How Many ZX Spectrums Does It Take to Draw a Cat?

An AI image of two cats took 30 seconds to generate. On a ZX Spectrum it would have taken millions of machines working together, and on a ZX81… well, you’d still be waiting. A light-hearted look at how far computers have come in just a few decades. The post How Many ZX Spectrums Does It Take to Draw a Cat? first appeared on Mad Black Cat.

For more updates, visit Mark Nolan's Podcast!

Episode Transcript

Speaker 1 (00:00):
How many ZX spectrums does it take to draw a cat?
The answer, as it turns out, is millions, because when
you ask modern AI to generate an image at say,
two cartoon cats in a living room, it takes about
twenty to thirty seconds. On a ZX spectrum from the
early eighties, you'd still be waiting and probably listening to

(00:20):
a tape deck screeching at you. Let's put it into context.
The Sinclair z X eighty one, launched in nineteen eighty one,
had just one kilobyte of RAM. That's smaller than the
text of this podcast script. Even expanded to sixteen kilobytes,
it struggled to print out a stickman. Asking it to
generate a high resolution cartoon image would take decades if

(00:43):
it was even possible. Then came the ZX spectrum sixteen kilobytes,
or if you splashed out forty eight kilobytes color graphics
enough to bring us classics like jet set Willie and
Mannic Minor and who can forget Dizzy the Little Egg.
But in today's terms, it's tiny. Generating one AI cat

(01:06):
picture would require millions of spectrums chained together. Now compare
that to your smartphone sitting in your pocket. With thousands
of times the power of those early machines, and the
AI used doesn't just run on one device, It runs
on vast networks of servers, the kind of thing only
governments or research labs could have imagined back then. So

(01:28):
next time you sigh because an AI image takes twenty
seconds to appear, think of it this way. Once upon
a time, we waited twenty minutes for a cassette game
to load, only to see the words are tape loading error.
By comparison, today's AI isn't frustrating, it's astonishing, by the

(01:50):
way in our example we've spoken about creating a single image.
In reality, AI tools rarely work one at a time.
When you ask for an image, the system used generate
several versions in parallel, often four, sometimes more, before you
even see the result. That means every thirty second wait
actually hides the effort of creating multiple complete pictures at once,

(02:13):
discarding most and only showing you the best. It is
like asking an artist to sketch four paintings in half
a minute and then letting you choose your favorite. When
I was first given a ZX spectrum as a young teen,
I could not have imaged for a minute, telling my
future self that I would one day create this article

(02:35):
on a mobile phone tablet sat next to me, before
starting my chromebook to actually do the publishing, all tools
than me on my z X spectrum could never have
imagined possible, even though I was able to imagine that
we would be flying around in Eagle transporters by nineteen
ninety nine. Computers may test our patients now and then,

(02:55):
but only because they've spoiled us with miracles