Slashdot Mirror
Moore's Law Is Becoming Irrelevant, Says ARM's Boss
holy_calamity writes "PCs will inevitably shift over to ARM-based chips because efficiency now matters more than gains in raw performance, the CEO of chip designer ARM tells MIT Technology Review. He also says the increasing adoption of ARM-based suppliers is good for innovation (and for prices) because it spurs a competitive environment. 'There’s been a lot more innovation in the world of mobile phones over the last 15-20 years than there has been in the world of PCs.'"
...and by this, Intel's fab advantage will eventually make ARM irrelevant.
CEO of a company that makes more efficient CPUs than the competition says the future is in efficient CPUs. News at 11.
Guy who works for company thinks said company's product will dominate product of other companies: Film at 11.
Then GM and Ford would come up with some witty comeback about crashing several times a day.
Bill Gates's friend Jerry may have something to say.
But every newer version of operating systems has more bloat than ever. There must be some corollary to Moore's Law which states successive Operating Systems will still require higher performance, but users will now become accustomed to slower response times.
We could call it the Blort Law.
A feeling of having made the same mistake before: Deja Foobar
The guy says nothing of the sort, it's just the title of the article. All he says is that efficiency is becoming more and more important, and that ARM offers such efficiency. (He *also* says that ARM can offer performance as well.)
If this guy gets his way, then we may never have sentient computers.
At least keep Moore's Law alive through 2025.
-John Henry
" efficiency now matters more than gains in raw performance"
Sure, so why don't you start off by telling us why an Exynos Cortex A-15 chip running a web benchmark is using about 8 watts of power, with the display turned off so only SoC power is being measured, while Intel has already demoed a full-blown Haswell running Unigine Heaven at... 8 watts.
So when the miraculous Cortex A-15 uses the same amount of power as the supposedly "bloated" x86 Haswell, while Haswell is running a benchmark that is massively more intensive than a web-browser test, who is really making the most "efficient" platform?
Exynos Source: http://www.anandtech.com/show/6422/samsung-chromebook-xe303-review-testing-arms-cortex-a15/7
Haswell Demo Video: http://www.youtube.com/watch?v=cKvVdhkgAxg
AntiFA: An abbreviation for Anti First Amendment.
As a geek I love a powerful general purpose machine that can do all the things an ebook reader/music player/web browser can do AND a whole lot more like play 3d games, run a math or science simulation, allow you to record and edit video, memory and processor intensive image editing. To me a tablet is little more than a crippled PC with the keyboard removed (fantastic, why did I learn to type at 90wpm again??), and a smudge screen interface (hate viewing photos through finger marks!!!). It's really awesome that we have dumbed down our computers to the point of mediocrity. Even finding a decent e-book reading or music playing app - the things these pieces of shit are touted at making easier - is a nightmare. So many book readers don't even let you zoom on images. And browsing the web without flash support is like trying to surf with one leg. I don't mind that there are dumbed down idiot boxes for those who like to post pictures of food on Facebook, but I really resent the impact on general purpose computing.
Moore's law just predicts transistor density - it says absolutely nothing about computational power. Increases in transistor density can make electronics more efficient per watt, but this still is aligned with Moore's law.
The title is stupid, and the actual article says almost nothing like it.
It's called Gate's Law: Every 18 months, the speed of software halves.
Sure efficiency matters, but only in portable devices. Desktops or other computers connected to the mains don't have a problem.
Hey its winter already, a watt used by your CPU is a watt less that has to be used by your radiant or convective heater.
Moore's law is about the number of transistors on a chip. It has never been about performance.
Sigh. It seems there is a new, hip, propaganda trend on Slashdot: pro-ARM articles are posted, and a bunch of ARM zombies come out saying how anything ARM makes will (magically) be lower-power or more power-efficient than anything x86.
So I'll start a tradition of posting this same response every time (originally posted by me here):
"ARM isn't magic; there is nothing in the ARM ISA that makes it inherently lower power than x86. Yes, I'm counting all the decode hardware and microcode that x86 chips need to support legacy ISA. There just isn't much power burned there compared to modern cache sizes, execution resources, and queue/buffer depths which all high-performance cores need regardless of ISA. If you have an x86 processor that targets A9 performance levels, it will burn A9 power (or less if Intel makes it, given Intel's manufacturing advantage). If you have a ARM processor that targets Sandy Bridge performance levels, it will burn Sandy Bridge (or more) power."
