Slashdot Mirror
Why Intel Insists Rumors Of The Demise Of Moore's Law Are Greatly Exaggerated (fastcompany.com)
From an article on FastCompany: Intel hasn't lost its zeal for big leaps in computing, even as it changes the way it introduces new chips, and branches beyond the PC processor into other areas like computer vision and the internet of things. "Number one, too many people have been writing about the end of Moore's law, and we have to correct that misimpression," Mark Bohr, Intel's technology and manufacturing group senior fellow and director of process architecture and integration, says in an interview. "And number two, Intel has developed some pretty compelling technologies ... that not only prove that Moore's law is still alive, but that it's going to continue to provide the best benefits of density, cost performance, and power." But while Moore's law soldiers on, it's no longer associated with the types of performance gains Intel was making 10 to 20 years ago. The practical benefits of Moore's law are not what they used to be. [...] For each new generation of microprocessor, Intel used to adhere to a two-step cycle, called the "tick-tock." The "tick" is where Moore's law takes effect, using a new manufacturing process to shrink the size of each transistor and pack more of them onto a chip. The subsequent "tock" introduces a new microarchitecture, which yields further performance improvements by optimizing how the chip carries out instructions. Intel would typically go through this cycle once every two years. But in recent years, shrinking the size of transistors has become more challenging, and in 2016, Intel made a major change. The latest 14 nm process added a third "optimization" step after the architectural change, with modest performance improvements and new features such as 4K HDR video support. And in January, Intel said it would add a fourth optimization step, stretching the cycle out even further. The move to a 10 nm process won't happen until the second half of 2017, three years after the last "tick," and Intel expects the new four-step process to repeat itself. This "hyper scaling" allows computing power to continue to increase while needing fewer changes in the manufacturing process. If you divide the number of transistors in Intel's current tick by the surface area of two common logic cells, the rate of improvement still equals out to more than double every two years, keeping Moore's law on track. "Yes, they've taken longer, but we've taken bigger steps," Bohr said during his three-hour presentation.
Seems the size of the paragraph doubles every 2 stories.
Table-ized A.I.
More like Moore's Observation. This is not a law, and we really need to be careful in our terminology.
Moore's law is an empirical law in the sense of Ohm's law, which most familiar materials obey but some do not.
Lack of competition is what is slowing processor speed growth not difficulty. One or two major competitors just don't provide enough competition so there is no reason for them to innovate. They hold onto new technologies for years and years until AMD makes an announcement then all of a sudden intel reduces prices or make an announcement. Innovation comes from competition not from patents that seem to last way too long or force expensive licensing and large corporations gobbling up any business they even has a spark of creativity.
"This "hyper scaling" allows computing power to continue to increase while needing fewer changes in the manufacturing process."
This "hyper scaling" allows Intel to continue to milk customers who expect more than modest gains with every generation.
FTFY
Moore's law is in fact alive, but Intel added ME to its CPUs. That means a second processor in the same box as the first, only it's for the government to spy on you, not for you to use. That's a fixed overhead that was most noticeable when it was first implemented several years back - when CPUs seemed to get slower - they've once again started to catch up to where they were. If they replaced the ME architecture with more cores you would definitely see a ~2x boost in performance, but of course that won't happen so they'll just keep competing with each other over bus constraints (network, HDD, peripherals, etc) and you won't notice the speed enhancements as much.
Number of Transistors may still be increasing in density but computers aren't seeing the revolutionary jumps in power and performance- it's not scaling to us end users. I have a 5 year old PC at home I built, and it rivals most of the mainstream PCs being put out today. Even if Moore's law is still holding true, it's not really relevant anymore.
Computers aren't getting much faster any more. Processors may be getting smaller as transistors density gets higher, but your average home PC isn't getting much better.
"That's the way to do it" - Punch
I made sure I got a previous gen processor in my latest laptop so I could run Windows 7. If Intel wants new processor sales add explicit 7 support.
Is it just me or does this whole diatribe just ooze "pathetic marketing maneuver"?
It's one thing to admit that things are getting more challenging cause the low-hanging fruit is gone and Intel's having to put more time and effort into their manufacturing, but for the love of Pete, redefining Moore's Law is just lame.
