What is Digg?140 Comments
- inactive, on 08/04/2008, -5/+116MHz game ended in 06 when they figured out the P4 was teh suck
- ippey, on 08/04/2008, -11/+78So they'll effectively start the non-mhz game AMD did years ago?
- dilibau, on 08/04/2008, -6/+68if software is not going to be adapted AND FAST for multi-core processing, then we might have a problem Houston...
- marc123, on 08/05/2008, -1/+54this just in... study finds bears do in fact ***** in the woods
- SwabTheDeck, on 08/04/2008, -7/+43This is not news.
- Berkana, on 08/05/2008, -3/+36Hardware is already far ahead of software. The next chip may be a hell of a lot faster and here within a year, but the OSes that might take advantage of them best are still years off. Not a single popular OS out there (Mac, Linux, nor Windows) fully takes advantage of multi-core chips, and none are likely to thoroughly do so because there is so much legacy code.
In order to take full advantage of multiple cores, software needs to be easily and pervasively parallelized so as much of it can be simultaneously processed as possible (including the OS that serves the apps), but the only programming paradigm that makes it trivial to do pervasive parallelism is functional programming. (Think Scheme or LISP; every program is a tree in the form of a list.) In functional programming, the same parameters into a function always guarantee the same output (unless the function invokes a randomizer in its operation). Because of this, virtually every program can be expressed as a composition of functions, or a tree of functions. In *strictly functional* process trees, transitivities (dependencies) are very clearly built into the structure of the function, with each branch of the tree independent of (and therefore capable of being processed in parallel to) all other branches. Parallelizing imperative or OOP/imperative code is non-trivial and very labor intensive, and ends up being a needless re-implementation of functional programming, with transitivity manually tracked and parallelism manually enforced. (Correct me if I'm wrong or missing something; this has been my experience.)
Even OOP can be done in a functional manner, as OCAML and Scheme/LISP OOP systems demonstrate. Unfortunately, OOP is currently not done in a functional manner; modern OOP is an extension of imperative programming methods that came to dominance with the C programming language, including C++, Objective C, etc.
I really hope the rise of multicore processors will make computer scientists and developers give functional programming a second look. I suspect multicore architectures will not be used to their full potential outside of supercomputing environments unless functional programming becomes popular again. I just don't see manually designing and enforcing parallelism and manually separating sequences of transitivity in imperative code becoming widely popular; coding is hard enough as it is. - TwiceHephaestus, on 08/05/2008, -0/+26AMD lost so much ground around the same time they merged with ATI (and for a while, so did ATI). Now, I used to be a diehard AMD fan, but frankly, intel got their ***** together, and AMD no longer has the edge.
By the time they get a tech to the market, Intel has had it out for 6+ months, and is a week away from something new. All AMD can do in response is cut prices. What used to be my high end gaming vendor is now a discount processor company? - jonathandyer, on 08/05/2008, -0/+20He is referring (I believe) to how AMD labels their CPU's as in a 3800+ instead of a 2.2GHz. But I may be wrong with that.
- pauls88, on 08/05/2008, -0/+16I think my macbook is close to nuclear reactor heat, I've had to start wearing fire proof pants!
- Urzeitlich, on 08/05/2008, -3/+19They'll declare that their new CPU can scale up to 10 cores on one chip, but then stop at about 4 and go back to the "MHz Game" =)
Oh prescott. We don't miss you! - fraglessone, on 08/05/2008, -2/+16But you need the nuclear reactor to generate the 1.21 gigawatts.
- ineptsavant, on 08/05/2008, -1/+14Washington post is on the ball with news from several years ago.
- hollyminkowski, on 08/05/2008, -0/+12To keep up with Moore's law they will have to go lower V+, multi core, 3d layering, optical inter chip data handling.
I bet in 15 years the cpu will be a cube like thing with cooling pores sitting in a tiny liquid cooling container on a mini motherboard connected electrically only to the V+ and ground. V+ will likely be close to 0.7v and there will be no bus wires on the board...it will be pretty much all optical.
A cubical 3d cpu package will need to have cooling fluid slowly pumped through an intricate network of pores or channels to keep cool...at least this will be silent and cheap since the pump could be very tiny and simple. The fluid would form a continuous loop through the cpu and a heat radiator of some sort. The cpu would likely come as an integrated package with pump, coolant and radiator ready to stick onto a motherboard.
Cpu core voltage has been dropping steadily from 12v to about 1.2 today... 0.7 is about the limit given today's technology...but that might change, and with every reduction comes a lowering of power use and generated heat...heat is the biggest problem we have in cpu design. - EBFoxbat, on 08/05/2008, -2/+13He would, undoubtedly, be much funnier than that.
- dewbieZ, on 08/05/2008, -0/+10yes because CPU's make so much noise
- boobsbr, on 08/05/2008, -1/+11athlon xp FTW!
