digg

Join DiggAbout

Discover and share the best of the web!
Learn more about Digg by taking the tour.

See the original image at news.cnet.com

Intel says to prepare for 'thousands of cores' in processors

news.cnet.com — Intel is telling software developers to start thinking about not just tens but thousands of processing cores.the chipmaker is now thinking well beyond the traditional processor in a PC or server."The more cores we have the better. Provided that we can supply memory bandwidth to the device."

Bury
show profanity settings
expand all
  • lardoandkeg, on 07/01/2008, -28/+5Intel Intel he's our man, if he can't do it ...... GREAT!
    • easypie, on 07/02/2008, -5/+0Yes! yes! let's all cheer for the man.
    • thatspsychotic, on 07/02/2008, -1/+3Oh come on, no one else got the Aladdin reference?
  • jessechan, on 07/01/2008, -24/+19This is going to be interesting as NVIDIA and AMD/ATI already have hundreds of cores on their GPUs. So for Intel to have thousands of cores is not as far off as we may think :)
    • jumanous, on 07/02/2008, -9/+26hundreds? Which GPU do you have?? They have plenty of pipelines and pixel shaders.. but cores... no.
      • zyklon, on 07/02/2008, -3/+16Exactly. Unless he's spilt the beans about some mega-secret project he was alone privy for.
      • chazza125, on 07/02/2008, -3/+16The Unified Shader Architecture actually consists of hundreds of 'modules' which can actually be classified as self contained 'cores'.
        See: http://www.extremetech.com/article2/0,1697,2053309 ...
      • invinciblechunk, on 07/02/2008, -1/+14Nvidia GPUs have had hundreds of ALUs (128 on the 8800 GTX, 240 on the GTX 280) for a while now. Normally these are only used for shaders, but with the CUDA SDK, they are also exposed as proper general-purpose processing cores, which can be programmed using C++ and threading. Context switches are hardware-accelerated, too.

        If you're going to be ignorant, at least try to do it quietly.
      • sponeil, on 07/02/2008, -0/+9jumanous is wrong, invinciblechunk is right. Since the GeForce 8000 series cards came out with unified shaders, they have 100+ cores which can be used as general-purpose processors.

        invinciblechunk is wrong about one thing, though. Digg is not a site where people keep their ignorance quiet.
    • proficient, on 07/02/2008, -5/+5False.
    • Ramble, on 07/02/2008, -4/+2Hundreds of shaders, yes. Only one core.
    • Stormwern, on 07/02/2008, -1/+7Reminds me of "***** Everything, We're Doing Five Blades" :)

      http://www.theonion.com/content/node/33930
      • lengau, on 07/02/2008, -0/+3I think the best part of that article is that I got a free razor from Gilette the other day, and part of the ad on the box they mailed it in was that it has five blades.
      • CasinoJack, on 07/02/2008, -1/+2The Gillette Fusion Power Stealth is actually a damn good razor - shame it's got such a stupid name.
      • lengau, on 07/02/2008, -0/+1Meh... I'll try it eventually (I still have one from last year, too), but I think I'll stick with my electric.
  • Omek, on 07/01/2008, -10/+4Next it will be millions of billions of trillions of cores... cores infinity?
  • mark076h, on 07/01/2008, -3/+59The Intel KiloCore Extreme Processor, sounds pretty nice
    • zyklon, on 07/02/2008, -0/+16The name alone would sell me.
      • MattBD, on 07/02/2008, -0/+9If they get up to millions, it'd be MegaCore.
      • ortucis, on 07/02/2008, -1/+1I'd pay 10 bucks for zyklon.
      • kaiwai, on 07/02/2008, -1/+8Or you could just call it "Intel Hard Core - Throbbing power within your grasp"
    • 007isbond1, on 07/02/2008, -1/+1Intel 2^infinityCore..
    • Thud, on 07/02/2008, -5/+3I can't wait to get my first 1000-core processor. I bet it'll overclock nicely to 1400 - 1500 cores.
    • muzy, on 07/02/2008, -1/+2Crysis will run at 600 fps mark my word
      • freebird09, on 07/02/2008, -1/+1Blasphemy! They said it couldn't be done!!
      • nielkie, on 07/02/2008, -0/+2Crysis is not truly multi-threaded, so no advantage would come out of having so many cores.
    • eengineer, on 07/02/2008, -1/+1With all those cores we could redundantly check to see if your music/movies are pirated!

