Will Moore’s Law Become Irrelevant?

CommanderFrank

Cat Can't Scratch It
Joined
May 9, 2000
Messages
75,399
According to ARM’s CEO Warren East, the future will rest with ARM technology and will replace x86 processors in desktops, laptops and servers rendering the concept of Moore’s Law irrelevant.

We don’t believe there’s going to be a massive switch to ARM-based PCs overnight. It’s more of a gradual process as legacy applications become replaced by newer applications that are more up to date.
 
ummmmm, how does increased market penetration by a competing architecture in any way alter Moore's law?
 
Because ARM processors are less powerful than modern x86 processors such as gaming quad cores etc. Therefore they claim that Moore's law has failed because they think ARM processors and other low power low performance processors will dominate the market.
 
so, the market is not going to demand more and more powerful ARM setups?
 
If Windows is going to look like it belongs on a smartphone or tablet they may as well build PC's out of smartphone and tablet parts.
 
If Windows is going to look like it belongs on a smartphone or tablet they may as well build PC's out of smartphone and tablet parts.

That is hardly going to run the 3@ 192000x108000 480Hz display setup I want
 
That is hardly going to run the 3@ 192000x108000 480Hz display setup I want

Huh? why would you want that? I'd settle for a 4K resolution screen @ 120hz and then use the rest of the horsepower to ray-trace in real time.
 
"To me a PC is really just a smartphone in another form factor."

Oh god.

Indeed. If that's not the scariest you've ever heard. Let's just go back to the stone age while we are at it.

What type of devices do they expect to develop ARM software on? Other ARM devices? No. Development will always have to continue on general purpose processors. Which means the arms race goes on forever.

Besides, we have not even really begun the VR revolution. And the horsepower required for that will not start on your smart phone. LOL.
 
I'm not sure if he fully understands Moore's Law if he thinks it is going to go away. Moore's law says nothing about power or speed, it simply states that the number of transistors double every two years. Whether the base architecture is X86 based or ARM based the same general rules apply ... in a competitive market people will want more capabilities on a recurring basis. Since ARM has the same limitation as X86 in that unlimited speed is not an option they must both add more "features" to remain competitive (more features takes more transistors, hence the law ;) ). This isn't likely to change even if there is an evolutionary branch from x86 to ARM.

That said, I don't believe that branch will replace x86. Although ARM has many advantages in the mobile space where heat and power efficiency demands outweigh performance ... the desktop and server space is more balanced between the needs of ongoing and increasing performance, in addition to some level of heat and power management. ARM taking over this market completely is no slam dunk, since it is unlikely that Intel will go down without a fight and they have many talented PHD types who can keep the x86 architecture competitive in the non-mobile space.

I think the house of the future is also going to favor LOTS of HD video screens. If this favored anyone, besides Intel, it would be NVidia and AMD since they might be able to optimize their video chips into high powered SOC configurations easier than ARM could. As Abraham Lincoln once said, "The Hen is the wisest animal in the barnyard, she never cackles until AFTER the egg is laid." :D
 
PCs are likely built as browsers, gaming rigs, or workstations. Of these, the browsers (work pcs and granny boxes) are by far the most common. They also haven't needed a processor with the performance of a chip produced since 2000. A quad ARM A15 would be idling 99% of the time in such a box and the user would never notice the difference. I can only hope that any gains ARM makes are gradual enough to keep from harming the gaming rigs and workstations.

Of course claiming "Moore's law" has anything to do with this is pure marketing. Moore's law has made the size/per chip cost between an x86 and a RISC trivial, thus the market only buys x86 because of the software base and the fact that Intel can build smaller-faster-cooler transistors. The only advantage now is that Microsoft's historical ability to create ever more bloated software to slow the hardware down just hasn't kept up.

The final thing is that we have been in Moore's Laws (very extended) death throes. It used to be that a smaller chip was cheaper, faster, and more efficient. Nowadays you are lucky to choose two. While the A15 might be vastly more huge and power hungry than a standard ARM, it will also be vastly more powerful. I strongly expect it (or later versions) to be sufficiently close to x86 in terms of performance and wildly more efficient in power usage to take over plenty of tasks (the server is where Intel is panicking: huge profits and fixed software make it a natural target for a new architecture).

