Moore's Law is dead, but not for the reasons everyone says

2024-03-20

Here’s some strawman reasons that Moore’s Law is dead:

• Photoshop doesn’t edit twice as fast every 18 months
• GTA V framerate doesn’t double every 18 months
• Frequency doesn’t double every 18 months
• Hard drive capacity doesn’t double every 18 months
• My networth doesn’t double every 18 months

None of these are correct for the simple reason that Moore never made these claims. However, Moore’s Law is dead for completely different reasons that nobody mentions because they haven’t read the original paper. So let’s all do that, and keep in mind that Moore was writing before any human landed on the Moon.

https://hasler.ece.gatech.edu/Published_papers/Technology_overview/gordon_moore_1965_article.pdf

Here’s the key quote:

The complexity for minimum component costs has increased at a rate of roughly a factor of two per year

The version of the paper above has additional material from 40 years on, and clarifies the above:

So the original one was doubling every year in complexity now in 1975, I had to go back and revisit this… and I noticed we were losing one of the key factors that let us make this remarkable rate of progress… and it was one that was contributing about half of the advances were making.

So then I changed it to looking forward, we’d only be doubling every couple of years, and that was really the two predictions I made. Now the one that gets quoted is doubling every 18 months… I think it was Dave House, who used to work here at Intel, did that, he decided that the complexity was doubling every two years and the transistors were getting faster, that computer performance was going to double every 18 months… but that’s what got on Intel’s Website… and everything else. I never said 18 months that’s the way it often gets quoted.

Not only that, but Moore states that he was working with only a few datapoints at the time, and expected things to continue for a few more years. He never would have guessed back then that we’d be pushing it for decades.

So I looked at what we were doing in integrated circuits at that time, and we made a few circuits and gotten up to 30 circuits on the most complex chips that were out there in the laboratory, we were working on with about 60, and I looked and said gee in fact from the days of the original planar transistor, which was 1959, we had about doubled every year the amount of components we could put on a chip. So I took that first few points, up to 60 components on a chip in 1965 and blindly extrapolated for about 10 years and said okay, in 1975 we’ll have about 60 thousand components on a chip

The idea of 60,000 components on a chip probably sounded like huge progress at the time. That would drive a revolution, and it did. But keep going for another few decades? There was no expectation that it would, and he’d have been insane to suggest it. He was extrapolating from about 6 years of data, and it needed to be revised 10 years later.

Look carefully at that original claim: “The complexity for minimum component costs has increased at a rate of roughly a factor of two per year”. We’ll substitute in 18 months instead of a year. Now, it’s not that things will double in speed every 2 years. It’s not that frequency will double every 18 months. It’s that the cost per integrated component will be cut in half every 18 months.

Let’s do some extrapolation. The Intel 8008 chip was released in April 1972 with 3,500 transistors for $120 ($906 for inflation to 2024). There have been 622 months since then, which gives us 34.6 doublings. We would therefore expect a chip to have 3500 * 2^34.6 transistors, or about 90 trillion.

Absolutely nothing exists in that size. The largest currently released processor is the AMD Instinct MI300A with 0.146 trillion transistors.

Let’s be more generous and go with doubling every 2 years. That means there should be 26 doublings, for 0.234 trillion transistors. The MI300A was released in Dec 2023, so Moore’s Law is in the right range if we give it a 2 year doubling period, right?

Wait, can you buy an MI300A for $906 (the inflation adjusted price of the Intel 8008)? No, not even close. It’s in the “if you have to ask, you can’t afford it” range. Reports put it around $10k to $15k each.

So Moore’s Law is dead because you can’t buy a 200B transistor device for around $1000. Not even close.

All that said, Moore’s original paper is quite remarkable. He made an extrapolation that held on for a decade more, and then held on for a few more decades after a little revision. He’s also looking ahead to how this would affect everything from radar to putting computers in the home and cars. It was a remarkable prognostication.