What is Digg?Sponsored by Travelzoo
All-time Low Fares for Thanksgiving, Christmas & New Year view!
travelzoo.com - Flights $52 and up. Nifty all-airline calendar identifies absolute cheapest dates to fly.
91 Comments
- bhavi, on 10/12/2007, -7/+65All these cores will only be useful when software actually make efficient use of it. Without that it's a meaningless rat race between the processor companies.
- wingnut21, on 10/12/2007, -13/+69The PS3 is an excellent example of this.
- sysoprock, on 10/12/2007, -0/+40Countdown to when AMD counters with the "it's not about the amount of cores" argument and releases the AMD 90Core, featuring 20 cores.
- Roger, on 10/12/2007, -8/+46The OS should take care of everything.
The software shouldn't have to care about the number of cores (besides using threads). - colincornaby, on 10/12/2007, -1/+29"The OS should take care of everything.
The software shouldn't have to care about the number of cores (besides using threads)."
Technically this would be quite a feat. OS's today can automatically run different apps in different threads (and if you look at the number of processes normally running on your computer, this is still pretty handy). But so far there is no magical way for a program to be threaded automatically over a bunch of cores. This is one of the big issues with the PS3. The Core is so hard to optimize for that a lot of developers don't like the Core. - Roger, on 10/12/2007, -2/+30Will your ass blend?
- wang1011, on 10/12/2007, -1/+26What will they call it? Intel Core Octoginta?
- inactive, on 10/12/2007, -2/+21It is. 2^6.3219280948873623478703194294894ish.
- diggfinity, on 10/12/2007, -2/+20"Why buy an Intel Core 100 for $10,000 when I can buy an AMD 90Core for $6,000 and overclock it to 150Core?"
- phlux, on 10/12/2007, -0/+1647 posts and not one person here has any clue what they are talking about.
My friend works on the team that did this and he was in charge of the release...
We went to hike up in Saratoga today and I got to ask all sorts of questions about this thing. This isn't just simply more cores. And more specifically this is a research chip that is for the purpose of testing the routing mechanisms within it. The other important aspect of this chip is that it reaches teraflops in 62 WATTS. It was originally predicted that reaching a teraflop on a single chip would cost 1000 megawatts, then that number was brought down a bit - but it was never predicted to be so low.
The original design called for 64 cores - but the energy cost was too high - meaning that they couldn't reach a teraflop on chip under 100 watts. So they opted for more weaker cores.
Additionally the cores can be put to sleep if they have no work packets - resulting in lower power consumption based on whats being computed.
The beauty in a chip like this is not in the number of cores so much as in the ability to route work to any number of cores to meet the needs of the job.
One of the interesting things we were talking about today was the future of overlaying a meta computer onto large scale multi-core systems. Whereby you have many hundreds of cores available and you grab as many as your job needs and you form clusters of job cores. This came from the thought of just having a large large number of transistors that you would dynamically form into a core of the right size when needed. (my thought) - but Sean said that this was essentially what was being done with large core systems.
In the standard multi-core systems you basically just have two separate procs connected over the same system bus as you would a traditional multi-proc box. In the modern case though the cores are on the same board.
In the new chip talked about in this thread - the cores have a router built in - there is communication and switch fabric on die so that the work of routing information just happens on the chip. This means that you can have a very high number of cores of lesser "speed" resulting in much faster computes because all info routing happens on die...
anyway - I am still milking Sean for info on this. We will see many hundreds of cores moving forward. Wait till all those interconnects are optical.
As I said today: "So all these cores are great - but how long till we get fem-bots?" - jm1234567890, on 10/12/2007, -1/+16You mean the Cell?
- Roger, on 10/12/2007, -0/+14Surprised its 80 and not a power of 2.
- etnu, on 10/12/2007, -2/+13I'm sick of this argument!
Many programs already "take advantage" of multiple core CPUs! Multi-threading isn't some mythical beast that only the most expert of programmers can perform -- it's actually quite common. Virtually every major piece of software on the market today in common use is multi threaded. Most server daemons are multi-process, multi-threaded, or both.
YES, writing thread-safe code can be difficult, especially for novice (or incompetent) programmers. That doesn't mean it's impossible. If you write your code in a clean manner, making it thread safe is actually pretty easy. Thousands of programs are already out there that are multi threaded or multi process, and the multi core systems yield huge performance improvements in most cases.