I really wish Apple had a tightly held patent on their reality distortion field cause now everyone else is trying to use it and it's just... cringeworthy.
when they decided they hate those of us that use laptops when they hatefully decided to limit the amount of memory supported by mobile and have refused to increase it for five years and counting. It is hate that drove that decision. Now, Apple laptops are garbage since the max amount of RAM hasn't increased in five years so their laptops are useless for professional work.
They claim that emperor still has clothes...
Has parallel processing gone about as far as it can go due to difficulty in programming for it?
I originally read the title as "Why The Rumors Of Demi Moore's Demise Are Greatly Exaggerated".
Moorephy's Law: "If the processing power of a CPU can double every 2 years, it WILL double, and in the worst way possible. You will have a plurality of CPU cores that each want to do their own thing. And your compiler will not be able to get those cores to work with each other properly. If you code in Assembly, of course, things are very different. Your CPU cores WILL eventually learn to talk to each other, but by the time that happens in any meaningful way, you will unfortunately be a patient living in a psychiatric institution." (Moorephy's Law is licensed under the GNU Multi-Core Assembly Programming Brainfuck 3.0 license).
Why did the chicken cross the road? Because Elon Musk put an AI chip in its head.
This is a remake of a 1999 movie. Plot summary: AMD had a chronically weak offering, Intel was in the habit of dribbling out the performance gains. AMD suddenly came on very strong with Athlon, a completely new chip which was arguably faster than Intel and definitely cheaper. Almost overnight, Intel suddenly figured out how to make much faster chips, and so did AMD. Performance doubled, tripled, with AMD being the first to crack the 1GHz barrier the next year. That spiral continued for a few years and the users were happy, but AMD ultimately fell behind and Intel went back to their old tick-tock.
intel needs to up the number of pci-e lanes.
AMD has more.
We haven't had any noticeable gains in computing for a long while. Other than SSD, nothing got notably faster or bigger. What is not clear to me if we hit diminishing returns or lack of competition allowed market leaders to sleep on laurels.
If this is diminishing return on hardware - then next area is software optimization. So far, most of our progress was carried by hardware. This is not going to be popular view among programmers, but default mode of operation is "how much resources do I have, lets use it all" in software engineering. There is no thought given to making it leaner and more efficient, because it used to be that hardware gains over time would make such effort moot. Well, there might not be any more notable gains. We will hit the next nm fab level, get 3D layout process in place and not have a good way to move forward other than occasional specialized optimizations (e.g. AES) acceleration). This might take a form of optional co-processors.
I generally feel that Moore's Law is still well in affect. While Moore's law is just about number of transistor's will double ever 18 months, I take a broader look at this. I see it as processing power will double ever 18 months.
How do I define 'processing power'? That's up to a lot aspects. Obviously, A CPU's is raw CPU clock speed, bogoMIPs, etc. But there are other ways we can game power/efficiency too. Multiple cores and SMT. But also power efficiency should be considered, if we can do more processing power in the same amount or less of power we've gain processing power. I also consider cost too. If I can purchase two CPUs of the same speed as I did 18 months ago for the same price, I still consider Moore's law working.
No good deed goes unpunished.
However, what matters to me is that a 10-year-old desktop computer is not very far behind, performance-wise, a 2017 desktop computer.
Someone can post with detailed calculation, but I'm still retry certain Moore's law will end before the year 3000.
If memory serves, even if we turned the entire volume of the observable universe into a single computer, with a "transistor equivalent" device in every Planck volume, it still wouldn't be enough for Moore's law to apply after about the year 2600.
Such is the nature of exponential growth: it has to stop at some point.
I wonder if PC making is entering the same regime as the aircraft industry has been in for quite some time? Fifty years ago, one could do London to New York as fast as it can be done today - faster, for Concorde was debuting. Today, we have the so-called Dreamliner, an airplane that is said to be revolutionary. What do we, travelers, get to see? Well, its windows are very small, rather than teeny-weeny; the air pressure in the cabin is slightly higher than before, but still not quite sea level; and the humidity is higher than before, but still very dry. That's it. That's the revolutionary Dreamliner. Oh, and tickets are not any cheaper than they were 15 or 20 years ago.
