Phrenetical - I'm not a programmer, but neither are complete ignorant. I might be wrong, however what you presented as simplified way how to see more core is helping to speed up things is a bit wrong for two major reasons.
First. Class person(); //command new person 1 - your idea is that every person is entierly independent on the others. That is major flaw in your logic, since (let's assume these persons are bots in game) actions of person 2, 3 and 4 can't collide with person 1. We can give it as simple, as all four persons can't stand in the same spot, just because that is the most optimal way to play right, now, yep?
So, you need to process person 1 first and update his (we say simple, so we do it simple) position. THEN (and not a picosecond sooner) you can process person 2 WITH the already known position of person 1, witch he can't use, so he go for the second best shooting position at your a$$
That was logical reason why your example is entierly wrong.
Second reason is highly technical. To do ANY kind of work, processor need to use memory, where the data are. Even todays big caches can't help notably, because if you ever did low-level assembler coding you know that many stuff you need to write to memory, and not only to cache. And the position of the soldier/person above is exactly this example. What is the problem with dual/quad core there? Oh, wery simple. Only ONE memory buss. So, using dual or quad core you can't effectively say you got 2 or 4 processors, because they are all sharing one memory bus (it get deeper than that, they also sharing one incomming instruction pipeline, but that is way deeper we want to go) and therefore they can't run in paralel at all, except for what they store in their caches. (hint, that is why latest intel models come with 8MB of L2) But as we said already, many informations are need to update in memory and increasing the number of core's only add another reasons for that. However storing informations in memory are time consupting. Average CPU run at 3 000 Mhz, while average memory run at 250 Mhz, somewhere 12x slower and I did not even go to the details like the ram is not CL1 but Cl2, 3 and even worser + refreshes, bank timings... etc.
Hence adding core only put more stress on the memory bus and it is not coincidence that many things are on same machines using dual/single core CPU's a little slower on dual core ones.
Funny part.
Your imagination how "CPU render pixels" is laughable at best. Yes, some demos/intros and other experimental stuff use realtime raytrace to produce 3D scenes, but usualy games todays use GFX cards to draw pixels. The difference (CPU commanding GPU to draw a pixel) might not sound big on the first time, but we get there. 3D world on computers are created using geometric models that use polygons. (no nurbs curves yet, eh) The very simplified version of explaination 3D engine goes like this. In BSP (Binary Space Partion) are found the position of the player and his rotation of his look. Given this, processed is everything with the fog of his view for preparation of screen draw. (important - at the time, everything has to be computed already, no paraelizing again, sorry) Now are taken every polygon in the fog by his normal is determined the Z-distance and possible overlapping to sort out polygons that are just hiddenand stay hidden for player, so there is no need to bother with them anymore. Ater this, these geometric data are being feed to the GPU, that know that for each polygon he has to use certain texture + another textures to create the effect desired...
The only way to paraelize that stuff is to proces geometry on one core, and AI + sound + memory management in another. Your imagination that from quad core each core draw (or prepare to draw) 1/4 of the screen is pretty vivid and clearly it can catch some users, however it is completely false, as each step require that the step before it is already done (!!!).
And even paraellizing the 3D screen setup for 4 cores willbe insane, because many polygons are overlaping between the 1/4 of the screen rendered and these data has to be shared using main memory (slow) - witch will lead to massive slow-down...
its more then possible to divide something up and put it back together
Certainly. But while you spend your time by dividing and then putting it back together, the single core is already twice time done with it
need a way to detect if a machine can even do paralell processing, creating heaps of overhead in most current languages and slowing down the app
That is exactly why multicores support are next-to none even in professional apps. One thing is paraelize a benchmark and another application where steps depend on each other. Thing that obviously did not cross your mind yet.
multicore software programming, they have proven you can run things like windows ona CPU with 32 cores each running at 1.5Mhz
If you bother to look into computers history you know, that things better that lazy winblows can be run on 7.15Mhz CPU. (Amiga 1000 in 1985 and Amiga OS that can do much more that winblows can do today)
Immortal_Hero -
Do you really think that the CPU makers are just spending millions of R&D money to develop multiple cores because they can market them?
