After quad core?

[∞Velocitymaster∞]

I hope next year around june quad core should drop by half the price or more. I have P4 3.2ghz and im building a new rig mid 07 hoping to get DDR3 as system ram and quad core.

After quad core will be 6 core or 8 core. Or something like new socket size and FSB speed to handle future cpus?

 

[yaric]

Well eventually they're going to mini cores, where there are 32 cores on one chip, then 64 and so forth.

 

[[LYL]Homer]

I would imagine 8, 16, 32, 64, 128, etc.

 

[∞Velocitymaster∞]

I would imagine 8, 16, 32, 64, 128, etc.

thats just what i was thinking, thats how they usually do, like in ram also. computer is binary "2" so they double each time like 16, double of that is 32 and so on.

Now the this what I want to know too, I hope they keep the same socket LG775 cause then if they have 8 core with different socket, means new mobo

 

[tpfaff]

I can't imagine an actuall program using that many cores, besides an intense multi core optimized game. Perhaps an extremely nice looking os, but idk I just don't see a non gamin/visual app needing this much computing power, for the average consumer pc buyer I just don't see why they'd want to spend their money on a 128 core processor.Of course their will be programs and a dual core processor will be laughed at, I just can't really imagine it

 

[acetic]

I can't imagine an actuall program using that many cores, besides an intense multi core optimized game. Perhaps an extremely nice looking os, but idk I just don't see a non gamin/visual app needing this much computing power, for the average consumer pc buyer I just don't see why they'd want to spend their money on a 128 core processor.Of course their will be programs and a dual core processor will be laughed at, I just can't really imagine it

The only real purpose is that eventually you'll hit a wall of how small and fast you can make your processors.. The answer to that if you want more processing power is, well, more cores. It's a simple way around a rather large wall.

 

[jchen012]

I don't think we'll be seeing 8 cores for awhile. After quad core, they will have to release a 45nm version, and then after that, a FSB1333mhz version. Then maybe a 8 core cpu =P

 

[NulloModo]

I can't imagine an actuall program using that many cores, besides an intense multi core optimized game. Perhaps an extremely nice looking os, but idk I just don't see a non gamin/visual app needing this much computing power, for the average consumer pc buyer I just don't see why they'd want to spend their money on a 128 core processor.Of course their will be programs and a dual core processor will be laughed at, I just can't really imagine it

While I don't see 128 core processors in the next decade, I do think multi-core is the only way we can continue the trend in CPU performance for the forseeable future. We are looking at a paradigm shift in programming at the same time we are seeing it in the engineering and design of CPUs.

Multi-core is just getting out of its infancy, programmers are going to have to learn to optimize for multi-core no matter what the application to survive in the future.

 

[fenton06]

The only real purpose is that eventually you'll hit a wall of how small and fast you can make your processors.. The answer to that if you want more processing power is, well, more cores. It's a simple way around a rather large wall.

The OTHER way around the wall would be to get optical computing perfected. (using photons instead of electrons) WE're talking about insane performance increases because of how much faster photons are.

 

[Astrogiblet]

Dont they already have 64 and 128 core processors for business grade servers? Like real high end ones.. I seem to remember reading an article about this..

 

[Astrogiblet]

Here is a 128 core processor, used for extreme networking situations:

http://www.commsdesign.com/news/product_news/showArticle.jhtml?articleID=170704430

 

[∞Velocitymaster∞]

most apps are good on 8 core but after that, you are not gona see performance increase,
Before they were keep increasing the clock speed to like 3.6 or 3.8 max on the P4, till the first dual core came out "Pentium D" it stop climbing.

I know high clock speed = more heat and power, I think there is a need of increase of FSB and the North bridge on the mobo architecture to handle more data.

