Aardvark Daily

Please visit the sponsor!

Reviving old skills

There is no doubt about it, computer prices are about to go through the roof.

These huge price rises are being driven by the unprecedented demand for RAM and GPU silicon created by the AI industry.

The end result of this is may be a renaissance in the ancient skill of code-optimization.

Both the size and speed of almost any piece of software can be significantly improved by careful optimization and it will be very interesting to see if new value is placed on this skill.

Back in the days when I was writing full-featured accounting software that had to run under the paltry memory and CPU limitations of 8-bit hardware, a good 30 percent of my programming time was dedicated to making the code ever-smaller and ever-faster.

In those days, an 8-bit processor like the Zilog Z80 could only directly address a maximum of 64Kbytes (yes, that's kilobytes, not megabytes or gigabytes) of RAM and generally speaking only a maximum of 48K was actually usable because the remaining 16K was allocated to the BIOS and memory-mapped video display.

Even the processor itself was usually only clocked at leisurely 2MHz so every cycle was precious and not to be wasted on inefficient code.

Despite these hurdles, we managed to create some amazing software that still ran so fast that the operator was usually the slowest component of the entire system. Rarely would you have to "wait" for the computer and most of the time it waited for you.

Of course there was no facy GUI and a mouse was an unheard-of peripheral but those systems still worked amazingly well despite having orders of magnitude less power and memory.

These days however, there's nowhere near the focus on speed and efficiency of code. Why bother wasting valuable programmer time trying to squeeze that last bit of performance out of your code when you can simply specify that the minimum system requirements are a few more GB of RAM or a processor with more cores?

However, we are now about to hit a wall. DRAM prices mean that any software which relies on inefficient code which demands the user install extra memory will be at a disadvantage in the marketplace. Likewise, systems that rely on raw GPU grunt to deliver acceptable performance will not be as attractive to those who can't afford a king's ransom to buy a new card.

Contributing to today's problems will be AI-generated code produced by those who are simply vibe-coding applications. These are unlikely to be well-optimized in the same way that code cut by an experienced and competent human programmer could be.

Yes, you could try to get AI to optimize your code for you but I wouldn't bet that the results will be what you're looking for. Once programs get to a certain size, AI tends to end up wrecking things in its attempts to make changes -- just look at the most recent updates to Windows for proof of that.

I'm now wondering whether any programmer who wishes to remain employable in the AI-era ought not specialise in code-optimization. Those who can reduce the RAM and GPU requirements of a package through careful code auditing and revision may well be in high demand -- at least until hardware prices return to some semblance of normality.

If I'm honest, optimizing code is one of the most fun things to do as a programmer -- or at least that's my experience. Trying to squash a program by just a hundred bytes or so in order that you don't have to use overlays was a fascinating challenge in the days of CP/M computer systems.

Some people bought crossword or jigsaw puzzles for a fun pastime, I optimized code, often even dipping into assembly to save just a handful of bytes or to improve performance by a few percent!

Ah, happy days.

Carpe Diem folks!

Please visit the sponsor!

Here is a PERMANENT link to this column

Rank This Aardvark Page

Change Font