ANINTERVIEWWITH OVERCLOCKING CHAMPION 8PACK
Ian ‘8pack’Parry is a professional overclocker
and system builder for Overclockers UK
Asus, which is my motherboard of choice on
LGA2066, I boot with 1.1V SA and IO but tune the
IO down to 0.92V. Doing these small tweaks
saves 8-10°C, and don’t forget to apply an AVX
offset of minus three or four and even 11 or 12
with AVX512 to reduce its frequency a lot for
these workloads. This is even a tweak we do
this with customer systems.
What about AMD CPUs?
On Socket AM4, you need to benchmark and
overclock in your specific programs to see
if it helps. I’ve found that overclocking rarely
helps in lightly threaded applications as the
single core boosts are high on AMD CPUs, so
you’re usually better off using Precision Boost.
However, in multi-threaded applications, you
can easily gain 20 per cent performance using
a manual overclock. It’s worth looking at the
voltages Precision Boost is using too. I find that
a manual overclock can lower heat and power
consumption in multi-threaded workloads. On
Socket AM4, I’ll use around 1.35v maximum,
which is often less than PBO, so that can save
heat and temperatures under load. On Socket
TRX4 and Socket AM4, you can usually save
temperatures by tuning SOC voltages and
again lowering memory voltages.
What about memory overclocking
and frequencies?
With LGA115x, there’s very little difference above
3200MHz in real-world benchmarks, so adding
more heat by increasing voltages isn’t a good
idea. In fact, it’s worth lowering the DDR voltage
as much as possible. Many 3600MHz kits can
run at much lower voltages anyway. In any
event, I’d start by getting the most out of the CPU
core. Then I’d tune the cache or Infinity Fabric,
and the last thing would be memory. It’s always
a balancing act between different types of load,
so you need to work out what your priorities are
first. That’s how we tune systems for customers.
With AMD, 3600MHz, memory is usually
the best option, as most CPUs will do this linked
with Infinity Fabric. Occasionally I might reduce
the ratio to get a higher CPU frequency, but you
lose efficiency. Also with AMD, the main things
I tweak are loadline calibration to prevent droop
on both core and SOC. Also max out the power
and current limits where possible, but pushing
memory higher using the dividers doesn’t really
aid 24/7 or extreme overclocking.
In your experience, does spending more
on cooling on a home setup scale at all with
overclocks? For instance, going from mid-
range air to high-end air to high-end AIO then
DIY liquid cooling?
It depends on other limiting factors really. Right
now, with soldered heatspreaders, you can get
100-125MHz extra with good water cooling
compared with air and of course, it results in a
much quieter PC too.
The thermal paste Intel used for a while
was definitely a limiting factor. With AMD
mainstream CPUs, though, not so much. A lot
depends on workloads too. If you need good
AVX512 performance then on LGA2066, there’s
not a chance you can do that with air cooling.
This is why the best stability test for anyone’s
overclock is an actual workload.
S
ome of us are lucky enough to
overclock various systems on a
regular basis, but we chatted to
someone who does it on a daily
basis and in extreme ways too. Ian Parry, better
known as 8Pack, is one of the world’s best
overclockers and currently builds crazy PCs for
Overclockers UK. We caught up with him to get
his top tips on other settings you can tweak and
opinions on all things overclocking.
So Ian, what would be your top tips for the
various CPU sockets around at the moment
when it comes to overclocking?
I’d start by setting the XMP profile on the
memory in the EFI. Then I’d make sure the
board is setting correct voltages and timings
for the memory too. I’d also set a degree of
loadline calibration, so the voltage doesn’t
droop under load and cause instability. On
Intel CPUs, I’d max out the CPU and memory
current limits, as these can hinder even modest
overclocks on some boards. In addition, I’d also
max out the power limits.
From there, it can often be a case of lowering
some voltages to rein in temperatures. On all
Intel CPUs I save degrees of heat by tuning
down SA and IO voltages. With LGA115x, you
can also lower the memory voltage quite a bit,
usually to around 1.15-1.25V, especially with
3200MHz kits. Just this change can result in
between 5-10°C temperature savings.
On LGA115x, I also tune the cache to gain
further efficiency. All processors can do a x45
cache ratio, so start at that. Increasing the cache
ratio by two or three points is the same as
100MHz on the CPU, so it’s worth doing.
For gaming-only systems, I would disable
Hyper-Threading too. You can easily save up to
10°C under load and gain an extra 100MHz. If
you have a 6-core or 8-core CPU then Hyper-
Threading is probably not even helping in
games and can actually result in slower frame
rates in some. On LGA115x, you should apply
a -3 ratio for AVX workloads. They generate a
lot more heat, so you need to make sure your
overclock is stable if you deal with them.
On LGA2066, you can lower temperatures
again by tuning SA, IO and cache voltages. Once
you find these in the EFI, and once you have
a stable overclock, try to reduce them below
their stock settings. For example, 0.92V on
the cache voltage still allows for a x28 ratio. On
FEATURE / CUSTOMISATION