Single vs Dual Core

[AleleVanuatu]

Extelleron, one reference to shut your mouth up: Chuck moore and forth chips. Check out ultraforth.

You have zero clue about how hw engineering works, simply put, improvements are always possible. For crissakes, look at quantum computing.

You sad sad kids, already so dry and grumpy, agreeing that WORK PER CYCLE is going to remain low, that moore's law is FINISHED, god it pains me for the future. It's not even close to finished. You think throwing more cores at the problem is gonna solve it quicker? LoL. Not even close. Buddy, look at JS execution. Just think about that. Simple problem for you kids who know html and JS to grok. Tell me how they are going to multi-thread rhino. Please. LoL.

For godss sake, Please, start getting optimistic. I'm going to let this thread die now, you guys have drunk the koolaid and are honestly so pessimistic and ready to accept the stagnancy of the marketplace.

At least you've made me feel young again, I have vigor and urge for change! Remember, you're only as old as you feel.

Oh and Excelleron, Nice work on Excel 2007. Now optimize the gajillion other products! LoL, Just LoL.

If you keep up the personal attacks, you will be getting a vacation.
Markfw900 Anandtech Moderator

 

[Extelleron]

Originally posted by: AleleVanuatu
Extelleron, one reference to shut your mouth up: Chuck moore and forth chips. Check out ultraforth.

You have zero clue about how hw engineering works, simply put, improvements are always possible. For crissakes, look at quantum computing.

You sad sad kids, already so dry and grumpy, agreeing that WORK PER CYCLE is going to remain low, that moore's law is FINISHED, god it pains me for the future. It's not even close to finished. You think throwing more cores at the problem is gonna solve it quicker? LoL. Not even close. Buddy, look at JS execution. Just think about that. Simple problem for you kids who know html and JS to grok. Tell me how they are going to multi-thread rhino. Please. LoL.

For godss sake, Please, start getting optimistic. I'm going to let this thread die now, you guys have drunk the koolaid and are honestly so pessimistic and ready to accept the stagnancy of the marketplace.

At least you've made me feel young again, I have vigor and urge for change! Remember, you're only as old as you feel.

Oh and Extelleron, Nice work on Excel 2007. Now optimize the gajillion other products! LoL, Just LoL.

Are you seriously telling me that coding for multiple threads is too hard and will take 15 years, and at the same time suggesting quantum computing as a solution for immediate performance problems? It's my turn for a "LoL."

You're the one with no clue how hardware works. Saying that nothing is impossible in computing is pretty much true, but there are boundaries. You cannot just increase clockspeed and keep on doing it. It's not possible. Even if it were possible, increasing clockspeed is not at all a power-efficient means of increasing performance. Doubling clockspeed will roughly double performance, but power consumption will not rise linearly. While power increases linearly with a pure increase in frequency, an increase in frequency requires a corresponding increase in voltage. Voltage increases exponentially increase power consumption, so quickly you have 3-4x the power consumption for 2x the performance. At high voltages and frequencies, leakage becomes a serious concern and this is one the serious issues that the Pentium 4 faced. Compare that to multi-core, where you can double the amount of cores, double power consumption, and increase performance by up to 80-90%. The future of CPUs is small, power-efficient increases in IPC, small increases in frequency, and increasing numbers of cores. Ask anyone who knows anything about hardware development and they will agree.

Suggested reading in this area: http://www.anandtech.com/showdoc.aspx?i=2343&p=1

Moore's Law isn't anywhere near finished. Moore's Law has nothing to do with clockspeed, cores, or IPC, it says that chips will become significantly more complex as time goes on. That is certainly proving true and that will continue to be the case. So long as fabrication technology can be scaled down, Moore's Law will continue and so will increases in performance.

I don't see what more I can say to you. You act as if you know it all, but the vast majority of what you say is simply false.

 

[Idontcare]

Originally posted by: AleleVanuatu
Extelleron, one reference to shut your mouth up: Chuck moore and forth chips. Check out ultraforth.

You have zero clue about how hw engineering works, simply put, improvements are always possible. For crissakes, look at quantum computing.

You sad sad kids, already so dry and grumpy, agreeing that WORK PER CYCLE is going to remain low, that moore's law is FINISHED, god it pains me for the future. It's not even close to finished. You think throwing more cores at the problem is gonna solve it quicker? LoL. Not even close. Buddy, look at JS execution. Just think about that. Simple problem for you kids who know html and JS to grok. Tell me how they are going to multi-thread rhino. Please. LoL.

