The Ridiculous Advancements of Hardware

Digital computing hardware has advanced tremendously over the last 70 years. It's common to say a cellphone is thousands of times more powerful than the computer that took astronauts to the moon. Or that it's a supercomputer. While these statements can help us understand hardware advancements in the abstract, they don't tell the whole story. More importantly, they don't tell the story on an equal field. Sure, your phone might be more powerful than the Apollo Guidance Computer or a supercomputer from a few decades ago, but that doesn't mean it could be used for either of those tasks. We need an equivalence, not an analogy. We need to compare like things in a way that helps us truly understand the scale of difference.

To start, let's talk about the two fundamental properties of any modern processor: transistor count and clock speed. Throughout computing history increases in transistor count and clock speed have led to more capable digital computers and more efficient software. Moore’s Law is exponential, but that's challenging to grasp on the scale of decades.

For simplicity, we'll combine transistor count and clock speed into a single property, transistor speed, and use this to compare computing hardware across time.

But what computing devices should we compare? Ideally, we want two devices that are the same except for being created years apart. In the 1970s, things like C, Unix, x86, and TCP/IP were created, so we can choose computing hardware from this era as it's reasonably modern. Apple was created during this decade and still makes computing hardware today. Using computers from the same company allows us to control for the most variables. We'll compare computers made by Apple used for roughly the same purpose, by a similar audience, that cost the same amount in today's dollars.

In 1977, Apple released the Apple II. At the time of release it costs $1,298, which is equivalent to $6,530 in today's currency. In 2023, Apple released the second iteration of the Mac Studio. At the time of release, a Mac Studio configured with max processor and RAM specs costs $6,599, roughly the same amount. Both of these machines are used for desktop computing by professionals. We have our two devices for comparison.

There's a problem though: humans are bad at comparing large numbers. For example, the difference between $1,000 and $1 million sounds like a lot more than a difference between $1 million and $1 billion, despite the difference being the same: 1,000x. If we were to use the Apple II's raw transistor speed it would be around 3 billion. Comparing three billion to any larger number is too abstract. So instead, we'll use money per year. Many of us can understand our resources by calculating how much money we’ll earn in a given year. That influences our lifestyle and the choices that we make, much like transistor speed influences what choices we make with a given computer. We'll equate transistor speed per second to money per year.¹

Let's start with the Apple II and it's microprocessor, the MOS Technology 6502. What dollar amount should we choose? In the 1970s computing resources were constrained. We could do a lot, but we didn't have a lot of extra clock cycles to throw around and waste on things. An equivalent amount of money would be roughly $60,000 per year. One can do quite a lot with $60,000 per year, but there isn't a lot of room to throw around money, just enough to live a somewhat comfortable life. The MOS Technology 6502 contains roughly 3,500 transistors and the Apple II clocked the processor at 1.023 MHz. If we want that number to match out desired $60,000 number then we set the comparison of $1/year to 1 transistor at 59.675 KHz (which has an implied “per 1 second”).

If the Apple II is $60,000/year, then what is the Mac Studio? $11 quadrillion/year. In case that's too abstract, here it is written out fully:

$11,228,822,800,000,000/year.

With an M2 Ultra processor containing 192 billion transistors operating at 3.49 GHz, the Mac Studio has what is effectively infinite transistor speed resources.

Here's something to compare that $11 quadrillion number to: estimates for the total amount of money in the world range from the hundreds of trillions on the low end to a couple quadrillion on the high end. That means between the 1970s and the 2020s, we've gone from the computing power equivalent to the fiscal resources of a modest household to around 100 times more money than exists in the entire world. The Mac Studio has, comparatively at least, infinite resources. Meaning that, in reality reality, transistor speed isn't a useful property: it's become so unconstrained as to be useless.

That's not to say we are unconstrained, just that using transistor count or clock speed have become a useless metric. Not just because the numbers are large, but also because clock speed stopped increasing decades ago. Adding more transistors does help some things, but with more transistors comes more relative distance between them. This effectively forces hardware designers to turn processors from serial to parallel systems. Software written for serial systems does not run well on parallel system.

I hope with this equivalence, you'll be better able to understand just how much computing hardware has improved over the years. But even though hardware has improved, software has not. The gains of hardware have been spent creating modest gains in software. Quite literally, our hardware resources have increased a hundred billion times. Has our software, the foundation of which was built during the 1970s, advanced to take advantage of that hundred billion times increase? The answer is, quite obviously, no. Subsequent improvements will come, but will be wasted if we do not make the same foundational and architectural changes to software that were made to hardware.