What is Digg?85 Comments
- nunogt, on 10/12/2007, -8/+66I think AMD got it right. Increasing the number of cores increases the code complexity. It's not uncommon to see games or encoders performing worse in dual core machines than in single core CPUs. And in quad core architectures, such as AMD's own QuadFX (former 4x4), the performance tends to be even worse. This is also true for SLI and Crossfire (which I guess can be considered a multicore-GPU setup) - if a game doesn't include optimizations for it, the performance increase is negligible, and may even be worse.
It probably makes more sense specializing certain parts of the CPU to perform specific complex tasks than to add an insane number of cores to it. - geminitojanus, on 10/12/2007, -2/+40"Increasing the number of cores increases the code complexity."
On the other hand, AMD has been stuck in this position because they simply cannot keep up with Intel on this war, and they know this, way ahead of time: AMD doesn't have the money nor the factories to build multicore chips at extreme rates, and so instead of trying to compete with Intel, they're going in a different direction. It's a good idea for them, as the technological components are all there (HyperTransport for meshing said cores, ATi and several embedded players for various functionalities), but realizing this idea is going to take a lot longer than Intel's current game-plan of "Let's see how many cores we can put on one piece of silicon".
Of course, both companies understand that the future of computing is Asymmetrical cores, but Intel intends to get there on one piece of silicon after they've exhausted current options (see Terascale for more information), and AMD intends to get there on multiple pieces of silicon connected by a high-speed bus, which is another one of those odd reversals in strategy; Intel usually prefers the solution that allows them to sell more individual chips, and AMD typically goes for the integrated approach. - Llanowar, on 10/12/2007, -2/+29Right now Intel has better products than AMD.
Let's hope AMD's new view will get them an advantage again. - cquinnd, on 10/12/2007, -1/+27The truth is, every approach that provides some performance gains eventually will hit a wall. Then you will see the companies switch over to another method to keep the pace going.
- msgyrd, on 10/12/2007, -1/+22"Increasing the number of cores increases the code complexity."
Completely correct. Writing good threaded applications is a PITA.
Also, while quad core has amazing processing potential....as I look at my CPU meter that sits on my panel, I notice that most of the time my CPU usage is well below 5%. Sure, if I power up VMware and start a couple OSs, then start a movie and then open up some rendering application, I can max out both of my cores, but really, even then, most of my performance limitations stem from a limited amount of memory, hard drive speed, and my ability as a human to focus on a task.
Performance you say? Give me solid state hard drives. - nozol03, on 10/12/2007, -1/+12"The Dual core is similar to Internet."
Worst....analogy.....ever. - MatB, on 10/12/2007, -1/+11The problem with specialized cores is context switching. Think about the cell processor systems, where context switching requires ages. Good for single applications, but gets huge penalties in generic multitask environments, where also the specialized cores get shared.
Something on this line, was considered with the introduction of the MMX extension, which share the registers memory with the FPU. - inactive, on 10/12/2007, -3/+11Not sure why you're getting dugg down. AMD really screwed Intel till the core 2 came out!
I guess Intel had bought so deeply into the "Mhz at any cost" netburst mantra in order to trick consumers that their CPUs were faster that it took them a couple of years to come up with a processor that was able to compete, and a damn fine processor at that!
I sell lots of AMD based systems, and quite a few Intel ones, the number of Intel systems is steadily increasing though, and will keep doing so till AMD comes up with an answer to the Core 2. Hopefully the wait won't be too long! - Desolite, on 10/12/2007, -0/+6i didn't digg him down personally, but i'm guessing its because he used the non-word "pizzwn."
- sporkmonger, on 10/12/2007, -0/+6This is probably one of the main reasons they bought ATI. And I wouldn't be surprised if it is the best strategy for AMD going forward, since Intel clearly already has a leg up on AMD in the multi-core arms race. However, they're once again going to have to break out the old defensive marketing campaigns that explain the "multi-core myth" about how having more cores doesn't necessarily mean you have a faster CPU. Not a good position to be in, IMHO, even if they do end up making faster CPUs as a result.
- omatsei, on 10/12/2007, -1/+6"Don't also forget that you have to run as administrator to run AutoCAD. "
No you don't. We have an entire lab full of computers with AutoCAD on them, and the users are only users. You only have to enable the right permissions on the right directories. - geronimo, on 10/12/2007, -1/+6Maybe for the consumer but for the server, the more cores the better.