- mynameistux, on 08/05/2008, -2/+12The MHZ game finished, its the GHZ game now.
- MeatPlow, on 08/05/2008, -3/+13If a bear ***** in the woods and nobody is around to smell it, did it really *****?
- mikephimikephi, on 08/05/2008, -5/+14Or as Jon Stewart would say...
They're planning on playing the 'many core card' - Metasquares, on 08/05/2008, -0/+9It will still give you a performance benefit across multiple applications.
- jamesdew, on 08/05/2008, -0/+8I want both
- Elranzer, on 08/05/2008, -0/+8To be fair, AMD has always been less pricey than Intel. This part about the Intel-vs-AMD battle has been steady since the 1990s.
- Stupidumb, on 08/05/2008, -3/+11So what you're saying is that faster processors are faster than slower processors. Is that what your trying to say? Because P4 "was teh suck"? Single cores are slower than multi-cores, so you disagree because higher speed multi-cores are faster than lower speed multi-cores. Right? IS THAT WHAT YOU MEAN?
- gurellia53, on 08/05/2008, -3/+102002 called. It wants it's headline back.
seriously, why would anyone still publish this "news"? - caseycoold, on 08/05/2008, -0/+7I'm at RIT where microelectronic engineers are everywhere. One told e that the holes int he processors for cooling are already possible, but expensive (something about how their made). So I think you are right on.
- hexydes, on 08/05/2008, -0/+7I'm assuming this is about when we'll start moving from electrons to photons to transport and process data. It's still a ways off, but it's definitely on the way.
As the article said, Intel (and AMD) pretty much squeezed all they could out of raw processing power with electrons. Physics began to dictate that in order to have a faster CPU that still functioned it would have to use water cooling and have a radiator the size of a car's. It just wasn't practical.
So now they're moving to multiple cores. This is probably something that should have been done along the way since it just makes sense to use multiple cores in an architecture, but companies like Microsoft had little interest, and Intel and AMD were busy having their MHz/GHz pissing match. Now that we've hit the wall with raw CPU processing speed, they're moving to multiple cores. Eventually we'll hit a point where that has diminishing returns as well (you can only subdivide process executions so many times), and once we hit the point where we can't process any additional processes (no more multi-cores), and can't do it any faster (no faster CPU speeds), then we'll be faced with a full architecture shift.
It will happen sooner rather than later. I mean, at some point you're running 128-256 cores at 3-4GHz; you're not going to be able to subdivide tasks into anything smaller than something like that, and you won't be able to make the cores run any faster because they'll be too hot. At the pace we're moving, I would bet that is probably no further than 7-12 years away, at which point hopefully Intel or AMD have something up their sleeves. You have to wonder though, do we hit a point where we don't need more processing power? With that type of power you can edit probably twice the resolution of HD video in real-time. Of course, we've been asking that question for a long time, and inevitably something comes out that demands more power. :) - benologist, on 08/05/2008, -3/+9We won't have a problem as such, it just won't take full advantage of the cpu. They might have a problem since without full support there's little incentive to upgrade.
- railwave, on 08/05/2008, -0/+6Hold on, didn't Sun already do this with the 8 core, 64-thread UltraSPARC T2: http://www.sun.com/processors/UltraSPARC-T2
- keviniskool, on 08/05/2008, -0/+6Legitmate Computer talk? I thought this was 4chan.
- hexydes, on 08/05/2008, -0/+6I really feel that AMD has a long-term plan with the ATI acquisition. I think they knew that in the short term, it was going to REALLY hurt the GPU division (as seen up until the release of the 48XX series which soundly competes against NVIDIA), and because of the resources required, the CPU division would take a hit as well (as they have in the last two years vs. the Core 2 Duo). I think now that AMD has the graphics division back on track, they'll be able to get back to concentrating on their processing division, use some of the technology that came from the graphics division, and create some very competitive technology.
Additionally, processing is starting to hit a wall, where most applications just don't need more processing power. However, the entire market is starting to move to more embedded devices (HTPC devices, smartphones, etc), and this is where the technology is still very behind, and I think a huge market is going to open up here. That's why Intel, AMD, and NVIDIA all have solutions for this industry in the works. - ParanoydAndroid, on 08/05/2008, -2/+8... Your comment just made me a little dumber than I was before ... thanks.
- solidus636, on 08/05/2008, -0/+6Wrong article or are you just an idiot?
- tugger, on 08/05/2008, -2/+7So the CISC processors ran out of steam, ARM and IBM are taking the [relatively cool running] RISC processors to higher and higher processor speeds, and say that can do so for around another 8 years at least.
Multithreading can only work to a point, there's only so much core sharing you can do in an application before the law of diminishing returns applies. Intel will need to have something on the back-burner or they will end up like motorola. - Dgen_X, on 08/05/2008, -0/+5Jigga what?
