AMD says transistor tech will keep Moore’s law alive for 6 to 8 years Chipmaker AMD has hinted that new transistor technology will keep Moore's Law alive for the next six to eight years, but as one might guess, it will cost more. Meanwhile, the company still plans to market new chips based on its Zen 4 architecture next year, including Bergamo, which is intended to compete against Arm-based …
> "We could have made that. We had a design approach that was going to make the custom Arm design for AMD equally performant to the x86, but the ecosystem wasn't there. So we kept our focus on x86, and we said, let's watch the space in Arm,"
I think in truth they were out of money and had to streamline on x86 to avoid going bankrupt, but it was exactly the right thing to do.
"Moore's law died when frequencies stopped increasing past 4 GHz."
"Moore's Law" never really existed. A passing comment in a paper got magnified into a "rule" by others, but not by Moore. Calling a it a "law" implies that it must be complied with or is fundamental (like a "law of physics"), whereas it was no more than an observation of a trend that happened to persist (very approximately) for some time. There's absolutely no reason why that trend should persist in perpetuity, so it not doing so is no surprise.
Moore's law simply mentions transistor density, not frequency. And anyway, frequency scaling achieved by ever-longer pipelines (I am looking at you NetBurst) proved to be a dumb and inefficient way of increasing performance, it's why the Athlons and later the Core 2 Duos wiped the floor with the NetBurst CPUs. The thing everyone misses from Moore's law is power scaling (aka Dennard scaling). The number of transistors still doubles roughly every 24 months, but the power consumption per transistor doesn't halve every 24 months anymore. Which is a problem because, with the exception of smartphone and tablet chips, die size has not been a problem for a while, power consumption is. We have reached a stage where GPU chips are getting smaller with time but their power consumption goes up.
>Moore's law simply mentions transistor density, not frequency.
Precisely, Moore's law noted that wafer size increased and feature size decreased linearly with cost - while squaring the number of transistors per wafer. It's not clear that this was true in the EUV era, we had to make features smaller for power and to fit into a phone - but I'm not convinced that 3nm is cheaper "per transistor" than the previous generation
Well, AMD always touted how their chiplet design was more cost effective so. I have no inside data to prove either way but, if you believe AMD's past propaganda, it's not. There was 2 ways this release could of read: 1. AMD is letting everyone know they're raising prices. 2. AMD is experiencing exponentiall benefits from the design choices they made several years ago. It looks like they're focusing on #1 and pretending to forget about #2 (for now).
The latest AMD products seem to be doing well based on the fact that Intel can't or won't deliver on their promises about Rapids (or Max or whatever it's called). I suspect if they do deliver AMD will be talking about price "reductions" or "scale backs" or wtf ever.
AMD is the new Intel.
Somehow, I don't think I will be in that much of a rush to upgrade from a 5950X for some time. Maybe two, if not three whole CPU models. I really didn't need that CPU either to be brutally honest; it was a lockdown boredom splurge more than anything that prompted picking one up.
Software of interest is harder than ever to come by despite all the power on tap.
...considering that Moore's original observation referred to lowering the cost of transistors. Here's a quote from his 1965 paper (see e.g. https://montcs.bloomu.edu/~bobmon/PDFs/Gordon%20Moore.Law.pdf)
The complexity for minimum component costs has in-
creased at a rate of roughly a factor of two per year (see
graph on next page). Certainly over the short term this rate
can be expected to continue, if not to increase. Over the
longer term, the rate of increase is a bit more uncertain, al-
though there is no reason to believe it will not remain nearly
constant for at least 10 years. That means by 1975, the num-
ber of components per integrated circuit for minimum cost
will be 65,000.
I believe that such a large circuit can be built on a single
wafer.
The Register 
Biting the hand that feeds IT
Copyright. All rights reserved © 1998–2024