      The RIAA/MPAA will approve.
    • schneb, on 07/02/2008, -0/+1And Snow Leopard will be ready with its Grand Central Station.
    • darthjure, on 07/02/2008, -0/+1Definitely want a 2 giga-core processor.
  • scamerica, on 07/01/2008, -2/+24So... does this mean that the army already has it?
    • lengau, on 07/02/2008, -12/+5The Air Force already has computers with ZILLIONS!!! of cores!!!!111 not Millions, or Billions, or Trillions, but ZILLIONS!!!!!!!!111one
      No seriously!11! It's true!!! And you know what???????????// They're 800 GHz Pentium 8s with 497GB of L2 cache!!! and they all fit ON THE PILOT'S WRISTWATCH!!!

      THEY'RE USING THEM TO SPY ON US!!!!!11111111111111eleventyone

      /tinfoilhat
  • awesometastic1, on 07/01/2008, -8/+56the problem is of course that it doesn't matter how many cores you have, if you can't increase speed to memory, it does you absolutely no good. further, it is near impossible to break up a lot of processing problems into many pieces. So sure having 20 or 30 cores is going to increase speed (say if you have a bunch of apps open). But with current programming languages and the upper limit on efficiency between breaking problems up onto separate processors (see The Corollary of Modest Potential by L Snyder), having 1000's of cores is going to do us absolutely no good whatsoever in the personal computing realm.

    I mean, my laptop only has two cores and they pretty much sit at 0-5% usage the vast majority of the time. I guess, when we have 1000's of cores we can all marvel at them all sitting at 0% usage all the time except for the one or two that are actually handling running apps.

    Now of course the place where having 1000's of cores each with a separate bus to memory could be really handy is web farms. Instead of literally needing tons of servers you could hook one server with a few hundred gigs of ram up to a nice sized RAID array and vuala, web farm.
    • grumpyrain, on 07/02/2008, -4/+26* voila
      • pwnerofnoobs, on 07/02/2008, -2/+23he said vuala and he meant it godamn it.
      • kjd84, on 07/02/2008, -0/+17Was I the only one that read Vuala as Vulva??
      • bugsysservant, on 07/02/2008, -0/+0Thanks. I don't know why that bothered me so much, but it did.
    • mcduck, on 07/02/2008, -4/+41We must mail Intel this comment before they waste more money researching this. Clearly they have no idea what their talking about.
      • mythril, on 07/02/2008, -2/+10http://en.wikipedia.org/wiki/Amdahl%27s_law

        specifically:
        http://en.wikipedia.org/wiki/Amdahl%27s_law#Parall ...
      • easypie, on 07/02/2008, -7/+1Thanks for your brilliant idea McDuck.
      • lengau, on 07/02/2008, -0/+2@mythril -
        NOTE: I am not a parallel computing expert (although the concept really fascinates me, especially with regards to things like brains, which are, as far as we can tell, amazingly parallel computers).

        Amdahl's law refers to accomplishing one task at a time. The nice thing about parallel computing for desktops isn't really the making stuff faster bit, but more the multitasking piece behind it. This is why 100 cores in a machine would be handy in, say, servers. It's not that you want to make a single web page in, let's say, one nanosecond rather than 0.1 µs, but rather that we want to make 100 unique pages in 0.1 µs. This is a lot more reasonable.
        Also, with more and more cache in each core, slower access to main memory becomes less of a problem. If there are enough processors to have each task allocated to one core, and each core has enough cache to store everything in its task, fast access to main memory becomes less necessary.

        http://en.wikipedia.org/wiki/Amdahl%27s_law#Limita ...
      • Trevahaha, on 07/02/2008, -0/+2There's also a lot of research work that is going on on how to automate parallel computing (embarassingly parallel computing)- even automatically detect these in somewhat traditional linear processes. For instance, if you have code that does a foreach loop on a large set of data, the system could detect that the code within doesn't impact or depend on the results of the other loop and automatically run this processes in parallel. This is really geared toward HPC (high performance computing) clusters, but the same idea could be applied to desktops with many cores as well.

