On 1/3/25 03:23, tkoenig--- via cfarm-users wrote:
Am 03.01.25 um 02:40 schrieb Jacob Bachmeyer via cfarm-users:
> I have a philosophical view that idle time on servers is essentially
> wasted: a sunk cost.
In the age of TTL, ECL or even NMOS, that might have been true, power
usage was pretty much independent of the computational load in
those days.
Now we have CMOS and dark silicon, gate switching (aka computation)
determines energy consumption and the things that go with it (cooling
costs, semiconductor aging, etc).
For smaller machines, yes, with a few caveats that not all architectures
actually do that: if the clock is not gated off, the gates still
switch. Larger systems tend (as far as I know) to be less likely to
actually take those power-saving measures. The very newest models
probably do, however, I admit.
For typical rack servers in datacenters, I believe that cooling and
aging are still sunk costs. Cooling and power are aggregated across
many nodes, (and possibly tenants in co-location facilities) and the
servers will be obsolete long before they appreciably age in any case.
Now the cfarm may have secondhand nodes that *are* already "obsolete"
from datacenter use, and running in someone's apartment or office, where
cooling is less of a sunk cost, but where they also contribute less of
an incremental cost, if any. (If the server burns 1kW, but you have to
dump 50kW of heat from insolation anyway, the server is not
significantly contributing to your cooling costs---and only slightly to
your electric bill, at least compared to your air conditioner.)
-- Jacob
_______________________________________________
cfarm-users mailing list
cfarm-users@lists.tetaneutral.net
https://lists.tetaneutral.net/listinfo/cfarm-users
