Working Up

by Dwayne Phillips

The most important computer processor ever made was the Intel 8087 (well, a little exaggeration). Spend more money on hardware or more money on smarts? The decision is still with us.

It was 1980. Intel had just released the 8086 CPU—a computer on a chip that worked on 16 bits at a time. WOW! Great for powerful computing, so long as you only used integers (fixed point as we called it). Want to use real numbers? (floating point) Well, buy the 8087 co-processor. It only cost a couple hundred more dollars. Uh, well that increased the cost of the computer 10% or 20% or something.

Ten years later, I was writing software in the C programming language to process images. It was the kind of things that Photoshop did. There were several companies producing co-processing boards to pop into the IBM PC and the clones of the time. The co-processing boards cost several thousand dollars, but if you wanted to really do amazing things, you spent the money.

Well, I didn’t have the money and I wanted to really do amazing things. Solution? I figured real hard and used integers to do everything. I took great pride in telling people that my software removed the need for these extra processing boards. And this all worked. It actually did.

Spend money on hardware? Spend money on brain power? The question never leaves us. We are still trying to decide if we use super-large language models or think a while and use smaller language models. Do we have the hardware to do what we want? Can we afford that? Maybe we should just think more. Same questions.

The answer? Oh, I am sure someone has modeled this question and can show curves of hardware and time and brainpower and all that. They have the universal answer. Did they run simulations on state-of-the-art hardware or think about it a lot? Oh, same questions.