        With Steve Job's comment about difficulties to programming parallel processes, I wonder if Apple Research has found a practical solution to this for the next release of OS X. I know Microsoft is also putting a lot of effort into this too.
    • goldenmug8, on 07/02/2008, -1/+14Computers are used by people other than normal everyday people...such as scientists and engineers who NEED that speed.
      • jman583, on 07/02/2008, -1/+14Or gamers, who NEED the speed... or else.
    • ba5e, on 07/02/2008, -1/+9The whole reason for this article is to state the developers must prepare to write code that is efficient over many more than 2-4 cores. So if they do this, for example with 100 cores then each core may be loaded with 1%. on a regular chip (one core) in the same class as the 100 core chip it would be 100% load.
      • senfo, on 07/02/2008, -0/+6That was kind of the point of awesometastic1's comment. It's easier to say we should split up tasks so that they can be spawned across multiple processors than it is to actually do it. Furthermore, it takes a much better software engineer to write multi-threaded code. In addition to this, context switching is still expensive in modern operating systems. Truth is, it's rare that I see or write CPU intensive code that executes across more than two or three threads. More often than not, the worst latency occurs because of disk IO while the CPU patiently waits for something to do.

        Software design patterns for writing multi-threading applications are still being developed. Even Microsoft is getting in on the fun with its Parallel Extensions Framework, which Microsoft hopes will make it easier to write applications that scale to take advantage of parallel hardware.

        http://msdn.microsoft.com/en-us/concurrency/defaul ...

        We're really just getting started. I wouldn't be so bold as to say that we'll never require 1,000's of cores (I'm sure there are many data centers and super computers that could make use of this technology), but it's probably going to be a while before desktop computers will come close to requiring such a large number of cores.
      • Trevahaha, on 07/02/2008, -0/+1You're exactly right. It's also usually recommended not to spawn off more than 20 threads, so it will definitely have an impact to programming models if we start talking about hundreds, or even thousands of threads that an application can use. It's going to be an interesting change in the approach to software development!
    • BassJunkie, on 07/02/2008, -0/+5I wholly agree, multiple cores on a CPU are pretty much redundant without the programmes to utilize them properly. I think we are starting to see the beggining of this with Adobe making use of multi core system's on Photoshop with the multicore mac's, where rendering task's are divided up amongst the available processor cores.

      As the OP states, where this sort of technology will come in useful is in servers, especially with the growing interest in virtualization, being able to cram more cores into something the same physical size as current servers means more can be done at once. This is all based on the assumption that the other system buses can be developed at the same speed, so we need much faster ram and HDDs to support the CPU.
      • Trevahaha, on 07/02/2008, -0/+1Do you know how well this scales for Photoshop? Does it take advantage of 8-cores as well as it does for 2? Will it know if there are and utilize 20 cores?
    • bfisch1983, on 07/02/2008, -1/+9I am shocked that intel would fixate on a somewhat meaninless number, tell us it's the most important number, and then market they chips by constantly increasing the super important number. They just aren't the same company they used the be....
      • lengau, on 07/02/2008, -1/+6You're right. Their meaningless number is now number of processors in the computer rather than number of Megahertz of the one processor. Oh boy... How times have changed...

        (That said, when they first got into the business, they were a lot better about it, but this sort of thing has been going on for over a decade).
      • AdamFromMyspace, on 07/02/2008, -1/+2I wish I could digg you up more than once.. so true
      • ammundsen, on 07/12/2008, -0/+2Intel has good marketing. Clock speed being a current barrier they need something new to emphasize. 45 nanometer is better than 90, but by this measurement smaller is better. Marketing smaller numbers as better is contrary to most peoples expectations. So all they are left with is the number of cores.

        I recently switched to a dual core. I'm underwhelmed with the increased performance. The biggest advantages for me have come from the larger L2 cache.
    • turpialito, on 07/02/2008, -0/+9@ awesometastic1:
      I disagree with the core usage part as follows: I'm an architect. I designed a 5 br house with lots of detail (trees, ashtrays, plants, utensils, appliances, etc). It took me a full night to render an 11-second animation on a P4EE@3.73GHz (6 GB RAM). On a Core2Duo@2.66GHz (same RAM) the same animation took about a third less. Okay, so the cores would sit idle while I'm not rendering or while I read my mail or whatever, but I sure could use as many cores as I gan get my hands on to increase my productivity. Plus, I'm sure you can break up a CPU-intensive render among however many cores you have. I think the same goes for CGI movie makers, video editors, etc. I think you'll find plenty of market for multicore processors.
      • lengau, on 07/02/2008, -0/+1I'll bet ILM and ORNL will love having 100 cores/processor. You're right in that.
        Also, isn't part of this also that you can read your e-mail/do whatever while the machine's rendering without affecting the render as much (or, possibly [though improbably] at all)?
    • abrasion, on 07/02/2008, -1/+5Yep, and when it is slow, 99% of the time it's due to the dopey 1960's technology of a magnetic hard disk :( good lord where are decent, fast, cheap SSD's?

