AMD Says That CPU Core Race Can't Last Forever

[dark_knight33]

[citation][nom]awood28211[/nom]Here's a idea (maybe a prediction). Instead of "cores" let's focus on "computing entities". Put 16 cores on a chip, design the system with tons of ram, drive space, etc and let the hardware turn this one machine into 4 logical units... Powering off cpu's, ram and resources entirely whenever an "entity" is shut down. Make this box 4 perfectly independent machines (or even 8 or even 16 all on one motherboard. and whatever other devices that are necessary. Instead of putting a computer at every desk in cube-topia, just put one for every 8 cubes... connect all the peripherals wirelessly...keyboards, mice, display, speakers, printers, so on and so forth... Let this box become the "terminal server" of the cube but instead of sharing the server space, you actually get a full blown machine logically cut out of a box... This would be much like the server blade concept except that the "blade" would be physical cpu's in the core with it's dedicated (or dynamically shared) ram and other resources. No competition for CPU time, it's all in 1 box, saves power, removes the wires and keeps it all independent. The major drawback would be when the entire box fails (power supply or whatever) each box goes down together.[/citation]

It's called Cloud Computing. People stupidly think it's something new, but it's actually an old concept dated from the 70's spread over a wider area using IP instead of RS232. Look up a DEC Vax 11/785. Also, I really *really* hope software as a service so utterly fails in the consumer market no company would be willing to touch it with a 10ft pole.

 

[the_krasno]

The future is quantum computing. Operating at relativistic speeds means that theoretically a PC would do what you want it to do before it actually finishes processing the data!

 

[wishmaster12]

what is it "nothing is faster than light in a vacum" make a light and vacum processor

 

[tayb]

[citation][nom]Stardude82[/nom]Or maybe the technology will just hit a wall. For instance, passenger jets don't travel faster today than they did 50 years ago despite years of development. A lot of technologies just didn't work out like supersonic flight and right now huge efforts have to be made for incremental improvements in efficiency.[/citation]

Bad comparison. There were known physical limits to flying before flying was invented. You can only push so much air out of the way. These "walls" we are talking about in terms of the number of cores are from a power constraint perspective and are purely hypothetical.

Anyways, we will move on to quantum computing before we see 128 cores so in the end I suppose he is right.

 

[Guest]

The reason software isn't more multithreaded is because Windows multithreading sucks.

If you're willing to tolerate GUI freezes and an application with a single thread, you can have stability easily. As soon as you start adding more threads, stability goes all to sh.i.t. Even the most basic examples of multithreaded code will cause random crashes and instability.

 

[theshonen8899]

Make sense. But in terms of server products, I think the core race still has quite a way to go.

 

[kelemvor4]

[citation][nom]Linux_Multithreading_FTW[/nom]The reason software isn't more multithreaded is because Windows multithreading sucks. If you're willing to tolerate GUI freezes and an application with a single thread, you can have stability easily. As soon as you start adding more threads, stability goes all to sh.i.t. Even the most basic examples of multithreaded code will cause random crashes and instability.[/citation]

troll fail.

 

[Pyroflea]

I've been saying this forever. You can't just keep adding cores and increasing frequency. Eventually you'll need to redesign the wheel and just start utilizing new technology.

 

[Shin-san]

I agree. The 16-core CPU AMD is coming out with is ahead of many applications that can actually use all 16 cores. The Law of Diminishing Returns has got to hit sometime.
[citation][nom]back_by_demand[/nom]The whole point of multicore was not to run a program on each core, although that does quite nicely when your AV kicks in, because who has more than 6 serious CPU intensive programs running at the same time?The idea, and rightly so, is to split a single program up between several cores to make it run faster. Great idea in principle but where are the slew of multicore programs? ..... Silence.Developers, please get off your collective fat asses and write the next generation of programs that can actually utilize all this expensive hardware I already own.[/citation]
Easy to say, but hard in practise. There's issues with multicore programs like synchronization. It gets tricky to run multicore programs at 100% CPU utilization. However, you do get cases where you get a dual-core program running at 70%/70%, where you gain 40% more power.

[citation][nom]molo9000[/nom]How about replacing x86 instead of adding more and more stuff to it?I'm not very familiar with x86, but is a 32year old instruction set still useful?The number of transistors in a processor has grown from 29 thousand to well over a billion in those 32 years.[/citation]It's still pretty useful. Still, the heavy computation usually gets done on SIMD units now (MMX/SSE/AVC/etc)

 

[ahumphers91]

Wow, 128 cores...10ghz chips heh. As Tim Allen on tool time would say.."uuhhhh" haha.

 

[palladin9479]

A system can always use more cores. The way hardware scheduling works is that each thread is giving a certain amount of time on the CPU to execute its instructions then it's interrupted and another thread its put in its place. All instructions fed through a CPU are binary machine code, best represented as assembly language. These instructions operate on a set of registers, a stack and a bunch of instruction and memory pointers. Virtual 86 mode (introduced in the 80286 / 80386 time) makes it so each thread assumes its the only thread executing on the CPU and therefor expects its registers and stacks to be maintained as if it was the only thread being executed.

What this all means is that if you have 300 threads running on a system then you have 300 sets of contexts (registers / stacks / pointers) that much be maintained simultaneously. Each core can only handle a number of simultaneous threads equal to its number of register sets. On AMD CPU's this is one register set per core, on Intel HT CPU's this is two register sets per core (HT is just using multiple registers per core), on a Sun SPARC T2+ CPU this is eight register sets per core. This swapping in / out of context can quickly become CPU intensive. If anyone is bothering to check they'll notice the majority of CPU instructions executed is local reads / writes and not actual math / compares. If you have 300 total threads on a system then those 300 contexts must be swapping in and out constantly depending on how often a thread needs CPU attention, even if its just for a small handful of instructions. You quickly end up spending more time swapping contexts in and out then actually doing the work.