For godss sake, Please, start getting optimistic. I'm going to let this thread die now, you guys have drunk the koolaid and are honestly so pessimistic and ready to accept the stagnancy of the marketplace.

At least you've made me feel young again, I have vigor and urge for change! Remember, you're only as old as you feel.

Oh and Excelleron, Nice work on Excel 2007. Now optimize the gajillion other products! LoL, Just LoL.

This is unnecessary and quite out of line with the spirit of how we like to treat one another on these forums. Please stop.

 

[Markfw]

Originally posted by: Idontcare

Originally posted by: AleleVanuatu
Extelleron, one reference to shut your mouth up: Chuck moore and forth chips. Check out ultraforth.

You have zero clue about how hw engineering works, simply put, improvements are always possible. For crissakes, look at quantum computing.

You sad sad kids, already so dry and grumpy, agreeing that WORK PER CYCLE is going to remain low, that moore's law is FINISHED, god it pains me for the future. It's not even close to finished. You think throwing more cores at the problem is gonna solve it quicker? LoL. Not even close. Buddy, look at JS execution. Just think about that. Simple problem for you kids who know html and JS to grok. Tell me how they are going to multi-thread rhino. Please. LoL.

For godss sake, Please, start getting optimistic. I'm going to let this thread die now, you guys have drunk the koolaid and are honestly so pessimistic and ready to accept the stagnancy of the marketplace.

At least you've made me feel young again, I have vigor and urge for change! Remember, you're only as old as you feel.

Oh and Excelleron, Nice work on Excel 2007. Now optimize the gajillion other products! LoL, Just LoL.

This is unnecessary and quite out of line with the spirit of how we like to treat one another on these forums. Please stop.

Yes, warning given. Vacations will be forthcoming if the attacks continue.

 

[AleleVanuatu]

Ok no problem, I see that serious debate can't be handled here, so back to towing the party line.

 

[SunSamurai]

Originally posted by: AleleVanuatu
Ok no problem, I see that serious debate can't be handled here, so back to towing the party line.

Not when a douche like you is involved. LoL

 

[Markfw]

Originally posted by: AleleVanuatu
Ok no problem, I see that serious debate can't be handled here, so back to towing the party line.

Debate is one thing. Personal attacks are another... If you don't know the difference, then you don't need to be posting here.

And as far as multithreaded apps, my quads do very nicely on F@H, thank you very much...

 

[bunnyfubbles]

Originally posted by: AleleVanuatu
Dual-core is a polished turd and enthusiasts are loving it. 99% of apps run on a single core, and are completely bound by single-path execution.

Look at Firefox or IE. Run a JS intensive page, and watch the rest of your browser windows get all laggy as well. It's all running in the same process, and that's crap right there.

It's all about raw clock-speed in the end, on Core 2 of course. So 4 GHz matters, dual-core or quad-core, not so much.

Software is going to take 10-15 years to work with this stuff. Regular software I mean, not stupid encoding crap that everyone loves to toss out as if everyone rips dvds to x264. LoL. Just LoL.

Uh, so what strictly single threaded programs are out there that absolutely require 4GHz+ from a single core on a Core 2 chip in order to tangibly improve user experience/productivity?

I think the thing you're failing to recognize is the fact that, while most people truly do not need anything more than a decently fast single core, those who do need something more actually run programs and/or work in a way that benefits significantly from multiple cores

Yes, you're correct about programs not running any faster when balanced between two cores (or more), but that's part of the point - a program consuming 100% of a single core at high priority can do so on a dual core either spread between the two or with a specific affinity to allow the remaining core to be ready for whatever else. While it doesn't speed up that one single process, it does leave you with processing power to spare so you can continue to do other things instead of waiting for that first task to finish - this novel concept is known as multitasking.

 

[Cogman]

Originally posted by: AleleVanuatu
Holy crap I think you guys must be kids. Guaranteed. LoL. Let's take this hardcore insane assertion and rip it apart.

Yeh, uhh, anyone looking at perfmon with a dual core should know by now that single threaded apps get spread across cores etc. Just get a quad and watch how 25 percent of each core is used when running one instance of prime.