      Furthermore, code is written poorly nowadays too, tragic stuff :/
      • lengau, on 07/02/2008, -1/+3Well three mediocre programmers cost about the same to hire as one excellent programmer, and three is more than one, so it must be better.

        ^This coming from a mediocre (at best) programmer
    • MindStalker, on 07/02/2008, -1/+5What if a compiler was written that used the thousands of cores for branch optimization, say you have a 1000 cores, you set 800 cores for branching and have each of those cores follow a different branch of the program before it even gets to the decision?
    • Charlotte_Web, on 07/02/2008, -0/+4My understanding of the problem is the diminishing returns of adding more cores. That is, you have to factor in the processing power needed to manage the traffic between the cores, and the more cores you add, the more traffic management you have to do in an increasing amount. The performance gains relative to the number of cores added slopes downward rather sharply beyond a certain point.
    • griz, on 07/02/2008, -0/+4What about a 1000 core system sitting in an office just waiting to chew on a problem that any number of smaller systems hand off to it? Centralized computing isn't new, but having 1000 cores to deal with hundreds of user inquiries would be mighty fine.
      • turpialito, on 07/02/2008, -0/+1I don't think I have room for any number of smaller systems, let alone the means to support the power requirements : (
  • swordedge, on 07/01/2008, -5/+46The problem as I see it is code. Code MUST be written for multiple cores to run any faster. While I don't know what percentage of programmers are capable of coding for multiple cores, I believe it to be under half.
    Lets say you have a two D shootem up space game where there are 16 robot ships in addition to your own and your friends. Each ship gets its own thread. Code for each thread has to have some of its own memory space but also needs to access community memory. It must NEVER just write to that memory else it overwrite something that another thread put in. It has to issue a code that says, "I am writing this address", then wait till it gets permission, then write to that community memory, then undo the code. Also, keeping track of what 16 or how ever many robot ships there are means running a very active database. Game programmers are NOT database administrators. They often use a messaging system to communicate between threads. The main thread becomes a glorified message handler. This whole scenario gets real complicated real fast. Writing good multithreaded code is not easy. This report simply means that coders must step up and make sure they can code for more than one processor if they want to be hired in the future.
    • Steinr, on 07/02/2008, -6/+6I believe that apple is trying to solve this exact problem by developing Snow Leopard and Grand Central.
      • Kamujin, on 07/02/2008, -3/+3Apple couldn't even architect a multi-threaded OS. Who are you kidding?

        OS X is a FreeBSD fork. IE, some else designed the kernel.
        OS 9 didn't even have preemptive multi-threading.
      • nycmac247, on 07/02/2008, -3/+2@Kamujin

        Have you even read about 10.6?
      • Kamujin, on 07/02/2008, -0/+5@nycmac247
        Yes, I've also come to take what Apple says with a grain of salt. IE, their "64 bit" leopard. Which is actually a 32bit OS that supports 64bit mode switching. There is nothing in the Grand Central announcement that is even slightly novel. Many people are working on parallel libraries, so I just hope Apple is smart enough to clone the best one.

        Look, I have a MBP right here on my desk. Apple makes nice products, but most of the things they "invent" wind up on the trash heap of history. Apple does way better when it takes other people's technology and makes it shiny.
      • superkendall, on 07/02/2008, -3/+1So you haven't read about 10.6 then, or really understand Grand Central or OpenCL. Thanks for clarifying your astounding levels of ignorance on the topic.
      • Kamujin, on 07/04/2008, -0/+1LOL, OpenCL.

        Yeah, I've heard of CUDA.