Big issue today is everything is "bench-marked" with the least amount of background apps running. This gives the application the majority share of a CPU's time when in reality it would be fighting for that time with other applications. So while it seems programs aren't taking advantage of the additional cores, in a real scenario the system would take advantage of those additional cores for better scheduling and more efficient context swaps.

 

[liveonc]

Better code won't sell more hardware. So unless AMD/Intel really want people to use their PCs for the ten years it's made to last. OC the heck out of it so it dies after 3-5 years. If we ever reach a wall & it can't be breached. It's time to make better use of what we have. But until that day comes, nobody cares enough & if there's not enough power we'll make more power. But a day will come when people will ask if they want to invest in a thousand new servers every 3-5 years, or if they want to spend their money on making what they have run faster. If efficiency can improve 50% for less than a hardware upgrade which only improves performance by 33%, it's time to invest in software.

 

[dogman_1234]

Here is what we need to do:

Use graphene as new transistors for a decade
Get rid of x86 it has bee around almost more than 30 years
Introduce 128 bit 'parallel' processors
Integrate a cloud computing integer core into the physical core
Get rid of virtual cores, they eat up power

 

[Scott2010au]

Some people may want to read up on the different 'bits' that apply to processors (CPUs).

Namely: Data Bus width, Register width and Address Bus (or Address Space).

ie: It is possible to have a 256-bit wide bus, with only a 40-bit address bus and both 32bit, 64-bit & 80-bit registers.

Add in register renaming (http://www.google.com.au/search?q=register+renaming&ie=utf-8&oe=utf-8&aq=t&rls=org.mozilla:en-GB😱fficial&client=firefox-a) and you'll see that we already have '256-bit' processors today. (Many CPU's already have internal 256-bit wide paths).

 

[dEAne]

From integration comes disintegration.

 

[Scott2010au]

http://www.google.com.au/search?q=terahertz+processor

 

[Thor]

It's funny to hear that.
I think many guys forget that exist other matter than silicium. And until now companies who create electronic chip don't use.

Using for example gallium arsenide can make chip 10 time faster.
Other matter like artificial diamond can make possible to have 10 GHz or more speed clock.

Before to change matter and/or technologies, electronic companies want press more silicium to give all what they can.

And we even don't talk about optical CPU neither quantum CPU.

And we use again binary which can be replace too.

So electronic have a very open and nice future.

 

[jimmysmitty]

AMD is right. Its also the reason why Intel has been foraying into stuff like Terascale which has a lot of coes but isn't quite the same.

I imagine we will have a lot of cores in the future but not the same as we have it now. It will be more like Stream Processors, being many but set for multitasking instead of accelerating a single task.

 

[WarraWarra]

oooh all the AMD employees are upset about what we post here about their GOD or the leader of their AMD world.

I do really like AMD and would switch back in a heartbeat just like with Apple love it but they need to give me a valid reason to spend my hard earned cash on their products.

Yeah as much as I dislike Intel I still end up buying Intel as they give me overpriced expensive value for my money at this stage.

Just because AMD plays on all our suckers emotions about them being the underdog just like us their consumers we keep on falling for their lame excuses like this core joke.

Eventually cores and speed and everything will be limited DUH but for the mean time do something decent or go home.

 

[demonhorde665]

[citation][nom]waksksksks[/nom]he maybe right. quantum mechanics processor is in the future. let say who can make it out for the public. amd or intel. you never know. by the time when you look back probably everything you have now is all junk. always happy to see new technology for better tomorrow.[/citation]


i doubt when quatum computing hits the consummer markets that intela dn amd will lead the pack , consider this , while both companies do put money in that research what they throw at it is minimum compared to waht tehy throw at advancing the x86 design further. likely one of the many unheard of (comercially) companies that are working on quantum comptuers for teh goverment , will likely lead the field when they become comercially availble.
Further more pure quantum comptuers will be next to useless on the home market , mostlikely what consumers will see will be hybrid chips , and it's not likely that intel and amd will be leading the pack on these , but who knows perhaps when quantum hybrids become more feasible amd and intel willthrow a lot more money into develeping thier own , but they certainly won't be the leaders out the gate first ,unless they make a x86 hybrid that can run x86 code

 

[Tom Kamkari]

LOL wow, how long did it take this guy to come up with the most obvious answer to just about anything, like we all thought that in 10 years we would be dealing with chips with 10,000 cores, lol.

Clock was/is better/important cause the faster the clock speed the faster the cpu, number of cores are good to but to a certain extent, and of course technology will produce something in the future that is better then bunch of cores and great clock.

 

[moonwave]

What core race? core wars? LOL. Newell is trying to bring up a distant past Ghz competition between the two rivals, and basically pretend that it's still going on, but now it's all about cores. This was back when amd couldn't keep up in clock speed, so they began to bash Intel to say it wasn't important even though there were many other forces at work other than the clock. Then amd finally got a little ahead for a short span which has happened on and off since both companies have started making chips. Not surprising. But then they got arrogant and tried to embarrass Intel and publicly challenge them etc.. Something Intel silently ignored while they were working on their next chips. A couple of years went by and amd got their A-ss-es kicked hard. But now they are coming back again with the same "promises", and B.S. that got them in trouble in the first place. What they need to do is just create a better product. You know, like Intel has done all along without continually bragging and implying they are better etc.. And then come out with it. That's the way it's done.