Take a look at intel's 5355 monster:
http://www.spec.org/jbb2005/results/res2006q4/
208968 (2 sockets) vs 119641 for the opteron 885 (4 sockets). And it's much cheaper. On the server end it makes little sense to go with AMD right now. Keep in mind I really like AMD but I don't let that bias affect my purchasing decisions. - geminitojanus, on 10/12/2007, -0/+5The distinction is symmetrical vs asymmetrical; having many of the same core, or a few different cores, on one chip. AMD plans to go with the latter, but in order to get there, they've got to first make chips independent of each other that all talk on the same bus, so that they can be integrated into a single product cohesively. Intel, on the other hand, intends to build what we call "multicore" symmetrical chips until the asymmetrical designs are completed and finalized, then start shipping those and phase out the former.
Both companies are on track to get to the same station, but both companies are on two different rail-ways to it. - geminitojanus, on 10/12/2007, -2/+7I wouldn't say AMD is a step ahead, they're just looking at the game-plan differently. Both companies see that the x86 architecture is starting to show its age, and that the various components of it simply aren't working; SSE, despite its increased complexity and increasing speeds, is just now getting the kind of attention from compiler developers to make it a worth-while addition to the processors, at the same time, graphics card companies are now starting to pull developers on-board for their generalized cores. This obsoletes SSE. Meanwhile, the traditional pipeline simply isn't getting any faster, and the only thing left to do is throw more cores at the problem.
This is where you can start to take two different looks on the problem: AMD sees multicore as a convergence technology; as you get all of the pieces of the puzzle together, they will slowly converge and collesce into a platform, and you let the customers decide how that happens. Intel, on the other hand, sees multicore as a race, and thusly is going to try its hardest to quickly speed up its own cores, while keeping them thin and lean, and sticking a bunch of them together, sharing a giant cache (instead of building lots of itty-bitty individual caches).
They're both working towards the goal of putting a core for your application on their chips, but AMD is getting there quicker using technology we have today, while Intel intends to get there using highly developed technology of tomorrow, and all-at-once, on one piece of silicon. - msgyrd, on 10/12/2007, -0/+4Did you guys even read the article or just posting based of the Digg submission?
AMD isn't saying multiple cores isn't the future, they're implying that adding tons of symmetrical cores is pointless, and they're focusing on asymmetrical cores. One look at the linked article will show you they aren't regressing to a single core architecture.
You people are what make Digg comments lame. - inactive, on 10/12/2007, -1/+5A program needs to be multi threaded to get the most benefit of having multiple cores, however even if a program is non multithreaded they will still get a boost as the different cores are managed by the OS to perform different things.
I went from an Athlon 64 3800+ to a X2 4200+. The 4200 has 2 cores that run at the same speed as a 3500+ Athlon, so in terms of pure processing power of a single CPU the 4200+ is slower than my old 3800, however pretty much everything runs faster as the OS 'divvies up' the workload so (for instance) windows will mostly run on one core, while your application will run on the other. This results in better efficiencies as the app doesn't have to 'share' its core with any windows stuff like it would have on a single core system. - Gottschalk, on 10/12/2007, -0/+3While that is largely true what relevance does that have in today's market?
For PCs you can either go Conroe or 6600Q (next year) depending on your need for a dual or quad core solution.
It's not about first but best. I'm sure AMD will continue to be fine. Intel has been behind AMD for a while and now they are ahead. There is no need for anyone to fanboy for either one.
Saying 'lolz Intel sux0rz' doesn't make sense when the core duo 2 is clearly superior. Saying 'AMD is teh doomz' doesn't make sense because of AMDs stunning track record. I don't understand fanboyism. Competition makes both companies better.
Since Intel is hellbent going in their direction I wish AMD good luck with a different one. This way the market gets offered two different paths. One may wind up being better but at least we're getting some interesting alternatives. This is progress. - Gottschalk, on 10/12/2007, -1/+4The Conroe is superior. This isn't a cheap trick from Intel they actually put some original thought in the design and came up with a better solution. Tomorrow maybe AMD will have the better solution. I don't understand these biased comments. AMD has historically been ahead of Intel. So? Who cares. If AMD is ahead tomorrow I will say so freely but today the Conroe is nice. I'm not as sure how the quad core market stacks up though.
- msgyrd, on 10/12/2007, -1/+4@cypherz:
Yes, *nix handles processes better than windows, but as I touched on in my previous post, there is somewhat of a ceiling of how many processes a system will use, and at some point, you need to handle processes better, not just more of them.