Buddy, buddy buddy. It's called a context switch.

No I'm not a software programmer, but that does not mean that I am wrong. Windows does not by some magic create two threads out of a single thread with some kind of "reverse HyperThreading." But it spreads the load over various processing cores. That is why unless the CPU is utilized 100% or affinity is set, no single core every reaches 100% utilization. Threads are not tied to one CPU in Windows.

Again!

Guys, you gotta get with the program. what's happening here, is really simple ==> A process is started with one single-path execution. That execution is divided into time slices, and distributed over the cores, but IS not load-balanced. What that means is that for time 1 it runs on core0, time 2 it runs on core1, then back and forth flipping like that. But my friend, please understand, this DOES NOT SPEED IT UP -- its still single-path execution! In fact, when you set the affinity to use only one single core, you will notice it runs FASTER!

Why?

Because of less context switches. Try it Read about it. Check the amount of time Prime95 takes to generate 1MB of results. Jesus H Christ of Latter-Day Saints. Oh Gilgamesh, Oh FSM!!!!

Cogman, Cogman, Cogman.

Office apps are too simple to make a difference if they run slow or fast? Cmon buddy! Sort a 400MB excel table, and tell me, does it make a difference? Granted this is nuts, but real users, in the real world, DO THIS!

Yes, again, I said, game logic sure, easier to parallelize. EAX in software on one core, Enemy AI on another, Path finding on another, yada YADA YADA.

This is not most apps. This is not even close to 10% of algorithms. You guys need a a computer science lesson!

Read about parallelizing quicksort. Please, thank you!!!!

LoL. You guys are fun at least. LoL.

Lol, Quicksort isn't at all hard to make parallel due to it "divide and conquer" methods. But if that fails you can always use mergesort as well. What did you just finish your CS classes but have been scripting in VB for 13 years? I suggest you go look up the wikipedia article on quick sort under Parallelizations.

And you still didn't answer my question, What are most apps? already I have hit the big ones that 90% of consumers use. And you would be blind to say that games aren't that big of a deal.

Want to go into the server market? Ok. Tell me, why since the creation of the internet has every webserver, File server, and Database server been hosted on machines with multiple cores (not always on the same chip). Do ISPs and data centers just love to throw away money. I guess they must. Oh wait, more then one user connects to an ISP. Databases often have tree structures (Easy to parallelize). so multiple cores are in fact beneficial in this situation.

So please, don't question my CS skills and just so you know. Sorting is an especially bad place point to.

And btw, saying game logic is easy to parallelize is quite a bad statement. Games are by their very nature hard to parallelize and see much benefit. The fact that developers have found ways to evenly split the load speaks more against your argument that "Nothing can me made multi-threaded for the next 30 years" then it does for it.

Oh, and I just had to laugh. You point to quantum computing as a possible improvement to computational power yet you are debating against multiple cores. Truly you just don't get it. Quantum computing by its very nature excels the best at applications that require many many processors. it does not, however, benefit in the least with in-order operations. Yes it can decrypt pretty much any form of cryptography known almost instantly, but that is a special case. You can take an FPGA and make it into an encoder that is faster then any computer available but that doesn't mean it will play your games faster.

 

[aigomorla]

Originally posted by: AleleVanuatu

Yeh, uhh, anyone looking at perfmon with a dual core should know by now that single threaded apps get spread across cores etc. Just get a quad and watch how 25 percent of each core is used when running one instance of prime.

ONE MORE TIME! Run task manager and notice that it's not 25% of each core, but one full core at 100% load! LoL!

easy there!


i will tell you one thing.

all my quads go on load 100% cuz i do use them.
So please dont say quadcores are usless.

Also the games im selective in use quadcores.

And one of the rules in AT is to respect others. We love a good debate, we hate a good flame tho.

 

[myocardia]

Originally posted by: AleleVanuatu
Also, please GUYS, remember that this is going to take 15 years to see any real benefits! Also, just look at games, your pet software, and read developer interviews where they ADMIT that they can't parallelize beyond quad-core,

Let me tell you a story about a guy who, two years ago, said the same things about gaming. He bought the fasest single core processor he could find (besides the ~$1,000 Athlon FX's, that is), and thought he would be "good to go" for a good, long while. He even recommended to other people that they should do the same, buy the fasest single-core they could afford.

