you may want to refer to it as. This
has little to no effect on overclocking
headroom when dealing with the air
cooling. It is not something that we
may want to acknowledge but it is
true. With the GM200 cores as used
on the 980 TI at least, temperature
seems to be the limiting factor
when dealing with air and water
cooling. Even with the low operating
temperatures courtesy of what is
arguably the best-crafted cooling
complex on any commercial graphics
card to date. There simply isn’t a way
to keep temperatures low enough
for the GM200 core to scale past the
already established and familiar
1555MHz~ ceiling. No tuning tool will
help alleviate this and certainly no
home brewed BIOS. This limitation
speaks more of the silicon and what
protection mechanisms NVIDIA may
have in place rather than any one
thing EVGA or any other vendor could
do. For the same reason that a CPU
that has a limit of 4.7GHz is not going
to get any better regardless of what
motherboard you place it in. There’s
no way around that with liquid and air
cooling.
So with that established, does that
mean the 980 Ti KPE isn’t worth
overclocking on air cooling? Not at
all, if anything it will still provide
far more fun than the vast majority
of GTX 980Ti cards on the market,
especially given that unlike what may
hold true for some competing cards,
there is no special software required
here. With EVGA’s Precision tool, you
can maximise performance on air
and water-cooling and perhaps even
with dry ice, which I will get to later
on.
For one, unlike many GTX 980 Ti
cards on the market, this model
uses Samsung GDDR5 memory.
You should be familiar with the
advantages these memory chips
bring over the regular SKHynix
sourced GDDR5. In general these
chips overclock further
and seem to have better timings,
thus leading to slightly better
performance. This isn’t something
you’d observe normally as most
people would not have two of the
same cards next to each other only
differing in where the GDDR5 is
sourced from. However, via trusted
sources and in house testing, one
can confirm that this general
perception of the performance
differences between the memory
vendors holds true. With that said,
reaching a 2GHz memory clock with
this card is guaranteed and perhaps
20 The OverClocker Issue 36 | 2015
even 2050MHz will work just fine
with no tuning at all. Where you may
need to add a little more voltage
to the memory via a 3rd party tool,
is past this point. Some graphics
cards have been known to reach as
far 2250MHz, but a safe bet for just
about all of the K|NGP|N cards is
2050MHz.
Again, do not take this memory
frequency for granted as there are
many cards which will not reach
1950MHz let alone the 2GHz mark. It
may only be 50MHz*, but it’s a 50MHz
that’s there for certain. You should
also keep in mind that memory
frequency is also dependant on the
memory controller within the GPU,
so if you are not able to go much
farther than 2GHz, that is more than
likely a reflection of your card’s
silicon quality rather than anything
on EVGA or SAMSUNG’s part. It
really is a silicon lottery and the
controls that you have at your
disposal to overcome that are
limited. Do not despair however, as
where an untouched DRAM
frequency may stop, the classified
tool available here, may get you that
much farther. Using voltages up to
1.75V exclusively for overclocking
sessions should not be an issue,
however if you’re planning on
gaming at the overclocked memory
frequency and voltage, it is best to
keep to a frequency that obviously
produces no artifacts and more
importantly needs no additional
voltage applied, this will ensure no
long term damage to the memory.
Where the core is concerned, well,
there’s not much to say. This
particular sample would do 1555MHz
3DMark FireStrike Ultra stable,
and not 1MHz more regardless of
the additional voltage applied. For
HWBOT Unigine Heaven Extreme