    • eengineer, on 07/02/2008, -0/+4I can tell you this. I used to code for a living and still do when I need to. My role has changed and I can say I was a fair programmer and we have some good and great programmers albeit no one that I would say is a "superstar" Google level coder. I doubt that anyone here could handle more than a handful of threads. I think the way to utilize hardware like this is to get the OS and compilers to do this kinda work.
      • Kamujin, on 07/02/2008, -0/+3I think it depends on your problem space. In my industry, the problems tend to be appropriate to highly parallel coding techniques. Thus, we have all been doing highly concurrent multi-threaded for 10 years now. 10-100 threads running with mostly lockless synchronization.

        But many problem do have interdependencies that essentially serialize them, which does reduce the effectiveness of adding cores to increase speed.
    • ca0abinary, on 07/02/2008, -0/+4You have the basics, but it's not actually as hard as you make it seem. You -could- have one thread per ship, or just a thread for game logic, or beyond that, maybe a thread for "ship" A.I. The memory thing is handled with simple memory locking, and databases don't really go into the equation. I know that in business software development databases are huge, but in games they may be used to keep track of state in an MMO, but most games only use databases for assets such as textures, sounds, etc.
    • brianpeiris, on 07/02/2008, -0/+3Quick! Learn Erlang! http://en.wikipedia.org/wiki/Erlang_(programming_l ...
    • ell0bo, on 07/02/2008, -0/+1Code needs to be improved to handle multiple processors, but languages must too. Sometimes it's not inherently obvioius how to split your code up to take advantage of multiple cores. I've done it in some places, but all too often people have become far too comfortable with sequential programming, since that's what most languages lend towards (even OO ).
      Worst has to be trying to multi-thread in Java though. Granted this was a year and change ago, but when I did it still only ran on one processor because the JVM was using only one hardware thread. I'm hoping they fixed this by now...? Anyone?
  • rac1234, on 07/01/2008, -1/+27I'm ready to start coding for it if Intel are ready to give me one to practise on.
    • Russelllucid, on 07/02/2008, -3/+7If Intel gives me a test chip I'll actually learn to code!
    • eengineer, on 07/02/2008, -0/+1More power to you man. Good luck trying to keep it thread safe and keeping track of what is accessing. I dugg you up, but I don't feel the same, I wouldn't want that burden.
  • Mathieugothax, on 07/01/2008, -3/+8I can see this coming miles ahead. The two titans clash, Intel vs Nvidia. Oh yeah... =) More competition = better products released faster, with lower prices. Competition = Good for us, the customers.
    • AppleGeorge, on 07/02/2008, -5/+2Last time I checked nVidia didn't produce processors.
    • Slade605, on 07/02/2008, -2/+3Man just think Intel GPU's and nVidia CPU's wait... a second.
    • tsunamisteve, on 07/02/2008, -0/+1wow. you some kind of sorcerer.
    • Timmyftw, on 07/02/2008, -0/+2If Nvidia cant figure out something as simple as how to get a driver to work with S.T.A.L.K.E.R i dont know how competitive it is going to be.
      • Schmich, on 07/02/2008, -0/+1Um when it comes to GFX I think Intel is the one who can be quiet.
      • RonBurgundy76, on 07/02/2008, -0/+1It's not nVidia's job to make their driver work with a particular game. If the game's developers can't figure out how to code properly, that's not nVidia's fault.
  • Steinr, on 07/02/2008, -0/+3The future is bright, for processor cores. They are multiplying faster than the rabbits.
    • Schmich, on 07/02/2008, -0/+1Yeh the problem has been going from 1 to 2 cores as multiplying 1 by 1 still gets you 1! It took them decades for that but now multiplying will get easier. Rabbits are going down.
  • m60dude5, on 07/02/2008, -1/+2I think we should work on getting the software to support multiple cores before making 1000s of them. It can't hurt, but it won't do much if nothing can utilize it.
  • seltaeb4, on 07/02/2008, -5/+3While no doubt true, this is basically an Intel press release that a mid-level PR exec probably encouraged the author to write up. "I can get you an interview with a [insert big cheese's name here]."

    Hey, writers have to churn it out, and that's why PR exists: "Hey, you have to write something, why not write about my guy? It'll look good in your portfolio. And if you ever get off the 9-to-5 kick and become a 'consultant,' you can say you know the guy. Hell, put 'has worked with chief technology advisor to market-dominating chipmaker Intel' on your résumé."