Finally, where is Intel? The Pentium Pro was the beast that put the fear of Intel into the RISC chips. The "Core" chips that took the crown from AMD were clearly inspired if not based on the Pentium Pro (P6) microarchitecture. There is clearly room to make an even smaller "Molecule" design that would match the A15s dual issue, and cut the bloat that is Haswell*. Look to AMD to see what happens when you assume that Intel will continue to neglect to eat your lunch.

* note this isn't a criticism of Haswell not being a [H]ard chip. Intel can claim it is a high efficiency chip all they want, but it is really a high performance chip all the way. It is largely power efficient to prevent itself from burning up. Count the transistors and ask yourself if they could easily make a moderate performance chip with 10% of the transistors.
 
People that make arm should be more worried about intel catching up in power usage on x86, then arm creeping into x86 space.
 
my 1 cent: I think silicon will still be around for a decade or more, we might see more 3D chip stacking of more components 'dram and or/ L4 cache with more system on a chip features etc. Maybe when we get to that point intel will have absurd prices since average consumers don't need a lot to get by.
 
Someone please correct me if I'm wrong.

I think the 3770k can do like 100gflops, something in that range.

A quad core Cortex A9 gets uhmmm 300mflops maybe?

If someone can post real numbers I'd greatly appreciate it.
 
Also if East read the stuff that came out of the intel dev conference regarding Haslam (http://www.anandtech.com/show/6355/intels-haswell-architecture), you'd think he'd be shitting his pants. Intel is basically concentrating all their development efforts on power efficiency including things like better idle states and a ram video frame buffer in the display so that the vidcard and cpu can idle.

If Intel can get their power usage down to Arm levels, arm is out of business. It's a race between intel getting arm level power efficiency (OR batteries getting significantly better) vs ARM getting intel type speed (while intel gets 10% faster every gen cpu). My money is on Intel in the long run.
 
No one has a monopoly on efficient architecture. It just depends on what your design objective is. Yes, there is nothing inherent in ARM's architecture to become a powerful CPU. But you do need to tweak for performance. if you are designing for performance, you give up something on efficiency. There is no free lunch, otherwise Intel would have eaten it already.
 
Someone correct me if I'm wrong here...but isn't X86 based chips STILL faster than ARM?

I mean, Intel came out with the Atom based CPU for cellphones and tablets...same power and heat as a ARM CPU but crushes ARM speed wise.

So even IF smart phones and tablets "take over", why does this dumbass think ARM will be at the forefront of this ever emerging market? Intel or AMD just simply isn't disappearing.

I think this is a classic case of "dontknodafuktalkinboutitis"...:rolleyes:

ARM is only the big boy in portable electronics because AMD or Intel really haven't TRIED to get into the market. The moment they do is the moment ARM will be beat down.
 
What we even are needing more of too is a software/programming revolution, simple things getting more standardized on linux etc.. A more effecient instruction set for x86. The last major breakthur in programming was C++ which is a dominant language invented in wait for it... 1979!
 
Specs aside, the winner, at least for the next few years, will be the tech that does mostest cheapest. The bean-counters for the big data centers where CPU big-bucks reside have it down to absurd measures like watt-per-served packet, pennies-per-BTU-removed, etc.

Whichever architecture can maximize those kinds of values is where the money is. I frankly don't know if there is juicy enough low-hanging fruit in ARM architectures to motivate the binary switch you'd need from x86 for, say, Google. But I don't think anybody else has the operational experience to say so either. Only time will tell. But power and heat are becoming cost considerations for the biggest computing as well as the smallest, leaving the desktop box to get squeezed from either end as its part of the total CPU market shrinks in money, hence attention.
 
I can't believe given the nature of this forum and it's age that no one has that name here.
 
I'm pretty sure that Moore's Law isn't something most people even care about at this point. Only a few reclusive sorts of people bother thinking about it and, really, it just doesn't matter whether or not its relevant because it never has been important to anyone aside from being a conversation piece for nerds to impress each other and show off to normal people. So yeah, it's less "become" and more like "never has or will be" in the whole relevance thing.
 
hate to tell you scribble

to date Moore's law is fully in effect, manifest on the material plane, the facts of the matter

do a little reading...we have come a long way from punch tape and mechanical relays
 
Someone please correct me if I'm wrong.

