Shooting straight for -125’C 1.3V
and 2,40MHz+ will lead your GPU to
crash. So it turned out that the best
way to exceed these low limits was
to slowly build up the clock speed
and increase the voltages. Based on
30 The OverClocker Issue 42 | 2017
scaling from lower clocks, I would
raise the voltage to say 1.325V
(lowest under load, so higher than
that with light workloads at around
1.34 to 1.35V) run the exact same
test (3DMark TimeSpy’s GT2 is very
sensitive and a good test for GPU
stability I found) and if that passes
you’re good to go. To test just how
long I could keep the GPU functioning
and if I had proper stability I also
used the HWBOT version of Unigine
Heaven (Extreme preset). As many of
you know, nothing is as long as this
test when you’re cooling with LN2
and it feels like an eternity getting
through all 26 tests.
Once those passed, I simply
bumped up the clock frequency by
25 to 50MHz (usually 50MHz) and ran
the tests again. Using this method,
was slow, but it is what eventually
allowed me to reach 2,500MHz which
I had struggled with for a long time.
The 2,500MHz was stable as well
and it is where the 2nd round of LN2
overclocking will begin. Sorting
out the memory clock and system
efficiency, in addition to knowing how
to approach overclocking of this card
should see a significant increase in
scores.
Right now they are low, but that
is not a reflection of the GALAX
card at all. One needs only to take a
look on HWBOT, where overclocker
OGS using a HOF GTX 1080 Ti has a
frequency of 2,872MHz for GPUPi 1B
and a matching time of 7.985 seconds
for the calculation. The same user
has a 2,645MHz run for 3DMark
Firestrike Extreme with a top five
result. So that may be a silicon
lottery GPU, but the point is that the
rest of the graphics card seems to
have what it takes to propel that very
same GPU to heights few, if any other
GPUs can achieve.
This speaks to the design of the
card I suppose, which for those who
are interested in the technical parts
is covered beautifully as usual on
Actually Hardcore Overclocking
channel here by BuildZoid. So
check it out and get more detail on
everything related to power that can
be discerned visually about the card.
This is where the claims of
superior electrical design and so
on are tested. Plenty of vendors
claim to have better power designs
than NVIDIA has on their Founders
Edition cards (Reference). However
that doesn’t translate into much for
obvious reasons, because barring
temperature these NVIDIA 1000
series cards all have the same limits
which is around 2050MHz when using
air or liquid cooling. So inevitably the
vendors end up selling you a cooler,
through no fault of their own but just
inherent GPU silicon behavior.