    So this particular piece of writing is mainly for the finance-side people, and is meant to keep AMD and others in line. It's not so unlike Microsoft's campaigns of FUD and announcements of vaporware, but the difference is historically Intel is to be believed, because they've always had the goods to back it up BEFORE they started selling the product.

    And that, boys and girls, is why you should have been shorting Microsoft stock all these years, and using your profits to buy Intel.



  • benologist, on 07/02/2008, -0/+2Ok, I'm prepared. Can I have my thousands-of-cores processor now please?
    • sanman, on 07/02/2008, -0/+1The catch is that there won't be enough energy to power it ;P
      • easypie, on 07/02/2008, -0/+0The catch is that in future there are going to be multicore nuclear reactors to power the processors.
    • funkywood, on 07/06/2008, -0/+1So you know how to parallelize all your code now do you? You're way ahead of most programmers then, who wouldn't know a mutex from a mutant.
  • Premier, on 07/02/2008, -7/+7So I'll finally be able to run crysis now?
    • sanman, on 07/02/2008, -1/+3That's why they're coming out with Crysis 2

      And also Windows 7
  • ufia, on 07/02/2008, -4/+10It's over NINE THOUSAAAAAAAND!
  • Dylson, on 07/02/2008, -6/+1ONE HUNDRED MILLION CORES!
  • malekov, on 07/02/2008, -2/+12Just like the brain. Trillions of cores (neurons) though they are a bit slow (due to electro-chemical limits) = great capacity of pattern recognition and don't generate almost any heat.
  • badart, on 07/02/2008, -3/+7Damn! Then my new laptop will really sear the flesh off my lap!
  • Jikul, on 07/02/2008, -2/+6So much cpu power... and we're still with mechanic hard drives. bah!
    • andrewtheart, on 07/02/2008, -4/+6We're slowly migrating to solid state memory, if you haven't noticed the trend.
      • Frost9999, on 07/02/2008, -0/+2Not sure why you're being dugg down. It's starting to happen and will eventually be ubiquitous. Spinning-platter hard drives are a huge bottle-neck in current machines despite all this multi-core goodness.
      • Schmich, on 07/02/2008, -0/+1I'm still waiting for the spherical hard-drive/memory.
      • eengineer, on 07/02/2008, -0/+0spherical? I would prefer a more gel based memory medium that I could mold to consumer goods. Think 500GB silly putty or squish ball drives.
  • CSharpSauce, on 07/02/2008, -3/+4I've been ready for thousands of cores for a while... of course I write enterprise scale server applications. Then again I imagine the cost of a thousand core processor would be far to high for a home user.
    • Pixelpaws, on 07/02/2008, -0/+4Weren't people saying that about the cost of quad-cores not long ago?
      • freebird09, on 07/02/2008, -0/+0And for most of us they still are. :(
  • andrewtheart, on 07/02/2008, -1/+8Too slow for me. I'm saving up for a simple quantum computer, which will make these sort of processors worthy of a McDonald's toy.
  • seltaeb4, on 07/02/2008, -1/+26Just in time for Duke Nukem Forever.
    • Voodrake, on 07/02/2008, -0/+2Why do you think they added Forever to the name?
  • tinus, on 07/02/2008, -2/+5How does this news relate to Mac OS Snow Leopard? Their introduction to Grand Central seems directly related to this story.
  • zlinger, on 07/02/2008, -3/+2Not quite powerful enough to run future versions of Windows. But for Win 95 it should be fine.
  • proficient, on 07/02/2008, -0/+4I would like to see fractional cores...
  • craighoxton, on 07/02/2008, -8/+3...and it STILL won't run Crysis
    • arjie, on 07/02/2008, -1/+1Is Crysis that multi-threaded that it'll actually be able to take advantage of so many cores?
  • hapax, on 07/02/2008, -5/+2Let a thousand cores bloom!
  • mooseysavage, on 07/02/2008, -6/+1fap fap fap
  • unluckier, on 07/02/2008, -2/+18That may be just about enough to run Symantec Antivirus without slowing the system to a crawl!
  • arjie, on 07/02/2008, -2/+1Ah, but I remember some outlandish promises they had when Netburst was still their favourite baby.
  • kiedesu, on 07/02/2008, -1/+5The next step: thousands of processors in one CPU
  • M724, on 07/02/2008, -4/+3Just when I bought a new computer, it's on its way to become obsolete...
    • unluckier, on 07/02/2008, -2/+1That is a tautology.
    • Pixelpaws, on 07/02/2008, -0/+4It was obsolete when you bought it. That's just how computing works.
  • goph, on 07/02/2008, -2/+6I just upgraded to core 2 duo, too