 

[Scali2]

I can't imagine an actuall program using that many cores, besides an intense multi core optimized game. Perhaps an extremely nice looking os, but idk I just don't see a non gamin/visual app needing this much computing power, for the average consumer pc buyer I just don't see why they'd want to spend their money on a 128 core processor.Of course their will be programs and a dual core processor will be laughed at, I just can't really imagine it

There is already software that makes use of as many cores as you have.
The only problem is the diminishing returns. At a certain point there's more overhead for adding another core to the calculations than the time it takes to perform the calculations themselves.
But audio/video encoding and 3d rendering are some areas that scale extremely well to multiple cores, processors or systems.
A renderfarm at Pixar already has hundreds of processors working together.

Games are actually a pretty bad case for multicore, because they consist of many dynamic/interactive processes that depend on eachother, and therefore cannot be processed in parallel. I would expect games to hit the point of no return much faster than other software.

 

[AntiSocialMunky]

Actually, games can have many many systems independent of the standard graphics output/main game/sound threading.

You can allocate additional cores for physics especially for independent systems(eye candy) like smoke effects, bubbles, realistic movement, or more sophisticated AI... etc.

The sky's the limit with games, and they'll always be power hungry as long as there's something that's not as real as possible... which is alot.

 

[Scali2]

You can allocate additional cores for physics especially for independent systems(eye candy) like smoke effects, bubbles, realistic movement, or more sophisticated AI... etc.

How is that independent?
If you want smoke and bubbles etc done correctly, you have to take all objects in account (smoke curling around objects, being sucked away by the wash of an object passing by etc).
Same with movement and AI.
Objects have to move around eachother, or collide. And with AI, objects need to interact with eachother. Just think about it... if one actor talks to another, the actor has to look at the other actor, and the other actor has to face the actor that addresses him. Which means AI depends on movement, and movement depends on AI.

 

[AntiSocialMunky]

It is really dependent on the way it is implemented and what type of threading methods you run into and whether or not you allocate a process to a die or whether you allocate several dies for one process - things like that. Valve was experimenting with multithreading particle systems, AIs, and subprograms. Its somewhere on Anandtech.

 

[Scali2]

It is really dependent on the way it is implemented and what type of threading methods you run into and whether or not you allocate a process to a die or whether you allocate several dies for one process - things like that. Valve was experimenting with multithreading particle systems, AIs, and subprograms. Its somewhere on Anandtech.

It's not about processes and dies at all.
The entire engine is a single process.
The Valve engine is just capable of using multithreaded algorithms for certain operations, and it can vary the number of threads per algorithm.
That doesn't solve the dependencies, because they cannot be solved. But it does allow for some multithreading where previous engines did everything in a serial fashion.
I'm just trying to make the point that even though it uses multiple threads/cores, it's probably not going to scale all that well to many cores, because there's just so much dependency inside.
What it does do, however, is that it increases the potential processing power available to each independent piece of work. Think of it like hardware T&L. It won't increase the framerate, but it will allow you to push more triangles at the same framerate.

In the case of physics that would mean something like you can't have more particles bumping into eachother at the same time, but you can have more 'clusters' of bumping particles.

Another nice example is the skeleton-animation. You can't have more objects interacting with one object (like one person aiming at the other), but you can process multiple objects that are interacting with one other object.

So it does scale, but mainly in the independent parts of the dependent process, not across the board... However, that means that you need to do extra work in order to figure out what is dependent and what is not.

The other examples, like media encoding or 3d rendering, are generally pretty much completely independent, and will therefore scale far more linearly with the amount of cores you add.
It's quite trivial to see how it can be split up aswell... like a movie or audio file, you can cut it into pieces, let each core encode a piece, and glue the results back together.
With 3d rendering, you can split up the screen in multiple regions, and let each thread render a region (much like what SLI/CrossFire do, but with offline rendering the balance between per-frame setup time and rendering time is generally far different, so it's much less of a deal when things like animation or physics cannot be split up that efficiently).

So the point is not what threading model or implementation you choose, because there really isn't much of a choice. Valve's hybrid threading is the only way to get any kind of gain from algorithms that aren't very parallel by nature.