It's worth noting, since people seem to not be fully aware of it, that
temperature on modern nVIDIA cards (particularly Kepler and on, but to an
extent even Fermi) is NOT to be considered the way you are used to.

While for marketing reasons nVIDIA suggests it's a dynamic overclocking of
sorts (GPU boost), modern cards are designed to operate at peak and
constantly throttle down from there (so in reality it operates like a
safety measure does on CPUs).

What that means is that, in example, a Titan will actually be pretty much
at 80+ Celsius all the time try to hit max clock (some will hit 1k and
above, some 960). That isn't a bad thing, it's designed to operate at that
temperature.
What the hardware does is vary several parameters, chiefly the clock, until
the temperature reaches a cap, and then constantly vary those parameters to
keep it there or just under.

While in the past constant throttling could be bad for performance, and
still is for CPUs on most tasks, Kepler and after GPUs are designed around
it.
So if you buy a 9xx, or a top end 7xx that's not a Maxwell preliminary (not
the 750 basically), and it's constantly at 79 on die, don't freak out.
To keep it below that your only option is to forcefully clock lock it, or
lower the caps for temperature or for the clocking plateau, depending on
what options the manufacturer's control panel offers.

Quadro budget Keplers like the 4200 and the 5200 work the same way I
believe, except that for cards like the 4200 they have such crap specs that
even boosting to cap they are unlikely to even scratch 80C, most of the
time they don't have the transistor count to make it to 70 :p

On Sat, Dec 13, 2014 at 12:36 AM, <hk-v...@iscs-i.com> wrote:
>
>  Read many horror stories but never that bad. Even 3D mark never cranked
> it past 70 C.
>
>
>

Reply via email to