And threading an applications does increase complexity. You don't always want to fork a new process. Mutual exclusion is easier to handle in threads than forks, gracefully closing your application is easier, sharing CPU usage between threads is easier than forks (as far as I know, you can't yield in a forked process), and race conditions are easier to control or prevent.
I'm not saying that more cores is a bad thing, having more hardware available is always nice, but I think the benefit of more is on a scale of diminishing returns. I predict, for what its worth, greater gains will be made by AMD by pursuing this, than Intel will receive from making a 10 or 12 core chip. - sc0ticus, on 10/12/2007, -0/+3I'm hoping someone can help solve my confusion about multi core processors.
To get a performance boost, do programmers need to explicitly take advantage of multiprocessors? Is there some kind of automatic threading going on behind the scenes? If I wrote a program using threading, would the OS automatically spread those threads out over the different cores?
If anyone has a link to a thorough explanation of multi-core processing, I would much appreciate it. - ronin2040, on 10/12/2007, -0/+3Is it just me, or are the comments on this article the most coherent and intelligent in the history of all digg articles?
- Psych77, on 10/12/2007, -1/+4Dual core is rubbish for gaming. Yep, my E6600 can't perform...
- pcgeek101, on 10/12/2007, -1/+4Did you ever consider the fact that Autodesk's software does a lot? Sure, they could strip out some of their ***** licensing bloat, but is it really all that bad on an average machine? My experiences say "no".
In case you're curious, I did work with AutoCAD 2004/6/7 at my old job. Not to mention, Maya 8.0 is awesome too. :) - geminitojanus, on 10/12/2007, -1/+3Well, serial protocols tend to get outdated very quickly, so they're making a second generation of PCI Express that has about twice the theoretical bandwidth of PCI Express as we know it now (32x links). It's essentially the same protocol (I believe it's forward compatible), but with some updated and fixed things.
http://www.theinquirer.net/default.aspx?article=25774 for more info. - geronimo, on 10/12/2007, -0/+2actually you can't buy the 5355 yet but even the 5320 and 5345 should easily beat out the 885.
- batmang, on 10/12/2007, -15/+17Intel seemed to have done the whole jamming thing with MHz. Whereas AMD came out with a superior architechure that could pizzwn 3.2 GHz with a 2.2 GHz
- kelbear, on 10/12/2007, -0/+2Gottaschalk is right on the money,
It seems that AMD wants fewer but more capable cores, while Intel is gunning for a large number of cores to handle the task.
Two different approaches gives us (the consumer) a better chance of being offered the best solution for our problems. Or perhaps, both approaches may need to be adopted in the future to give us flexibility in how we handle the problems. If we knew which company had the better idea, of course we'd want that one to succeed(and it naturally would). But since nobody knows, having both of them work on their own approach is our best shot at finding the answer. - Gottschalk, on 10/12/2007, -0/+2No games support dual cores...
Except these apparently, flukes I say!
http://forumz.tomshardware.com/games/Dual-Core-supported-games-ftopict94969.html
Besides I'm sure that in the future dual core support will be even worse. In fact if you go quad core most games will refuse to run, on principle. - HerbertScrunge, on 10/12/2007, -0/+2"The Dual core is similar to Internet."
I'd say it's more similar to a system of tubes - Kwipper, on 10/12/2007, -4/+6*reads the topic*
Umm yeah.. but wasn't it AMD that decided to do Quad Core first? Oh now that their 4x4 got beat out by Intel's new quad chip, hell it got beaten out by a few of the dual cores, they decided that multicore CPU's just aren't the future? Yeah. Nice way to back down there guys. :/ - Marthinus, on 10/12/2007, -3/+5Again AMD will innovate and Intel will just cling to what works until the world realizes their game. The MHz war was a prime example of this. AMD made slower MHz CPU, better perfomance
- Gottschalk, on 10/12/2007, -0/+2AMD does not have the resources to strike hard in the sym. multicore market. So they are going straight for asym. multicore. Intel is going A, B, C. AMD is going A, C.
- MacGyver2210, on 10/12/2007, -0/+1"Tell that to Autodesk, their software while bloated *needs* good hardware."
That's a valid point, but at the same time if AMD were to make a special 'CAD-Accelerated' processor for a special Autodesk workstation, then they probably wouldn't "need" so much other extra hardware, and I'm sure they could find a way to trim down on the software overall as well, like finding a better hardware alternative for their copy protection. - geronimo, on 10/12/2007, -0/+1It used to be that you would use multiple processes + ipc over threading since some OS'es ran faster that way. But for linux, the difference between thread and process is negligeable. If you are writing a portable app that must be fast, use multiple processes + ipc and be safe. But if it's running only on linux, why would you not use multiple threads? pthread_* functions are vastly easier to use than the myriad of IPC mechanisms. And with Java and python et al, making multithreaded apps is 100x's easier than multiprocess+IPC apps.