I think the 3770k can do like 100gflops, something in that range.

A quad core Cortex A9 gets uhmmm 300mflops maybe?

If someone can post real numbers I'd greatly appreciate it.

A quick bit of Googling indicates that the A9 can peak at 4gflops.

A15 is supposed to be about 40% faster. So best case scenario, 6gflops.

The reason ARM can potentially compete so well in the server market is power consumption. The A9 uses about 5.5 watts to get 4gflops. The 3770k uses 165watts to get about 60 gflops? (comparing at stock speed, because you don't OC in a professional server environment). Obviously that is a pretty significant performance per watt advantage to the A9.
 
I'm pretty sure that Moore's Law isn't something most people even care about at this point. Only a few reclusive sorts of people bother thinking about it and, really, it just doesn't matter whether or not its relevant because it never has been important to anyone aside from being a conversation piece for nerds to impress each other and show off to normal people. So yeah, it's less "become" and more like "never has or will be" in the whole relevance thing.

People might not worry about it by name but I don't see too many people saying that their electronics are all fast enough and no more improvements are needed ;) ... with electronics companies trying to get us all on a 2 year upgrade cycle for phones and a 3-4 year upgrade on computers and other technology they will need to follow Moore's law or change their business model ... most people will not change a phone every 2 years unless the features are dramatically different (and that will likely require them to continue to follow Moore's Law :) ) ... people don't usually worry about Gravity either (at least until they are falling off a cliff or an Apple drops on their head) :p
 
Someone correct me if I'm wrong here...but isn't X86 based chips STILL faster than ARM?

I mean, Intel came out with the Atom based CPU for cellphones and tablets...same power and heat as a ARM CPU but crushes ARM speed wise.

So even IF smart phones and tablets "take over", why does this dumbass think ARM will be at the forefront of this ever emerging market? Intel or AMD just simply isn't disappearing.

I think this is a classic case of "dontknodafuktalkinboutitis"...:rolleyes:

ARM is only the big boy in portable electronics because AMD or Intel really haven't TRIED to get into the market. The moment they do is the moment ARM will be beat down.

Something i've been wondering. If you were to take out all special commands like SIMD (MMX, 3DNow, etc.) and hyperthreading. How much less power would an x86 CPU use if applications were to strictly adhere to legacy x86? And how would it compare to ARM then hz to hz?
 
If anyone actually skimmed over the article they would have realized that it's the reporter that doesn't seem to understand Moore's law. ARM's CEO said nothing of the sort...
 
I'm pretty sure that Moore's Law isn't something most people even care about at this point. Only a few reclusive sorts of people bother thinking about it and, really, it just doesn't matter whether or not its relevant because it never has been important to anyone aside from being a conversation piece for nerds to impress each other and show off to normal people. So yeah, it's less "become" and more like "never has or will be" in the whole relevance thing.

Eh, its been an important thing in the development of computers, in a sense that there is always a need to cramp in more transistors into a SoC die and Moore's law is a reflection of that. We continue to shrink the transistors size not for the sake of keeping Moore's law alive, but because there is a need for more transistors.

Here's a simple example. Cache memory is an important part of a processor, the more the better but how much of it we can have depends on how many transistors we can cramp into a single chip, and in modern processors they do take up a significant area. There's the need for more transistors to improve performance.

SRAM, the type of memory used to implement onboard cache do consume a significant amount of power and voltage scaling by itself is not suitable to reduce its power consumption because then the SRAM becomes unstable. So what we do is to implement better SRAM design that use more than the typical 6 transistors per cell design, and with that we could run them at very low voltage and maintain its stability. So, here is a need for more transistor to improve power consumption.
 
I think you can do that already.

Depends. FTP Server? Yes, i use this all the time. PnP Server? yes, it's the one that allows the phone to stream videos to Windows Media Player on the desktop. Web server? Yes, try downloading AirDroid.

If by server you mean a full PHP, SQL, server on your phone, i don't think i've seen that yet.
 
Back
Top