You don't even have to look for server processes to see where multi core is productive. Most of the common desktop applications in use today are multi-threaded:
- Firefox
- Internet Explorer
- Windows media player
- WinAmp
- iTunes
- MS Office
- OpenOffice
- Eclipse and all the IDEs that are derived from it
- Most games released in the last 3 years
I could go on and on. The only people who still complain about how "hard" multi threaded programming is are the same people who can't code their way out of a wet paper bag, and/or people who don't know what they're talking about in the first place. Yes, making multi-threaded systems is more complicated than making single threaded systems -- but it's not that big of a step, especially if you're using a system like Java or C# that has thread-safe training wheels for you anyway.
PS3 comparisons are idiotic. The 360 is not a single core system, and it's doing remarkably well. The PS3's shortcomings have nothing to do with it having multiple cores -- it has failed because it's way too ***** expensive relative to the games that are available for it, period. If Sony had put the thing out at the same price points as the 360 and/or put out a bunch of "must have" exclusive games, it would have sold fine, but sony would probably have gone bankrupt by now. Any other arguments are uninformed at best and fanboyish at worst. Don't blame the concept of multi core for Sony's colossal ***** up, it makes you look like an idiot. - econoar, on 10/12/2007, -6/+14Octopus X 10
- Vigile, on 10/12/2007, -2/+10A lot more technical info and less fluff crap on this one: http://www.digg.com/hardware/Intel_Terascale_CPU_does_4GHz_with_80_cores
- inactive, on 10/12/2007, -3/+10I'm going to go out on a limb...
Yes. - Vigile, on 10/12/2007, -0/+6ANyone else sick of Engadget repots up here already? Talk about asking to link to the SOURCE, not a different blog/news site...
- steepdecline, on 10/12/2007, -1/+6look at the pic... they're calling it the 'Eye of Sauronium'
-- according to bigbignews.net - inactive, on 10/12/2007, -2/+7@ johngalt
Ahem... http://www.top500.org/list/2006/11/100
Almost all these run Linux. - Charlotte_Web, on 10/12/2007, -0/+4My (limited) understanding of multi-core systems is that you have a problem of increased overhead with each additional core added on. That is, you have to spend more processor time managing traffic to each of the different cores, and then reassembling the results on the other end. I also seem to recall from an Ars Technica article I read way back when, that another issue would be with data that is broken up among several cores, but the cores don't all finish at the same time, forcing everything to wait (though the memory cache size determines how big of a problem this is).
The end result being that you have the problem of diminishing returns; the more cores you add to a processor, the less horsepower you get per core. As the article indicated, early results seemed disappointing. - browwiw, on 10/12/2007, -0/+4But will it run Oregon Trail?
- sysoprock, on 10/12/2007, -2/+6@diggfinity
"Dude, I totally got one of those Celecore 50s and soldered like 75 more cores onto it, my performance is through the roof!" - coldfusion055, on 10/12/2007, -0/+4Well there goes my brand new "future proof" Core 2 Duo...
- jub0r, on 10/12/2007, -7/+11Imagine a Beowulf cluster of those!
What? This isn't Slashdot? Where am I? - davidrools, on 10/12/2007, -0/+4Octogenitalia...that would be very interesting
- inactive, on 10/12/2007, -0/+3"I think processor technology isn't really going to take off until we see silicon replaced with synthetic diamond plating for the semi-conductor properties."
Yeah, you're right. Processors are nothing more than a hobby people use in their garage. I can't wait for a revolutionary way of switching electricity into different states that is small and affordable. We'll really see uses for processors then. - Wisgary, on 10/12/2007, -1/+4I was just about to post a beowulf cluster joke while thinking of slashdot. Ahh the memories. I left slashdot once I got tired of the ***** "editors".
- DyDx, on 10/12/2007, -0/+3@bhavi: There is plenty of software that will take care of all these cores, and then some. There is a reason that computer manufacturers have been selling gigantic server clusters for a long time -- in any field requiring massive amounts of computation (analyzing DNA microarray results, for example) will benefit from them.