About three months after he bought said Athlon 64 4000, the game Oblivion was released, and every review of it said that it "definitely requires" a dual-core. That was okay, because he didn't plan on buying that game, anyway. A few months after that, Microsoft released Flight Simlutor X (a game, in case you didn't know), which is still to this day the most CPU-bound game ever. After the first service pack (for FSX, not XP or Vista) was released, guess what? It can take advantage of up to 256 cores. Uh oh, he had bought that A64 4000 mostly for FSX, which it ran pretty poorly, even when overclocked to 2.8 Ghz. So, he bought an Opteron 170, and overclocked it to 2.8 Ghz, and it ran FSX about twice as well as the single-core 4000 did.

Then, this past September, he (that would be me) decided to build another system. Guess what mistake I didn't make again? Right, this time, I built myself a system around the newly released Q6600 G0 (a quad-core), and haven't looked back since. The Q6600 performs at least twice as well with FSX at it's stock speed as the Opteron 170 did with an 800 Mhz per core overclock, and that was with the same video card.

Have you started to see where this is going yet? Yes, a considerable amount of programmers seem to be quite lacking in the IQ dept., and usually are the last ones to catch on to what's going on in the computing world, but you're way too late to convince anyone to skip dual-core processors, and around here at least, most likely too late to convince anyone to skip quads. Whether you like it or not, the computing world is going multicore.

BTW, what does JavaScript have to do with anything?? A 1Ghz PIII has no problem at all running it, so why would anyone care whether it's multithreaded or not?

 

[Idontcare]

Originally posted by: AleleVanuatu

Yeh, uhh, anyone looking at perfmon with a dual core should know by now that single threaded apps get spread across cores etc. Just get a quad and watch how 25 percent of each core is used when running one instance of prime.

ONE MORE TIME! Run task manager and notice that it's not 25% of each core, but one full core at 100% load! LoL!

A single-threaded application running full-out on a quad-core CPU will see the thread intentionally migrated from core to core to core such that on average each core is utilized 25% of the time...and there is no better proof of this than a quick glance at task manager.

Thread migration is very much a real and problematic issue for modern multi-core CPU's as it defeats the load-based power-saving and selective core-overclocking features.

 

[ch33kym0use]

Yes, dual core is good. If one core is maxed out CPU usage then other core can do work. Also apply to quad core or more possibly.

 

[VirtualLarry]

Originally posted by: nerp
Yeh, uhh, anyone looking at perfmon with a dual core should know by now that single threaded apps get spread across cores etc. Just get a quad and watch how 25 percent of each core is used when running one instance of prime.

I know it's hard to accept, but it's true.

I wonder how many people that are doing this, have installed the MS Multicore patch v4. Supposedly, it fixes the thread-bouncing problem. I have not tested it myself.

Apparently the thread-bouncing is a side-effect of a patch made to the scheduler in XP to support hyper-threading. Thus it trys to prevent one thread getting pinned down on a hyper-threaded virtual cpu, when a real cpu was free.

According to a book on the W2K kernel, threads are supposed to be "sticky", and NOT constantly bounce between CPUs.

 

[VirtualLarry]

Originally posted by: AleleVanuatu

Yeh, uhh, anyone looking at perfmon with a dual core should know by now that single threaded apps get spread across cores etc. Just get a quad and watch how 25 percent of each core is used when running one instance of prime.

ONE MORE TIME! Run task manager and notice that it's not 25% of each core, but one full core at 100% load! LoL!

It's not one full core at 100% load. I wish (for performance reasons) that it was. I blame MS.

 

[VirtualLarry]

Originally posted by: bunnyfubbles

Uh, so what strictly single threaded programs are out there that absolutely require 4GHz+ from a single core on a Core 2 chip in order to tangibly improve user experience/productivity?

Easy. M.A.M.E.
The developers (last time I checked, which was probably last year) were feverently opposed to multi-threading, due to the significant complexities involved in overhauling the entire emulation core.

I came up with some algorithms that I felt would allow for speed gains while multithreading, and in the same time improve the quality of the emulation, but they were not accepted.