It depends on what world you live in. In consumer PCs, Moore started fading in the late 90s. I seem to recall that as the transition period from "faster, but always $2000" to "about the same speed, and getting cheaper all the time".
I think speed was less important on consumer PCs once they became capable of video playback. That's the most compute intensive thing that most consumer PCs need to do. Further advances in the hardware were drivin by bloated OS and better audio-visual quality.
Step outside the consumer PC bubble, and speed still matters a lot. I don't think Google with its football field sized rooms full of servers thinks performance is no longer relevant. Since chips are a commodity, that performance tends to percolate down to the consumer level.
It is just expressing itself differently as we begin to hit the wall with process size decreases and speed increases. If wattage of the cpu goes down, you can pack more cores into the same area. Computing power is still going up.
Weaselmancer
rediculous.
'There’s been a lot more innovation in the world of mobile phones over the last 15-20 years than there has been in the world of PCs.'
While this statement may be true, it is true ONLY because mobile phones have been desperately trying to become PCs, and they're still a long ways off IMO. Without the innovation in the world of PCs we wouldn't have seen said innovation in the mobile phone world.
One system built on top of another.
Efficiency only really matters when supply is limited. On a cell phone or any portable system power is limited, and improvement in power efficiency will extend battery life. ARM is a good option when it comes to things like a tablet, but when you start to do everything an Intel styles chip can do they start to get more tricky. Sure ARM probably has a lower floor so it's minimum power usage is a lot lower, but when you start having it do everything in the same time span as an x86_64 does then it starts to look too similar to actually matter. Unless using an ARM processor can save me well over 500 bucks a year on my power bill I don't see their efficiency as actually that much of an advantage.
Please Intel, keep making those big, inefficient chips.
There's already a lot of comments along the lines of "of course he's going to say that, he's the CEO of ARM", but if you think about it, a lot of people who used to buy desktop computers and laptops don't NEED full-fledged computers. No matter which OS you choose, a tablet is more than enough for email, web browsing, instant messaging, taking and managing photos.
There's always going to be a place for full-fledged computers, but the ratio of regular users vs programmers makes us a rounding error at the end of the day.
Get free satoshi (Bitcoin) and Dogecoins
I think at the end of the day what really matters whenever moores law is invoked is the underlying issue of cost per transister... I don't see cost ever being relegated to irrelevant.
As transistors get cheaper you can take any combination of two paths:
1. Build cheaper gear with same capabilities.
2. Cram more into the same device to increase capabilities while maintaining price.
Either way moores law is still critically important to the industry no matter who wins a CPU architecture war.
With regards to ARM vs x86 I am content to make some popcorn and watch from the sidelines as both sides talk shit and throw down ever more energy effecient gear.
Same arguments from the 1990's... again. Remember when the 68000 was lower power than Pentium60 because it was RISC?
Yup, that all over again.
Even the magnitude of the design wins are the same today as they were in the early 90's.
Let's see who wins this round! Can the heavyweight (Intel) slim down to welterweight and defeat ARM?
Or will the both live together spurring on the same innovation (and crazy, low low prices!) we got from the AMD v. Intel race to 1GHz?
Tune in next week for more...
TALES!
OF!
INTEREEEEEEEESSSSTTTT!
https://www.accountkiller.com/removal-requested
Mobile chips are shit to people who need renderfarms, simulation farms etc. People still do real work out there.
People who keep crapping on workstations and servers seem to think everyone just needs a computer for texting, facebook and angry birds.
He misses another point (though reference to competition hints at it). With Apples switch from PowerPC to x86 and now this move to ARM, and Linux going mainstream via Android on ARM, software developers are getting ever better at making things portable and hence making the underlying CPU architecture irrelevant. Also notice that Android tried to make this explicit by running most stuff on the Dalvik VM.
Sure, power efficiency and die-area are important in many places, but don't think ARM is somehow going to have a lock on that.
I'm *very* glad you made your comment...it highlights a serious mistake in computing...a mistake that costs **BILLIONS** and effects people directly...
"If this guy gets his way, then we may never have sentient computers."
The idea that Moore's "Law" is somehow a scientific predictor that indicates the ability and future inevitability of humans making 'sentient' computers is...well...it's ridiculous.
1. Computers execute instructions...humans are beyond that, we have *free will*...For humans to make 'sentient' computers the system would by definition cease to be a computer. Something that has *free will* is completely different from something that just **follows instructions**
We can never 'make' a sentient computer...and even if we **could**....