This is why postgres, oracle et al are multi-process applications, it's a legacy holdover from a long time ago and it's just plain more portable across all the platforms they are able to run on. - geminitojanus, on 10/12/2007, -0/+1"Right now a 100 core chip would basically be useless, for most PC applications, if it doesn't have software support."
Which is exactly why nobody is building a 100 core chip, right now. That's the reason for the Terascale project, to examine what kinds of interconnects on chip and off are necessary to make these systems useful. How do you even connect a system RAM to 100 processors? Do you stick them on a packet-switched bus and hope they don't all talk over each other? Do you attempt to run busses to each chip, and hope your wire delays won't kill you? Ring or Grid topologies? How do you deal with individual dead cores, or heat generation from one core leaking into the other? All of these things need to be examined and a lot of information needs to be generated on how to make these devices work together. There are still too many _hardware_ problems to be solved, even before we start thinking about the software aspect of having 100 cores.
This is why Intel's deferring the problem. By giving research teams time to solve these problems, they can let their traditional architecture teams continue to do what they're good at: refine, rebuild, replicate, repeat. One to two, two to four, four to eight, eight to twelve, slowing down to give software time to catch up to the rate in which you're throwing more and more cores at it.
AMD on the other hand, thinks the solution isn't replicating more cores, it's simply abandoning the whole metaphor. Build more complex cores. But, their technology is so far behind that of what they need, they simply can't do it, yet. Instead, they'll continue the current metaphor, maybe to eight cores, definitely to four, but then start building MCMs like Intel's making, but instead of packaging two quad-core Opterons, let's make it a Quad-core Opteron, a Radeon X19xx, and a crypto core for home users, or an Opteron and two Radeons for a server, or a cut-down Opteron and two Radeons and an MPEG core for Nintendo's next game console.
AMD wants it this way because they simply won't have the technology to fab it all on one piece of silicon for years (at least 2, my guess is closer to 4 thanks to the ATi acquision costing them so damned much). But, they do have the money and the excess capacity to fab a lot of smaller chips and package them all nicely together (which is exactly why Intel's using that strategy now; they /could/ build a native quad core, but why bother? We've got plenty of these extra Core 2 chips laying around, and we've got the capacity to build lots and lots more, let's use that to our advantage and delay retooling costs).
The problem outsiders have is that they forget that these companies are businesses: they do things because they make sense in both the financial and in the technological realms. Intel is a huge, very conservative company, they don't like the make radical moves. AMD is a smaller, but still big company, so they can afford to take more risks, but they've still got to play it safe, especially in the shadow of Intel and IBM and now nVidia. - msgyrd, on 10/12/2007, -0/+1There are a couple uses for multiple cores, when the cores are symmetrical (both cores are the same). Intel is focusing on adding more symmetrical cores.
The first is kind of "implicit" usage. The programmer doesn't code for multiple cores, and the OS just does load balancing of processes on different processors/cores. Each instance of the application runs on one core.
The second is "explicit" usage. The programmer specifically writes a threaded application so that the application will run on multiple cores at the same time. The OS still handles load balancing, but each instance of the application can run on multiple cores. This is where multiple cores really shine.
However, this announcement from AMD implies multiple asymmetrical cores. Different cores will be better at certain tasks. From their pictures/graphs, it looks like several general purpose cores, then a bunch of specialized cores. For this, you'll see applications getting load balanced onto cores that handle the process better. I'm not sure how they're doing this, so I can't speculate on how it will work exactly, but I feel that its a better decision than Intels "lets throw more cores at it". - WiBu, on 10/12/2007, -1/+2Isn't this similar to that "anti-hyperthreading" thing that was being talked about quite a while ago. AMD where rumored to be researching a way to make multiple cores work on a single thread.
- geronimo, on 10/12/2007, -0/+1Yes but code optimized for multiple cores won't perform as well on a single core, so game designers will have to be more creative now. Do you know why your linux kernel comes with either smp or non-smp version? the smp version is optimized for >1 core(really - up to 4 to 8, beyond that you should recompile the kernel). With the non-smp version it is meant to run with one core so it doesn't try to grab expensive CPU locks - those instructions are commented out when you compile linux smp.