 

[soccerballtux]

Originally posted by: AleleVanuatu
And, being in the professional software industry for 15+ years, I can tell you that you're dreaming if you think that parallelizing algorithms are suddenly going to appear, be implemented, or hell be invented and debugged within the next 5 years. It's going to take 15 years, at least. Sure you'll get some simple parallelizable operations like encoding, hell even some game logic, but apart from that? LoL you're dreaming if you think everything can be split into parallel operations. Look into MPL and see how those scientific programmers have been trying this stuff for years, and how the threading bugs are biting them in the ass on a daily basis. And oh Erland? Yes it's natively parellizable, but look into its requirements and deficiencies. LoL Just LoL.

Throwing some water on the fire hype. That's my job, Alele Vanuatu is out.

Good posts. I stopped reading Cogman's when he said 1). anything can be dual threaded and 2). "I'm pretty sure the tabs in Firefox and IE are different threads." There's a debate in FF development now about doing this, because the temptation for the developers is to make these threads visible to extensions (easier), but that's suicide for Firefox (because you'd be relying on the extension developers to know how to, and follow, multi-threaded coding procedures for the basic stability of Firefox.)

 

[Cogman]

Originally posted by: soccerballtux

Originally posted by: AleleVanuatu
And, being in the professional software industry for 15+ years, I can tell you that you're dreaming if you think that parallelizing algorithms are suddenly going to appear, be implemented, or hell be invented and debugged within the next 5 years. It's going to take 15 years, at least. Sure you'll get some simple parallelizable operations like encoding, hell even some game logic, but apart from that? LoL you're dreaming if you think everything can be split into parallel operations. Look into MPL and see how those scientific programmers have been trying this stuff for years, and how the threading bugs are biting them in the ass on a daily basis. And oh Erland? Yes it's natively parellizable, but look into its requirements and deficiencies. LoL Just LoL.

Throwing some water on the fire hype. That's my job, Alele Vanuatu is out.

Good posts. I stopped reading Cogman's when he said 1). anything can be dual threaded and 2). "I'm pretty sure the tabs in Firefox and IE are different threads." There's a debate in FF development now about doing this, because the temptation for the developers is to make these threads visible to extensions (easier), but that's suicide for Firefox (because you'd be relying on the extension developers to know how to, and follow, multi-threaded coding procedures for the basic stability of Firefox.)

Well, great to see you don't have any reading comprehension because I never said 1.. though Two I was unsure of, but thought it would be pretty logical. I guess im behind in my knowledge of browser development.

And just so I don't sound like im saying everything can be dual threaded. Ill say it very clearly, Not everything can be split into two threads. However, that doesn't mean that a good portion of things can't be. There does need to be a big paradigm shift in the way code is thought of, maybe even a new low level language the exposes threads more readily (or makes using them more natural). However, that doesn't mean that threads are impossible to access or use as it is.

AleleVanuatu was pretty much arguing that dual cores are almost completely useless and that very few programs ever utilize both cores without having to battle some sever bugs. This is just not true.

But, I don't you might not have read that far.

Well, since I need to be more credible for this non-believer about multithreading picking up...

Excel Multi threading
Small list of games that are multithreaded
Intro to OpenMP a multithreaded development helper, including a tutorial how to make quicksort multithreaded.
Folding at home, a good example of scientific applications using multithreading

I suggest you read the quicksort multithreading article as it was a one that AleleVanuatu declared to be extremely hard to multithread.

 

[josephfuentes]

Well my understanding so far about processors is core is better than two for all the reasons stated above. however, the OS and more importantly the apps have to take advantage of this (through multithreading - is this same as multiprocessing?)

Now, let me throw some spikes on the highway. What about quad cores? Will a quad core system, desktop or laptop, run better, more efficiently than a dual core? if so is this by virtue that there is more engine to share the wealth with? or are apps including the OS taking advantage of quad core technology?

Finally, for high end systems, pitting the best dual core and quad cores centrino-2, extreme processors, phenoms, opterons, the whole lot, even through in xeon for fun, who/what proves its toughest mettle?

People talk about not just CPU speed as being the benchmark. everyone knows it isn't. Then what are the real benchmarks that will crunch a CPU until it whimpers for mercy. Games? Photoshop stuff? Rendering movies? Nah! How about the nuclear physics calculations in the teraflop range? how about 3D solid modelling like air flow analysis? how about real time meteological simulation? Does anyone ever take those into consideration?

 

[solog]

Originally posted by: josephfuentes
Does anyone ever take those into consideration?

No. No one ever does, just you. Sorry.