A big part of the work that goes into building hardware is the supply chain. It’s hard to fathom how complex and time sensitive it is without having worked on it. When we built a device, there could be several hundred components. Each of these had to be ordered, shipped and soldered on. Some components were only available in some countries because that’s where the volume justified selling them, we had to pick alternatives.

The hardest part was using chips that were not revealed yet. If Intel announced a new chip in March of a year, we probably would have been working on it for a few months already. Testing with prototype chips. These prototypes had silicon bugs and we had to write code to evade them.

Just like we had multiple board versions, Z1, Z2, A1, B1… Chips also had successive versions where bugs were progressively physically ironed out. The new chips were either already being assembled into devices when the announcement was made or sometimes a click away from production, already assembled and racked but under embargo.

You also end up having bugs that you can’t do anything about. When a CPU dies because of an erosion in the clock transistors inside the chip, there’s not much you can do. It’s bricked. It’s worse when that erosion happens just because the device is up and running. Especially when the customer reboots into a bricked server.

Behind one device, behind one chip or board there are so many steps and synchronization points it’s truly humbling. Software looks really simple in comparison. It could be that it’s the reason that software is everywhere, it’s so much easier to work with. Much faster and much cheaper…

When the post Covid logistic crunch happened in 2021, I could all so clearly understand and feel the people who could not manufacture because one of the components on the board depended on multiple actors and steps.

All that work is invisible to 99% of the tech world. It’s invisible precisely because the abstractions are close to watertight. It’s been made even more watertight since the advent of the cloud.

The cloud makes it very easy both at the user level and at the financial level to start something. I now joke that the cloud is a financial tool. It spreads your CapEx into OpEx but at the time I was much more into the trenches. Carefully handling prototypes that were mass manufactured and rented out to customers. The financial part was making sure we made the right decision to slice costs at the right place. With the subsequent years, my insight got broader and more nuanced.

A new set of assignments were to be my last ones in the lab, but I didn’t know yet.

​

If you have missed it, you can read the previous episode here​

​

To pair with :

• Last train home - Lostprophets
• The Reality Dysfunction by Peter F. Hamilton