2. Moore's "Law" wouldn't predict when it would happen at all, because it is annecdotal, not quantitative in nature. It shows an **interesting trend** that could **become** a predictive 'law' if someone made a theory of it and tested it.
So there you have it. IMHO, geeks create fictions by transmogrifying anecdotal data into quantitative data...why? Simple, quantitative data has less uncertainty, and therefore is easier for a 'geek-minded' person to sythesize
however, anecdotal evidence must be seen and analyzed for what it is...continuous, non-quantitative, noisy, and information-rich
Thank you Dave Raggett
I guess that's why all the low-power Pentiums with two cores and no hyperthreading have about ten or twenty reviews on Newegg, and all the Core I7s that score 5x higher on Passmark and use 3x as much power, while costing several times more, have hundreds of reviews.
More efficient processor (Pentium G630), 18 reviews: http://www.newegg.com/Product/Product.aspx?Item=N82E16819116406
Less efficient processor (Core i7-3770K), 357 reviews: http://www.newegg.com/Product/Product.aspx?Item=N82E16819116501
On Amazon it's the same pattern, with 47 reviews for the bitchin' fast processor and 7 for the futuristic low-power one.
I'm also noticing that the difference between the best GPU/CPU and the second-best model is a margin of 30-40% on a good day, just like it has been for the last decade.
So is the PC just an inefficient heater, then? Even my aluminum case is cold to the touch. If I didn't have so many fans (10 in total), would it make the room hotter?
I'm asking because I often see it claimed that PCs make great space heaters, but in my experience, this one plain doesn't. Under full load, it should draw quite a bit of power, but it outputs much, much less heat than lower-energy dedicated space heaters. I'm tempted to find my Kill-A-Watt and see what it says.
If you can't convince them, convict them.
I want more performance!
The display (monitor), input (keyboard) and sound have all increased in the relative value added. I think processor speed is not irrelevant, but it is less relevant to flat/touch screens, keyboard/voice recognition, and sound quality. The displays have never followed Moore's law, which is probably why they now glue them indelibly to the chip in tablets, so you have to replace them when the chip does go.
Gently reply
Just no. Even as mobile phones sell new ones every year because they're more powerful, which is driving your own damned business, this man remains brilliantly wrong. We've got so far to go. Video games alone, once the current horrid business model for triple A games is replaced, should maintain the drive for Moore's law in and of themselves. Not to mention machine learning, applications such as ever more data mining, heck eventually the entire stock market will be controlled by computer (which raises it's own questions).
Or just look at the top 500 supercomputer list. It keeps slamming upwards every year, and there are reasons people pay for them. Yes, I'm sorry that none of your processors are in that list ARM. I'm sorry your Cortex A15 appears to be behind Apple's new Swift cores in terms of efficiency. I'm sorry Intel is successfully starting to muscle in on your territory and you've yet to be successful in getting in on theirs. But denial isn't going to help you any. Especially since Moore's law actually helps you with power efficiency, so the statement itself is blankly a paradoxical one that doesn't work.
I think we need some expert analysis on this one.
The PC is used to create content. A smartphone is used to consume content. The PC functions autonomously (in a pinch). The smartphone is permanently welded to its cloud-nipple. The PC brings you smart ideas in shabby attire. The smartphone brings you shabby ideas in smart attire. The PC discourages walled gardens. A smartphone never leaves home without one.
Wake me up when my smartphone comes with a holographic projector capable of conjuring up 40" of viewing pleasure at a comfortable focal plane, and either a haptic keyboard (gravitational hologram?) or a brainstem feed a million times better than Swype.
Next we'll declare that mopeds and Harleys are the same form factor because there are more Asians than balding fat men. Clearly a modped is more like a Harley than a smartphone is like a PC.
To be more specific, Anandtech reports that Intel HD 4000 runs Skyrim playably. If it plays current gen games, it can't be all bad.
If wattage of the cpu goes down, you can pack more cores into the same area.
Most of those 64 cores will sit idle until programming techniques for making extreme parallelism reliable become taught in universities and vocational schools.
"America is all about speed. Hot nasty badass speed." Eleanor Roosevelt 1936
The real issue here is whether ARM can lock up the market before Intel's offerings become highly competitive. The answer to that is clearly NO, they can't. Intel wants to compete in the mobile SOC market and they clearly have enough of a technology edge with their Fabs to jam their foot in the door before ARM can lock it. Intel doesn't need to blow away ARM here, they only need to make sufficient progress on power consumption to put themselves on near-equal ground. They've already shown that progress. ARM has no Fab... they can build as many designs as they want but they have to team up e.g. such as their announced team-up with TSMC to compete with Intel on the Fab technology.