LOL Your "argument" is very laughable, mate. Listen to yourself. What you are saing is just "
because they do it, it must be the only and best way to do it"
Because milions of people killed themselves, it must be great thing to do and you should do it
ASAP!
Nah. I seen millions of bucks invested into way more stupid things so I can take this as something more that laughable taunt.
Besides my point was not that dual/quad cores are useless. My point was, that it is very hard to paraellize code and mainly games, so I see no point of multicore stuff other that sell CPU's for high prices.
Sure, servers (massive paraell requests) and encoding benefit greatly from them. And you can even carve some benchmarks to show speed-up, so, bingo, let's sell them not only for servers, but for general public too, jay!
aaronjb -
There is very real and often very expensive (development-time wise) coding that needs to be done to make efficient use of mulitple cores.
Carve that to stone and bang with it some people 24/7 till they understand how hard it is to paraellize stuff. Meanwhile correct me w/o mercy, I know that I'm not entierly right, however the post lenght is increasing way too much with proper explainations... and for some people this is just a waste of time, since if they come with "arguments" like "they are doing it, so it must be good" - come on, that is not worth reply.
Moto7451 -
Neither is your email client or most of the programs people use. Even if they could gain performance its just not worth the money necessary to implement it.
Wait, it is actually worse. If they implement it (and it is very costy and time consupting and the gain will be hardly noticable), it cause significiant slow-down on single core
4+ cores will be meaningless to those who just surf the web and type reports unless the marketing department at Intel, AMD, and the PC companies earn their pay.
Wait, they already conviced some that it will be faster!
So they earn their paychecks pretty well
BMac420 -
the program knows how many cores its dealing with it could then split the number of objects up equally
True. But how many object you need to:
1) execute in parallel
2) have absolutely NO relationship to other objects?
The second problem is the very essence of multiprocessing. You can paraellize only processes, that using data that are not altered in any way by these other processes you just distributing to the separate cores.
word processing and web surfing programs should never go multithreaded because its not necessary for the little jobs they do
It is not only not necessary, it is very impossible to do. All I can imagine is that flash plugin will run on other core, while the browser run on other core... But it is need on current CPU's? Nah, most of the time the flash wait in loops, so...
more intensive software for graphics, vid, 3D rendering.. that kidna stuff could benefit a whole lot more, but at the expense of program size, photoshop CS13 is gonna be a 27 gig install. you just wait
Clearly there is the problem. Size grow. And ask yourself, considering the size. Do the memory speed grow with it? No? Then how things can be faster, when they only grow in size, yet the ram is pretty slow, no matter the marketing BS we hear all over.
Yes, I'm very skeptical in the core way of slow-down done todays. Yes, there are speed-up things, but for home user I see a little gain. Except there is a big gain for the servers and companies that is convincing us that more cores = better... Eh, what a scam.
Anyway, one user mentioned (on O/C forum!) that there is "Mhz wall". No, there is not. We all know that these Conroes are not only damn fast, but they are also damn well overclockable, so what wall?
Each wall in history was overcomed by better technology and we aren't at the limits clearly yet. Nowhere near of them. If some can clock Conroe at 6Ghz, then untill 6Ghz is sold for 200 bucks, then I see no need for more cores - except for encoding and servers.
I think that the PR magic primary concern is to convince users that more core = better to make these users bear most of the cost to develop server multicore CPU's. So the servers got their CPU's cheaper, because enough people - that do not need them and can't utilize them - buy multicore CPU's even with their prices are way higher and unjustified for the little slow down in some applications and little speed up for another ones - or tell me, how much will my FPS in BF2 increase on dual core on same clock verzus single core???