- MrFurious2k, on 08/05/2008, -0/+5Multi-core has been quite beneficial to the server market, but so far, not so great for the single user workstation. Until we can really get a handle on utilizing multiple cores, we're going to continue to have one core doing all the work, while the other(s) sit around... sort of like a highway project. Maybe the other cores can wear orange vests and whistle at the hot instructions that come by.
- Aero347, on 08/05/2008, -0/+4I want faster =(
If I wanted quiet i'd use a bike instead of a racecar. - Morrison1002, on 08/05/2008, -1/+5no, you will need to hook up a wire to the clock tower and waite for the lightening to strike
- inactive, on 08/05/2008, -0/+4 They still have some tuning to do. If one core on my 6600 quad hits 100% then the computer momentarily freezes just like it would with a single core. In theory the others should pick up the slack but they don't.
- hollyminkowski, on 08/05/2008, -0/+4Thank you caseycoold, cant tell you how good it makes me feel that my speculations are right on.
- hollyminkowski, on 08/05/2008, -0/+4caseycoold... if there is ever any question you would like me to consider please ask. I would love it if it was passed on...no attribution is necesssary. I just want to see the future unfold as quickly as posssible, :-)
- mem2, on 08/05/2008, -0/+4my god you guys do drivel on. A simple solution already in use by many game developers is you split your application.
Now 1st it is only benefit to very large and complex programs.
Secondly, abstracting over the cpu to provide a single fast cpu again (or software layer or whatnot).... piss off
You simply break up your large program into daemons/threads/(your terminology here). A 3d game for example, you have your sound daemon (possibly threaded again - programmers call). The 3d engine, the phsyics and so on etc.
There are still obvious limitations to using ridiculous amounts of cpus in terms of real time preforms. eg: anything that has to happen in sequence.
Its the way its going to and is happening. As I heard in some cool Megadeth lyrics "If theres a better way I'll be the first in line". - bigsteve, on 08/05/2008, -1/+5Oh the mhz game, Intel, you played it so well. I remember selling computers for Best Buy 8 years ago, trying to tell people that this new AMD Athlon, despite it's lower price tag and lower clock speed could run circles around that Pentium 4 1.8ghz.
- hollyminkowski, on 08/05/2008, -2/+6To reach their full potential massively multi-core processors will need a layer a separation between the programmer and the cpu. The layer of separation will be software that reduces the incredible complexity to the point where the cpu can be considered by the programmer as the equivalent of a super powerful risc cpu with only a few dozen instructions. The layer would have to handle the complexities involved with making good use of so many cores. Little progress can be made until this separating layer is built...there simply are not enough man-hours available to deal directly with the complexities of massively multi-core designs. To the programmer it needs to be as simple as programming a single-core cpu, albeit one able to handle hundreds of billions or even trillions of operations/second.
Creating this layer of separation will be much more difficult than writing an ordinary compiler or OS of course, and the programmer will need to keep in mind that a multitude of things can be happening at the same time within the processor. - mynameistux, on 08/05/2008, -0/+4the sad part is, he is right.
Gone are the days where you could just make a joke.
sigh. - ChildeRoland420, on 08/05/2008, -0/+3@Gonthim,
You mean an OS? - Berkana, on 08/05/2008, -0/+3True; what I'm saying is that when your programming language naturally supports code that is parallelized because of its structure (compositions of functions) this layer becomes much easier to implement.
One real-world example of what I mean when I say that functional programming makes parallelism much easier is Google's MapReduce algorithm. Google's massively parallel and scalable MapReduce algorithm is based on functional programming methodologies:
http://renil.wordpress.com/2007/05/03/mapreduce-fu ...
Google implemented the algorithm in C++, but in so doing the basically re-implemented a very fundamental concept from LISP. The same is true of many very old features of LISP that are slowly making their way into C++, such as passing functions as data (first class functions). These ideas are not new, they've simply been overlooked because imperative programming and C came to dominance as UNIX came to dominance while LISP machines failed to keep up.
With lambda calculus (and therefore, functional programming, which is based on lambda calculus) being Turing-equivalent, there is nothing (algorithmicly speaking) that can be done in other programming languages that can't be done in a functional language, but by being done functionally, the same problem becomes trivial to parallelize. - NoCt1, on 08/05/2008, -3/+6Agreed. Thinking its going to be good as well.
- mem2, on 08/05/2008, -0/+3why not 1 core per keyboard key ? :D
- Metasquares, on 08/05/2008, -0/+3We've been around 2.2 GHz or so for quite some time now.
- deviationer, on 08/05/2008, -3/+6sorry but more ***** cores will NOT SOLVE ANYTHING!
why? because most software and games are still not multi-threaded. -
Show 51 - 100 of 141 discussions
Check out the new & improved 
© Digg Inc. 2009
— Content created and posted by Digg users is