    *sniff*
  • YodaJones, on 07/02/2008, -5/+5I want some Giga-Core processors. I want my PC to get so hot I have to pipe the ocean through it. I want to water exiting my PC to be boiling when I dump it back into the ocean and cook any life forms that may be surviving the pollution we flush down there.
  • abrasion, on 07/02/2008, -4/+6***** Everything, We're Doing Five Blades^H^H^H^H^H^H Cores!

    http://www.theonion.com/content/node/33930

    • freebird09, on 07/02/2008, -0/+2Don't know why you're getting dugg down, had me laughing.
      • abrasion, on 07/02/2008, -0/+1Not sure myself, happening a lot lately - can't be entirely sure.
  • Drizzit, on 07/02/2008, -2/+8Why should software writers need to write for thousands of cores. It's inefficient. The cores should be able to see an incoming task and split it up between themselves. Let the coder worry about the App not how it's going to be processed.

    Apple and Sun have already developed this type of technology. I think IBM has as well.
    • iXam, on 07/02/2008, -0/+1It will be up to the compiler to break up the work to suit more cores.
    • rideagain, on 07/02/2008, -0/+2There is tech to do that, yes. The problem is that it doesn't work very well: if your program is not already designed in a way that makes it easy to split, then it needs to be changed so much to efficiently be run in parallel that current compiler technology just can't do it.
  • logdesigner, on 07/02/2008, -2/+1Damn, I just got VSO working on my duo core.
  • vanguardanon, on 07/02/2008, -4/+1Does intel understand that most computer tasks are fundamentally not capable of running in parallel (asynchronously)? Do they know that as you add more cores you get diminishing returns?

    Parallel Process:
    Let's say you're going to take a picture and rotate it 90 degrees. Imagine that picture is a big grid of dots. You need to take the top row and copy the dots from left to right to another location but you'll write them from top to bottom. Repeat this until you're out of rows and the picture will be rotated. Now it's a fairly easy task to break this into 4 threads by chopping the original image into quadrants, doing that same operation on each one, then putting them back together. It takes a little more setup but it probably saves time overall. However, all that setup becomes worse than the actual processing at some point. You wouldn't want to break your image into 750 pieces and reassemble them all, that's slower than just doing it in a single thread.

    Single threaded tasks:
    *Most* tasks are actually dependent on the previous task's completion. When I display a web page first I have to know who you are to personalize it. Then I have to retrieve the data from a database based on your entitlement. Then I have wrap that data in html/css/etc. so it's nice to look at. I can not do any of these things out of order because each task needs information from the previous one. You might give me 1,000 cores but I'll be hard pressed to use more than one or two.

    Let's focus on things that matter like memory, network, and hard drive speed.
    • relyk, on 07/02/2008, -0/+1Quote:
      Single threaded tasks:
      *Most* tasks are actually dependent on the previous task's completion. When I display a web page first I have to know who you are to personalize it. Then I have to retrieve the data from a database based on your entitlement. Then I have wrap that data in html/css/etc. so it's nice to look at. I can not do any of these things out of order because each task needs information from the previous one. You might give me 1,000 cores but I'll be hard pressed to use more than one or two.

      What about doing the single threaded task for 1000 connections/users? On a single core system the process goes: Get connection from UserA -> Query DB for User A -> Get Connection from UserB -> Create html/css/etc for UserA -> Query DB for UserB -> Present content to UserA -> Create html/css/etc for UserB -> Present content to User B.

      In a parallel world those two transactions that have nothing to do with each other could run independently.

      In the home user world I could have a single threaded processes (decompressing a zip file for instance) completely thrash one CPU while I use another for interacting with the OS.
      • funkywood, on 07/06/2008, -0/+1That's great for web farms but you're only ever going to have a few 'background' tasks to do like compression. Any more than about 20 cores is wasted on a home user.

        Compiler optimizations like Map-Reduce will help but code still needs to be written to enable it and it will only result in short bursts of parallel execution.

