friendly rivalry between fellow hobbyists and
their teams, which in turn motivates people to
improve their folding credentials. It’s a great
way to show off the power of your setup, while
also contributing to a good cause.
‘We have some benchmarking machines
that we use as our standard,’ explains
Bowman, ‘And we decide to set the base
points based on how long one of these work
units takes on our benchmarking machines.
We’re then set up to award people based
on completing the work – the faster you
complete it compared with the benchmarking
machines, the more points you get.’
Core blimey
The Custom PC team (35947) has been
folding since 2004, so we’ve seen see how
the project has changed over the years. When
we first started, we all had single-core Athlon
XP and Pentium 4 CPUs (and no GPU folding
client), and it took a long time to hit 20,000
points, let alone a million. Later, the first dual-
core and later quad-core CPUs enabled
people to clock up many more points in the
same time.
Since then, GPUs have taken over in terms
of churning out points (more on that later),
but CPUs still play a very important part. You
can set up your client to fold on both your CPU
and GPU, and while that might be inefficient
in terms of points per watt, your CPU can still
work on valuable research that can’t always be
done by your GPU.
‘There are actually some things we can
do on CPUs that we can’t do on GPUs,’ says
Bowman. ‘With the current pandemic in mind,
one of the things we’re doing is what are
called “free energy calculations”, where we’re
simulating proteins with small molecules that
we think might be useful starting points for
developing therapeutics, for example.
‘Using these calculations, we’re able to
get a sense of how tightly different small
molecules bind to the COVID-19 proteins, and
we can then prioritise the tightest binders for
subsequent experimental testing. Right now,
these calculations are only implemented in
our CPU-based code on Folding@home.’
The CPU code also scales exceptionally
well to having more cores, as well as extra
threads using Intel Hyper-Threading or AMD
Simultaneous Multithreading. Open your
Windows Task Manager while you’re running
the CPU client, and you’ll see every core’s
activity box running flat out, whether it’s
physical or virtual. You can even fully stress a
Threadripper CPU, and there’s room to grow
further with more cores.
The number of cores that could be used by
the CPU client ‘really depends on how big the
protein and the surroundings we simulate
are’, says Bowman. ‘The bigger systems
could easily be running on an order of 100
CPU cores, and making effective use of them.
Some of the smaller systems wouldn’t scale
that well.’The Demogorgon opens up
An example of a big system that could run
on 100 CPU cores is ‘the simulation of the
COVID-19 spike,’ says Bowman. ‘We’ve beencalling it “the Demogorgon” because of this
opening motion it does – the opening of the
three receptor-binding domains reminded
us of the mouth of the Demogorgon from
Stranger Things.’
Analysing this opening motion was a key
project for Folding@home, as there was no
way to observe it using standard experimental
techniques. The aim was to understand how
the COVID-19 spike protein opens up to bind
to a protein called ACE2 on human cells.
‘Specifically, the spike has three receptor-
binding domains that directly bind to ACE2,’
Bowman explained on his blog. ‘For the spike-
ACE2 interaction to form, the spike’s three
receptor-binding domains must open up to
reveal the binding interface.’THEBIGGER
SYSTEMSCOULDEASILY BE RUNNING
ONANORDEROF100 CPU CORES
The CPU client maxes out nearly every thread on
an 8-core (16-thread) Ryzen 7 2700X
The opening motion of the ‘Demogorgon’
COVID-19 spike reveals the ACE2 binding
site for interaction with human cellsFEATURE/ ANALYSIS