Hacker News Reader

Comment by bb88 | Hacker News Reader

On the github page, he writes this:

> Normally for multiple teams with heavy traffics, Wey should not have any significant CPU usage, and RAM ussage is usually under 100MB. However if you have a team with more than 10k users in it, the memory usage may increase a lot.

Bloaty technologies was what made MS great in the 80's/90's, because 18 months later, the speed of the CPU would double.

I feel like we should think hard about optimizing for memory/cpu again.

View on HN

Now that Moore's Law is dead, that time is coming. We can't afford to be wasteful. Not that we ever should have, but we've spent over a decade now catering to "developer productivity", now it's time to put the user role back in the forefront.

The last time I checked (roughly a year ago) Moore's law was still holding.

Not sure why the downvotes. Moores law[1] refers to the number of transistors that are on the die. With the decreasing size of the transistors, this has remained true. Despite that, clock speeds have stopped growing thereby leading to serial performance not increasing as fast as it had in the past.

[1] https://en.wikipedia.org/wiki/Moore%27s_law

GPUs were stuck on 28nm for over 5 years before 16nm one launched. Intel has been stuck on 14nm for three and a half years and counting. This has certainly not been anything close to doubling every 18 months.

To repeat: Moore’s law is about transistor count, not size. From the Wikipedia article linked in the comment above yours: “Moore's law is the observation that the number of transistors in a dense integrated circuit doubles approximately every two years.”

Transistors can stay the exact same size for twenty years as long as the die size grows, and Moore’s law still holds.

But the die size didn't grow at nearly that rate, large dies remain very expensive to make. For example, comparing the HD7970 and RX 390X, similarly positioned GPUs released 4 years apart in 2011 and 2015, transistor count grew ~50% from 4.3 billion to 6.2 billion.

https://www.anandtech.com/show/5261/amd-radeon-hd-7970-revie...

https://www.anandtech.com/show/9387/amd-radeon-300-series/3

> But the die size didn't grow at nearly that rate, large dies remain very expensive to make.

But again, did transistors double in count on any processor, not just GPUs? If so, Moore held. Yes, according to this chart:

https://en.m.wikipedia.org/wiki/Moore's_law#/media/File%3AMo...

I mean if you want to go with the highest transistor count chip currently in existence, that would obviously be the 512Gb flash chips with ~170 billion transistors. Or if you want to constrain it to 'processors' that would be the 30 billion transistor Altera FPGAs. Which have been available for 3 years. No CPU yet has exceeded the 10 billion transistor count of the Sparc M7, also released 3 years ago. The Tesla P100, two years old, retains the highest GPU transistor count at 15 billion.

It's meaningless to compare these totally different types of chips that end up with extremely different transistor densities on the same node due to the wiring requirements.

I'm just the messenger. Apparently there is some sort of criteria for what a "dense integrated circuit" is for the chart previously pasted, and apparently all of the listed processors passed the required criteria for charting the law through time. What that criteria is, I couldn't tell you. Couldn't find it on the Wikipedia page. "Moore's law is the observation that the number of transistors in a dense integrated circuit doubles approximately every two years." (Wikipedia)

Yes. I read the wikipedia quote the first time you posted it. No need to say the exact same thing 5 times. The chart is clearly restricted to CPUs and ends in 2016. It's 2018 and the top chip on that chart still has the highest transistor count of any CPU on the market. Absolutely zero increase in transistor count for the type of chip that would go on that chart. Every category is seeing the exact same stagnation, Moore's law has been nowhere close to being matched in the past few years. Every chipmaker is struggling with their new nodes and falling way behind schedule on improvements. High-volume chips that normal people buy are really a far more realistic and useful barometer for Moore's law, and again you see the stagnation. Skylake and Kaby Lake are near-identical chips. Coffee Lake is ~25% bigger. There is no doubling of transistor counts going on in anything resembling an 18 month time frame.