But, game designers do some fancy stuff that I never understood so I'm sure they could learn this new dimension to programming. - Woknblues, on 10/12/2007, -0/+1not a fanboi of either company for the record.
I notice that AMD is willing to "experiment" more with their product line..... They will take a risk here and there, and seemingly cop to the "mistake" and move on. Intel in my view goes "all in" to borrow a texas hold'em expression. Intel's chips seem to me to be the next gen to me, but my AMD3700 single core (*gasp*) runs everything just fine for now. I can alt+tab out of BF2, and run anti-virus at the same time, no slow down. tell me again why I need to go dual or quad "right now!"??
I don't really know what the hell I am talking about, but, I like the idea of integrated cores if processes will be run "behind the scenes", effectively multi-tasking for me, rather than having me do the "multi-tasking"... I can't imagine photoshopping, encoding and playing BF2 while running an anti-virus and talking on skype and, and , and, and... guess I'm a one trick pony.. I can't even talk to my gf sitting right next to me when I am sniping on mashtur...... - MrViklund, on 10/12/2007, -0/+1To be honest. I think that AMD is right here. We will go from Quad core to eight cores and then maybe 10 or 12 cores but I don't think that 10s and 10s of more cores will do any good. At lest not for every day use. But we will see. It will be interesting to watch how AMD are going to integrate their products with ATi's in the near future. Like integrated GPU-unit in the CPU.
- benblack, on 10/12/2007, -0/+1Quad core processors have enormous potential for the segment of the market that is most interested - gamers.
Single-threaded games combine game logic, artificial intelligence, networking, and physics in a single thread. In order to maintain a reasonable frame rate, restrictions are placed on the amount of processing time given to networking, physics, and artificial intelligence, to the detriment of overall game quality.
In a quad-core environment, if physics and artificial intelligence run as separate threads, then they can run on their own dedicated core, and we may hopefully expect to see a quantum leap in both game realism and game AI. - Gottschalk, on 10/12/2007, -1/+2Yea screw what works!
- Satanael, on 10/12/2007, -3/+4What's this 2nd Generation of PCIe they speak of?
- BadgerOU, on 10/12/2007, -0/+1Actually, it's almost the opposite. They were being made smaller, so we could fit more transistors on a chip ... we just didn't know what to do with the extra transistors! So we just threw another copy of the same processor on the chip and called it dual core.
I know it's more complicated than that, but that's the basic idea. - Gottschalk, on 10/12/2007, -1/+2640K of memory ought to be enough for anybody!
- msgyrd, on 10/12/2007, -1/+2"Sounds suspiciously like the Cell processor to me. That can't be good..."
I think people misunderstand the Cell processor because Sony fumbled the ball with it by including it in the PS3. The Cell is a different approach to processing, and it has potential to be better, but current software isn't written to exploit it's advantages (optimizations). AMD will encounter the same problem if they introduce this all at once as an alternative, when it needs to be introduced slowly. They'll tout their advantages, someone will benchmark it with an OS and an application that doesn't really use the newfangled processor, and everyone will point it out and claim the new processor sucks. It happened to the Cell already, someone installed an unoptimized version on linux on a PS3 and claimed it benchmarked roughly equivilant to an 800mhz x86 processor. A Cell compiler didn't even exist until last year, you can't expect all the bugs to be ironed out just yet. - MacGyver2210, on 10/12/2007, -0/+1"Performance you say? Give me solid state hard drives."
Amen. Too bad they cost more than I make in a year for like 4GB of storage... :-/
http://www.superssd.com/products/tera-ramsan/ - Thank you RamSan - Jammerdelray, on 10/12/2007, -0/+1"It's not uncommon to see games or encoders performing worse in dual core machines than in single core CPUs"
Rubbish as far as gaming goes. Lower Cpu usage and better Latency in online games.
Numerous people have said the same thing. - spoonard, on 10/12/2007, -0/+1From a laymens point of view (And I certainly am the laymen here!) It seems Intel in the past thought it could beat out AMD simply by piling on the Mhz and win through sheer numbers. Turned out AMD had an Ace up thier sleeve with the Athlon processor. AMD seemed to be about smarter processing. That's why from the K7 up until the release of Intel's CoreDuo, AMD's processors were beating out Intels consistently even when Intels processors were running at up to a full 500Mhz to 600Mhz faster then AMD's.
-
Show 51 - 84 of 84 discussions
Check out the new & improved 
© Digg Inc. 2009
— Content created and posted by Digg users is