TSMC is trying to make a huge leap to 14nm, skipping several process sizes in-between and using a new transistor tech. It's a huge gamble on their part. I have grave, grave, GRAVE doubts that TSMC can deliver a power-efficient and cost-efficient 14nm process anytime soon. Intel, on the other hand, already has working silicon one step higher than 14nm and is already producing chips in bulk using processes that essentially still work at 14nm. Plus Intel has a completely spare Fab to do the transition on (Intel has three Fabs and their entire current production can run on just two of them). Plus Intel has the ability to migrate ALL their production to the new process whereas TSMC has to deal with hundreds of customers who are working across three+ generations.
At the same time, it is also quite clear to me that ARM will have a tough time competing against Intel in the PC and server space, where performance under load matters more. Performance at idle is something that Intel can fix (and has been fixing) on its consumer and server chipsets for the last few years... ARM has no 'in' there. So ARM is left with the much tougher job of boosting performance (a problem Intel has already largely solved). ARM has no leverage whatsoever at full cpu load. In addition, ARM is really unlikely to be able to compete against Intel for SMP loads. The best ARM will be able to do for their initial high-end supposedly competitive 64-bit offerings is virtually guaranteed to have inferior cache coherency characteristics across multi-core and will certainly not have any chance at being more power efficient. As you move into the higher end performance is determined more by transistor count... bigger caches, bigger TLBs, bigger branch prediction tables, better read-ahead, more write buffering, and so forth. This is very new territory for ARM.
Intel doesn't have to beat ARM in order to insert itself into the mobile space. ARM, on the otherhand, has to seriously beat Intel in order to insert ITSELF into the desktop and server markets.
-Matt
Man fuck Slashdot and their buttfucking captcha turning test. I had a fucking fantastic multi-paragraph response to this article and after failing the captcha test twice it told me to go fuck myself. So there it is... fuck it.
Something that has *free will* is completely different from something that just **follows instructions**
The religious right has spoken.
NEVERMIND THEN!
(whispering to everyone else: this guy is a nutjob.)
I own a netbook running Linux and over the last two years it has gone from a slideshow when playing high-dev mkv movies to playing them smoothly EVEN in windowed mode when the window is larger then the physical screen.
Now unless you are suggesting the hardware in a netbook updated itself, or the movie files morphed themselves, it can only be that the software has gotten faster.
It is true software tends to grow, but we also expect more of it to compete. Bookmarks in your browser? How decadent, what is wrong with a seperate bookmark program? Oh, you want it inside the browser? Well, you are bloating it then.
For movie players, it is not done anymore NOT to have fancy subs but the code for that increases the size of the movie player. I prefer mplayer and its variants because it comes with its own codecs meaning that even under windows, it just works, but it is bloated with codecs because of this.
Sure, WinAmp was an extreme example, do I really need to play games on my media player? No. And the market decided and it died. For countless other programs the market has decided that the bloat is wanted. See how many run mplayer with a shell just because they can't grock using the file manager as the play list and edit configs by hand.
Unless you use the cli for everything, you are the causer of bloat. You nasty person you.
MMO Quests are like orgasms:
You may solo them, I prefer them in a group.
Current Intel Compatible chips run much faster than similarly clocked ARM chips. I predict that in mobile PC's, arm will be the next big thing, while Desktop's, which have a constant power supply, will continue to be Intel Compatible.
put that dictionary away. it's doing more harm than good.
Is it b/c I used the word 'transmogrify'?
1. Being pedantic doesn't disprove my point
2. I chose to use a word that sounded like a made up technical word b/c, if you read my post, I think the whole idea of making a 'thinking machine' is...well...silly and quasi-technical...just like the word
3. http://www.merriam-webster.com/dictionary/transmogrify
Thank you Dave Raggett
Can you re-type your post? I honestly don't understand your intended message. From this: "(whispering to everyone else: this guy is a nutjob.)" I suspect that I am indeed said 'nutjob'
Which leads to the obvious question, why do you think my assertion about free will is related in any way to the 'religious right'?
I really have no idea how stating my reasoning about the human mind and the concept of free will makes me believe in god or a Republican.
If I said 'sapience' or 'sentience' would that make you feel better?
Thank you Dave Raggett