It turns out that a lot of this software is custom-built for the application, but there is still definitely a need. Consumers will almost certainly never need more than 2, or 4 processors.. unless the whole new multithreading paradigm that Valve Software came up with turns out to work well and catches on. As I understand it, making practical use of multiple cores is still very new to most programmers, and difficult to achieve. - etnu, on 10/12/2007, -0/+3The MHz argument was stupid to anyone who knows anything about how microprocessors work. Saying that a 5Ghz processor is automatically better than a 4Ghz processor is like saying that a 1000 Thai Bhat is better than 100 U.S. dollars, because 1000 is greater than 100.
- BlackAdderIII, on 10/12/2007, -0/+2News just in, computers with large numbers of cores DON'T run windows XP - let me repeat that, DON'T run windows XP. This is just incredible.
In other news, shock revelations that Red Storm was NOT built to 'play a little counter-strike and go on AOL messenger'. - mr804, on 10/12/2007, -0/+2DyDx:
I don't think valve came up with anything new. They just used hydracore by Havok.
Read more here http://www.havok.com/content/view/25/46/ - inactive, on 10/12/2007, -0/+2@johngalt:
SGI sells systems that "support up to 512 processors under a single instance of linux" so i'm pretty sure you can have a linux system running on 80 cores...
http://www.sgi.com/products/servers/altix/4000/ - BlackAdderIII, on 10/12/2007, -0/+2"All these cores will only be useful when software actually make efficient use of it. Without that it's a meaningless rat race between the processor companies."
Don't worry - systems of this size are pretty much geared toward making use of multiple cores, and so are the operating systems and programs run on them.
@etnu
I really mean no offense, but you seem a bit confused about the nature of the problems at hand.
Confused in a way that makes it hard to believe you're a programmer who understands WTH they're talking about. Not to insult you for that, nobody knows everything(I don't, that's for sure), but consider doing some research before trying to dispel myths you see floating around. :-) - maninblac1, on 10/12/2007, -0/+2I'm still impressed, if they "really" got 1 Tflop out of this chip. Most desktop CPU's are in the 15-30 Gflop range. Of course for any of this to be feasible memory is going to have to come up to speed, no way you could service a 1 TF chip with current memory technology. Intel will have to move to an integrated controller like AMD to make it even feasible.
Though really, if they can't hook it up to memory, how can they speed rate it!?
Anyways, 5 years, just in time to replace my current setup :-) - BlackAdderIII, on 10/12/2007, -0/+2"""The MegaHertz Myth was a myth to whichever processor maker was trailing behind in the MHz race at the time."""
No, it was, and is, a myth.
A processor is a machine, the clockspeed is the rate at which it runs.
To take an extreme example for clarification, if you took the processor out of your average digital watch and were able to ramp up the clockspeed to 3.5 gigahertz, this would not magically imbue it with the capabilities of a 64 bit Turion, or make it anything like as fast.
A machine is a thing designed to reduce required effort, and the thing which affects that most is *how it works*, not the speed at which you run it - you can spend all day trying to do something using more power or faster iterations which a better designed machine will do in one go. That's why you can't compare machine guns with crossbows in terms of trigger-pulls per second.
The MHz myth is absolutely that, a myth. The only thing a processor can be assuredly faster than with a clockspeed increase that leaves it still working properly, is itself with the lower setting. - atbnet, on 10/12/2007, -0/+2Just to clarify, AMD didn't label the XP line in accordance to what their Intel equivalent would be, they labeled them in accordance to what the Thunderbird socket-A class would have been at. Back then a XP-1800+ was 1.4 GHz itself, but would equate to a thunderbird at 1.8 GHz.
And the reason for the change is because they can only cram so many circuits on the chip so MHz starts to be less of the deciding factor. - inactive, on 10/12/2007, -2/+4"You don't seem to understand that pushing the clock speed envelope would be more expensive than adding more cores, right now at least."
Wrong. Synthetic diamond is as cheap to manufacture as silicon. Now convincing companies to abandon silicon might present a bit of a start up cost, but it'd be a VERY quick way to jack up clock speed to astronomical levels. - grumpyrain, on 10/12/2007, -0/+2It will actually be better though. Consider the reality that your processor cores may outnumber the number of processes you have running. If that is the case, each process can have a (or many) cores at any time it needs, effectively a real time process level. If some inefficient process goes awol, it is only 1 core down. Of course the universe will eventually hit back with idiot programmers to do the same inefficient work with multiple cores.
