Our Earth is dying. I don’t need to explain or prove this. All knowledgable folks on this planet already know how we have set it up for doom, unless we act on averting it now. If 7 billion humans can ‘mistakenly’ destroy an entire planet, they can also revive it back into full spring.

In all honesty, I have a terribly soft spot for the Earth’s bio-diversity. Be it the evolution of funky birds and their quirky dances to attract a potential prom-date or be it the wrestling of two beetles for territory - a small branch on a shrub. As a kid, I used to watch Animal Planet, a brilliant television channel that streamed documentaries about animals. Scenes from these documentaries kicked my heart in all directions, mocking how minuscule the scale of my experiences was. I have no shame in admitting that watching such documentaries, coupled with amazing cinematography, pushes me into an intense swirl of emotions, which seldom erupts out as a teardrop.
As a gift from my side, here’s a 2-minute video on walruses. Walruses dive in the oceans to hunt for prey and then come up to the icy shore to rest. However, since the ice-sheet is rapidly narrowing and thinning, the walruses have to seek refuge on rocky cliffs. What follows, can not be described in words alone.


This is just a couple of minutes of the second episode of the Our Planet series.



To realise that somehow, indirectly, I am part of the reason why walruses in Russia go through this, is incredibly heart-wrenching.
My interest to work on better compute-architectures is not a mere application of my passion for all things electronics and programming. It is significantly fuelled by a deep-rooted interest to avert the climate crisis.

Why should I optimise compute clusters though? Is it that significant?
Compute clusters, data farms, etc, use insane amounts of energy. For example, this Forbes article describes how Google alone uses about 12.4 terawatt hours of electricity in 2019. To put this in perspective, this much energy is more than enough to power entire Sri Lanka for a year!
Optimising an architecture by even 1% is a great contribution towards controlling irreversible damage.

But you might claim that making better engines for vehicles, moving to renewable sources of energy, etc are better ideas. What I feel is that some physical conditions constrain the efficiencies we can achieve. For example, the concept of Carnot cycle, on which the air-conditioner/refrigerator work, is only about 70% efficient in the present day. If your refrigerator is about 8-10 years old, the efficiency will be about 40-45%.
Because humans have an operating temperature range (outside which they simply die, pfff pathetic), all thermodynamics related to energy-conversion needs to happen around this temperature. Now, this temperature range is not a suitable condition for achieving maximum efficiency. Safe to say, it will be impractical to achieve extremely efficient devices in the near future.

We are going to need a lot of compute in the years to come. Everything in the world is shifting to a digital landscape - from research simulations to government documentations. I feel that a shortage of silicon is underway. I think that markets related to semiconductor materials will become an immense factor in economy.
To humanity, as it stands right now, finding a planet with rich reserves of pure silicon is much more valuable than finding a planet of diamonds!
This explosion of semiconductor requirement is in some ways, similar to the trends in the transport industry. Cars and bikes are akin to personal electronic devices; and cargo ships and airlines are compute clusters managed by large corporations. Everyone will be using these technologies at some point. It took several decades before regulations on engine-emissions were enforced. Consequently, new engines were designed, leading to cleaner alternatives.
However this time, we’re significantly aware of our impact on the environment. Emission standards and power regulations already set in place for compute farms or coin-mining facilities are doing a good job. But a better job can be done. Before we actually face a crisis of smoke, we can start building a better engine for the future.

That’s why I want to optimise compute, even though we have it fast enough for now.
That’s also why I turn off that extra light that’s still on.
Not because it will, later on, at some point in time, matter in the grand perspective. No.
I turn it off because it matters, right now, in this moment.
I turn it off for me; for the Earth that is dying.