Okay, so the comment is then that you believe Moore's law no longer holds because for the past two years transistor counts on CPUs have not increased. On that basis, I might agree. Not (as you were stating earlier) about the sizes of the transistors themselves, or about GPUs from 2011 to 2015, or about 512Gb flash chips. None of that applied to this conversation, hence the desire to repeat myself. Thank you for finally clearing that up. Now that we're finally on the same page, I might agree.

Are there no CPUs with higher transistor counts out there? Maybe ones you are not aware of yet?

Please show me this CPU with a higher transistor count. There is a maximum die size that each fab is able to make, mostly related to the optics used for their photolithography process. Most designs are nowhere near the limit because larger sizes are uneconomical to produce, though NVIDIA for example is on the record stating that TSMC is physically incapable of producing anything larger than their current GPUs.

> Please show me this CPU with a higher transistor count.

That was my question to you. You claimed, "It's 2018 and the top chip on that chart still has the highest transistor count of any CPU on the market." Which I take to mean that you don't know of any CPU with any higher transistor counts, which is an argument from silence. There could be a CPU outside of your knowledge.

So, we wait to see. Moore may be dead but "moore" time may be needed to hold the funeral.

I agree with both of you that transistor count per mm² is stagnating in the last years. However that has happened before. For example, 1982-1987 and 2007-2010 according to Wikipedia data.

It is too early to declare Moore's Law dead.

> Not that we ever should have

You say that like people were making these bloaty apps for the good of humanity—trying to make them objectively perfect, but falling short. While FOSS is done with that mindset, FOSS isn’t usually where bloat comes from.

Bad engineering is usually the result of overriding business concerns: being first to market, iterating quicker than your competitors, having more reach on more platforms with less expense, etc. More features before feature polish, and user-visible polish before good architecture.

Even without Moore’s law, this incentive is still in place. It not just has a slight countervailing force of nobody wanting to run apps that take too many resources.

What that means in practice is that apps will still be as bloated as possible (or rather, with as little optimization done as possible) while still managing to not choke the machines they’re on.

Moore's law is alive and well. If it is slowing at all it is evidently due to monopoly/lack of competition more than anything else. It feels like Moore's law is dead because there actually were a few more laws at play which have died. We went from exponential-exponential to just exponential, which feels like going from exponential to linear from our perspective, but is not the death of Moore's law.

Can't make transistors smaller? just add more! Speed of light becoming a problem? Stack em! Speed of light and heat becoming a problem again? Stack cores, improve multi-core architecture in both hardware/software and enjoy another many decades of Moore's law. It won't be dieing any time soon

If it’s slowing? Intel says it is. And Moore said dying in next decade.

Wikipedia:

Moore's law is an observation and projection of an historical trend and not a physical or natural law. Although the rate held steady from 1975 until around 2012, the rate was faster during the first decade. In general, it is not logically sound to extrapolate from the historical growth rate into the indefinite future. For example, the 2010 update to the International Technology Roadmap for Semiconductors, predicted that growth would slow around 2013, and in 2015 Gordon Moore foresaw that the rate of progress would reach saturation: "I see Moore's law dying here in the next decade or so."

Intel stated in 2015 that the pace of advancement has slowed, starting at the 22 nm feature width around 2012, and continuing at 14 nm. Brian Krzanich, CEO of Intel, announced, "Our cadence today is closer to two and a half years than two." Intel is expected to reach the 10 nm node in 2018, a three-year cadence. He cited Moore's 1975 revision as a precedent for the current deceleration, which results from technical challenges and is "a natural part of the history of Moore's law."

More background in recent article here on 40 years of processor performance:

https://www.eejournal.com/article/fifty-or-sixty-years-of-pr...

HN: https://news.ycombinator.com/item?id=16750733

The price of RAM has actually gone up in the past 5 years, so that's not a terrible idea.

How much RAM does wee-slack use? https://github.com/wee-slack/wee-slack

thanks for linking this, I was looking for a way to intergrate slack into weechat without using the irc gateway since they are dropping support for that

Reply on news.ycombinator.com