Moore's law has predicted future exponential growth of integrated circuit density fairly well since it was proposed over four decades ago.
It was wrongly stated anyway. Should have said by a golden ratio every year instead of doubles every 18 months (which is only a 4% difference). And it's much more believable that it would grow proportionate to Fibonacci numbers every year (parallel to release cycles) because the rate of tech progress is a recurrence relation to the level of existing tech.
Any guest worker system is indistinguishable from indentured servitude.
Wake me up when they manage to invent a time machine and deliver 10nm on time to keep up with moores law.
Until then I will continue saying that it's dead.
Why Intel Insists Rumors Of The Demise Of Moore's Law Are Greatly Exaggerated
Because Intel wants you to buy their CPUs which haven't seen worthwhile improvements in over 5 years.
Unless you need bleeding edge performance, just pay half the money and get a Ryzen CPU.
Have people forgotten that Intel promised us 10 GHZ chips by 2005? http://www.anandtech.com/show/680/6 . I think that Intel can get down to maybe 7nm but after that I believe we will have reached the end of what silicon can do. Intel and the rest of the industry dont want to switch away from silicon because they dont want to reinvest in machining, money, or more research for a new material (cant blame them) but they cant bury their heads in the sand too much longer.
I believe that towards the beginning of the 2020's we will need a new material if we want to reduce power or increase performance anymore. But eventually there will be a limit on how small we can make something regardless of the material. The 2030's will be a very interesting time. :)
Most of this is because of the proliferation of Virtualization and "Cloud computing"
If most of what you do is in a web app or a virtual machine or compatibility layer.
Your never going to see the hardware gains.
Intel processors may be 10X faster than they were 10 years ago but your ISP damn sure isn't.
When basically all programs have to phone home before they do anything, hardware improvements become irrelevant.
https://theinfosphere.org/Speed_of_light
Well of course the traditional numbers seem insurmountable by our efforts & inventions. That's why we change the numbers to allow our stuff to be worth anything!
"The number of people predicting the end
of Moore’s Law doubles every two years."
-Tryggve Fossum, Intel Fellow
No, If you want bleeding edge performance find a Ryzen cpu/mobo/memory combo that lets you boost the memory to 3600 and CPU to 4.1Ghz and remain stable. At which point you will be equal or better in performance to the 7700k @ 5Ghz(Interestingly, memory speed increases beyond 2400 have very little effect on Kaby Lake performance) in games and better than everything else in the other metrics.
The 7700k isn't the performance king. Some "Extreme" series (6870x or whatever) is. The 7700k is the popular gaming CPU.
Most Ryzen chips can't sustain 5.1 GHz on air. The highspeed memory situation is getting better, but it's still a crapshoot in many cases.
Another issue is that as you crank up the memory speed, increasing the speed of the "infinity fabric" connecting the CCX units, you have more power draw and more heat to deal with, which can actually hurt performance in certain workloads.
For anything heavily threaded or specifically coded with Ryzen in mind, Ryzen curb stomps Intel's shit. There are still scheduler issues to contend with as well on the Windows side (Windows 10 can't really properly distinguish between cores, virtual cores from SMT, and cores on separate CCX units which would require talking across the infinity fabric).
We're already seeing benchmarks and utilities and a few games get updates for Ryzen which lead to 10-20% increases in performance. Another big issue for games is the Nvidia driver. It recently came out that running "CPU" style gaming benchmarks on an Nvidia GPU fucks Ryzen over. Running with an AMD GPU (no Nvidia drivers) AMD gets a huge performance boost. I wonder if anyone will try to replicate this with older Nvidia drivers to try to pinpoint a time when this started happening.
I remember Intel trying to tell us that you improve the efficiency of a computer my making it run harder to complete tasks faster. They love to rewrite logic using words that happen to sound like there filling some sort of gap. Truth is Moores law has only been about increasing transister density, nothing else. Optimizing architectural changes has nothing to do with it, Intelies again.