Ask HN: Does Moore's Law apply to RAM and if so, is it finally ending? | Hacker News Reader

Ask HN: Does Moore's Law apply to RAM and if so, is it finally ending?

5 pointsLeoPanthera11 comments

I noticed recently that if you start with 32K in 1980, a reasonable amount of RAM for a personal computer at the time, and then double it every 2 years, you get a surprisingly accurate track of the "normal" amount of RAM for a personal computer in that year.

1M in 1990. 32M in 2000. 1G in 2010.

And... 32G in 2020... which while not unreasonable does suddenly seem quite high. And in 2022, you get to 64G, and I don't see many home computers with that much RAM.

So why have we suddenly slowed? Do we have "enough" now? And it just took 40 years to get there?

View on HN

There is also the rise of the cloud, aka someone else's computer, to consider. There was a time when a power user needed a power machine to do whatever it was that they did. When you can spin up a 400 GB RAM beast for a few bucks for the hour it takes you to run your model/do your analysis/render your scene, maybe spending money for local hardware goodies isnt the best choice.

I feel profoundly conflicted about the swing from mainframe to pc to cloud and now back to edge again.

Yes, Moore's law has nothing to do with computing power, memory, or any other specific chip. Moore's law only states that the number of transistors on a dense integrated circuit will double every two years. The number of transistors doubling then roughly equates to more computing power, memory, etc.

We have slowed for 2 reasons, most people don't need more and power consumption has become as important if not more important when it comes to mobile devices and laptops.

In general Moore's law is anticipated to start slowing because as you get smaller and smaller quantum physics comes more and more into play.

If you look at historical data for CPUs, RAM, and storage, all three have pretty much followed the same curve, modulo a delta for either the cost per unit or units at a given price.

RAM itself is transitors, so the same factors governing CPU production (mask and die size) apply. For storage, the situation is somewhat different in the case of tape and spinning rust, though for SSD again it's basically transistors.

Keep in mind that Moore's Law is less about the physics of the Universe than an agreement between those developing and producing semiconductors and their market. Eventually physics will intrude, but to that point it's largely a matter of markets, finance, and investments.

DRAM is mostly capacitors which have their own challenges and scaling limits.

Isn't there one transistor per capacitor? Or is your point that the capacitors are harder to create in semiconductors than transistors are, and so the difficulty of creating capacitors dominates the process of creating RAM?

The latter.

Moore's Law is more about what's available than what people choose to buy. PCs have been over-serving people's needs for years which is why many people are spending less and less money on PCs. This was predicted by Jon Forrest around 1995; it took longer than he predicted to happen but it did happen. https://jlforrest.wordpress.com/2015/12/18/the-forrest-curve...

DRAM technology is also stagnating; the industry has been on 8Gb for a long time.

It does seem like we have enough RAM to me. There's always a bottleneck to whatever you do on a computer, but it keeps shifting. I think the bottleneck nowadays is often latency rather than throughout/capacity, which sort of translates into latency or software that can utilize all the hardware capacity effectively.

Gamers are great at finding bottlenecks. They are now more interested in RAM latencies rather than amount of RAM.

I'm not sure about that timeline. 1GB of ram was already pretty common for workstations in 2000. My PC in 2001 had 1GB ram. The Sun SPARCstation 20 was released in '94 and supported up to 512MB. By '97 Sun's enterprise servers like the Sun Ultra Enterprise 10000 were topping out at 64GB.

For a laptop it takes power to keep memory powered up in sleep mode. For a 32GB laptop it is bad enough.

DRAM has hit a brick wall, look at how fragile the data stored in it is when RowHammer and a number of other things that should not ever work, are reliably exploitable.

Reply on news.ycombinator.com