- JavertHolmes, on 10/12/2007, -0/+2"Just to clarify, AMD didn't label the XP line in accordance to what their Intel equivalent would be, they labeled them in accordance to what the Thunderbird socket-A class would have been at."
Thanks for the correction. I sincerely did not know that. - Philodox, on 10/12/2007, -1/+3So Intel has replaced the megahertz myth with the multicore myth? Sounds like Intel just doesn't get what making processors is all about.
- phlux, on 10/12/2007, -1/+3@sparkey,
its not about being smug - its about learning something. Try it. Although your smooth cortex probably cant handle any thought that requires more than a fleeting amount of concentration...
If you noticed when I posted that - there wasnt a single post on here that had any info on this chip.
Here this link, then, is for you: http://arbitrary.name/spheres.jpeg - ngsayjoe, on 10/12/2007, -0/+2A parallel programming language such as Verilog, VHDL, could solve this problem. But the implementation of such language in the software engineering paradigm wouldn't be that easy. I hope OS can take care of distribution jobs to cores, where software just need to be multi-threaded, or parallel programming language.
There's some tests done now that multi-threaded applications running on multi-cores (4-core) CPU is slower on Windows. The reason could be Windows threads synchronization is slow, or it just never distribute threads to the corresponding cores correctly. Has anyone got knowledge on Windows threading implementation? - devinbunker, on 10/12/2007, -2/+3@etnu
Multi-threaded programming for an 80-core system wouldn't be difficult? Who are you kidding?
Sure, tweaking a web browser to have a separate thread for DNS lookups isn't too hard, but getting a web browser to efficiently parallelize across 80 processors is VERY hard. It isn't just a matter of worrying about thread synchronization -- which, despite your objections, is a very tricky problem that bites even the most experienced programmers in the ass -- it's also a matter of figuring out how to take your program and make it linearly scale with the number of processors available. - mgajda, on 10/12/2007, -2/+3Just another vaporware.
Tell me when they'll come up with a similar product (if ever).
PS This CPU may be not all that good given the constraints of memory-CPU transfer. - Himself, on 10/12/2007, -1/+2...... GRID-on-chip
- ngsayjoe, on 10/12/2007, -0/+1@chungthomas
>>>VHDL is strictly speaking not a programming language, it is a Hardware Description Language
Yes, that's what I'm talking about, I never said it was a software language. I said we said borrow the concept of the idea and implement it for software programming. Hardware are parallel in nature, whereas software has been sequential since the beginning. But we can totally change that mindset, and think in terms of parellism.
So how does hardware synchronize when there could be so many parallel components running at the same time? Well, the answer is CLOCK. So implementing a new programming language similar to that of Verilog could solve multi-threading problems. But the problem here is languages such as C/C++, Java, can easily be implemented is because they map directly to the processor instructions. On the other hand, there's no way you can do the same with Verilog as many necessary instructions required to do parallelism are missing. So, you basically need to create a virtual machine to get around that.
So, here's what I'm talking about. Create a parallel language, and create a virtual machine, and BANG you have a software programming language for multi-cores without worrying about thread synchronization and stuff. I feel if multi-cores is the future, then a new programming language is imminent! Just a matter of time. - littlebylittle, on 10/12/2007, -0/+1Can we just get on to "Infini-Core" and be done with it already. I'd like true Light Speed guts with that too (Motherboard, Bus, etc.). Thanks.
- phlux, on 10/12/2007, -0/+1@the Sauce - actually its the reverse... these are specifically floating point cores.
- thesauce, on 10/12/2007, -1/+2Also, these cores I believe are only capable of integer calculation. With no FP calculation, they aren't nearly as powerful.
- maninblac1, on 10/12/2007, -1/+2How can it be vaporware when the device actually exists and can be touched and held?
Read the full article. -
Show 51 - 91 of 91 discussions
The Digg Toolbar for Firefox lets you Digg, submit content, and keep track of Digg even when you're not on the Digg site. Download the official 
© Digg Inc. 2009
— Content created and posted by Digg users is 