        Actually compressing a decompressing stuff in the background might have benefits but again the bottleneck now is memory.
    • yabos, on 07/02/2008, -0/+1So, does that mean they should just leave the processors stagnating in the future? Hell no. And with your problem and access to thousands of cores, you could just break the problem into every row in the entire image and have a thread doing the matrix transformation for each row.
  • luckyguy2000, on 07/02/2008, -4/+1with todays approach to spread tasks onto more than one core i fear these 1000 core processors will have a average load of 5% with everyday applications and games.
  • voodooray, on 07/02/2008, -3/+10Zzzzz... Someone wake me when you can virtualise all those corez into a single cpu that every application ever made can take advantage of without having had to have been specifically built to do so. Until then, this is nothing but a marketing exercise.
  • SirZRX, on 07/02/2008, -2/+1ill just wait for Quantum computers
  • ff1959, on 07/02/2008, -0/+1With this type of technology, the cost running Oracle will go sky-high. Good news for the MySQL folks.
    • RyeBrye, on 07/02/2008, -0/+1I think the "per-core" licensing people would be forced to change their pricing model to reflect the change in processing technology.
  • aserer511, on 07/02/2008, -0/+1how subjective is the term core? if intel structured a chip in a certain way, could they get a couple hundred 'cores' out of it cheaply, or is a core a very distinct unit?
    • rideagain, on 07/02/2008, -0/+1It's fairly subjective, so they have a bit of leeway. Generally it's understood that if you have X cores, then you should at least be able to do X things at the same time.
  • Panda200x, on 07/02/2008, -1/+2I still wish they would put the proccessor speed up to 5ghz or something
    • tolgafiratoglu, on 07/02/2008, -0/+1the aim of this new core technology is to develop processors in terms of parallel computing. In this new era don't expect great increases in frequency. Because they're in the limits of both size of transistors and heat.

      That's why they switched from increasing frequency to change architecture.
    • clickwir, on 07/02/2008, -0/+1Frequency has a point of diminishing return. Example, say from 1-1,000 MHZ, performance is pretty much linear. As the frequency goes up, so does performance. But at some point it starts to level off. Say 2,500 MHZ. You can still get a little more performance from frequency increase but as you reach 3,500MHZ you stop seeing performance going up. So, theoretically, you get just about the same performance from a 4ghz cpu as a 4.5ghz. So why bother going through all that trouble to make it work at that speed when in the real world, it doesn't get you anything worth your time.
  • caoilte, on 07/02/2008, -0/+5In other words, "time to learn Erlang folks!"
  • MatthewK, on 07/02/2008, -0/+3Intel might be preparing for raytrace rendering, they've invested a lot of money in it already. Perhaps they've found a solution to the memory bandwidth issue.

    Another possibility is they've finally reverse-engineered the Terminator's processor. Remember Dr. Miles Bennett Dyson?
  • blanktarget, on 07/02/2008, -0/+1I use 3ds max and more cores = more render buckets. I'd rather have 1,000 at 2ghz than 500 at 4ghz.
  • Culyt, on 07/02/2008, -0/+4I can imagine 'cat /proc/cpinfo' will take a while.

    Or the number of cpu entries you will find in ControlPanel>System>Hardware under windows.
    • clickwir, on 07/02/2008, -0/+1I would imagine it would have to be summed up by how they are broken down. If there are 2,000 cores, inside the chip they are grouped into 100 cores per group. And there's 20 groups. So maybe seeing 20 groups instead of 2,000 would be easier.
  • palewook, on 07/02/2008, -0/+4scalability to any amount of processors should already be common coding practice. it's not widely done though. should be, but it's not.
  • SquidLips, on 07/02/2008, -1/+2
    Java 1.7 (with the Doug Lea fork/join and work-stealing goodies) or possibily Erlang are the way to go.
  • Fetching more discussions...
    Show 51 - 66 of 66 discussions
  •  

    Login or register to add a comment

    Create a new account or login to join in the conversation on Digg. You'll also be able to Digg stories to help promote things you like.


© Digg Inc. 2008 — Content posted by Digg users is dedicated to the public domain.
DIGG, DIGG IT, DUGG, DIGG THIS, Digg graphics, logos, designs, page headers, button icons, scripts, and other service names are the trademarks of Digg Inc.