU since all
    machines have different pheripherals, like sound card, number
    of HD:s CD-ROM/CD RW, memory etc...


/Thord.


--
To unsubscribe, send 'unsubscribe hardware' to majordomo at lists.distributed.net



More information about the Hardware mailing list