Intel 32nm

[MpG]

There are also many kinds of programs that simply can't be broken down any further. If operation B is based on a result from operation A, you can't calculate them at the same time (I understand that AI programming often falls into this category). So while there's a great deal of improvement to be done in the area of parallel programming, there are limits to how far it can be taken. That said, we've got a long ways to go before we legitimately reach that point.

[bojangles]

Originally Posted by MpG

There are also many kinds of programs that simply can't be broken down any further. If operation B is based on a result from operation A, you can't calculate them at the same time (I understand that AI programming often falls into this category). So while there's a great deal of improvement to be done in the area of parallel programming, there are limits to how far it can be taken. That said, we've got a long ways to go before we legitimately reach that point.

Yup. That is a major problem. It's simply a problem that nature has provided us, and cannot be fixed. Those dependencies are inevitable and we, as humans must wait for them.

[Mr.Blah]

Originally Posted by bojangles

Kind of. That's what pipelining was for. It was a way to act like parallelism, but also behave like a linear process. It allowed instructions to fetch, execute and store while others were doing the same thing on a different clock cycle. It solved most of the shared memory problems and wait times to store data in a cache or the RAM. This is why GPUs have more processing power than CPUs these days.

The CPU still uses a very large cache array, while the GPU uses a very complicated pipeline architecture. However, CPUs are best for solving static instructions like math calculations and things that are already written in stone (C++ code, etc), which is why the cache plays a vital role when fetching data and instructions. I don't know how the cache is set up in today's chips, but what I know is that they do separate the L1, L2, and L3 cache in such a way that it allows higher priority to either data or instructions, depending on what the processor is designed for (I'm talking the construction of any CPU, not just a Pentium, Phenom, etc.).

The GPU are extremely valuable when it comes to dynamic instructions, etc. because it needs to calculate rapidly changing variables, like in video games. Because of its pipelining architecture, it can solve vector algebra much quicker than a CPU (solving an N x N matric by a CPU is an order of N^2), but parallelism can solve that much more quickly), which is why GPUs are becoming widely used in solving highly complex environments. It is also why we put multiple CPUs in a server. However, with today's technology, we can push a higher amount of numbers out with less GPUs, than an array of CPUs.

A computer is stupid. Seriously. It can only do one thing, but it can do it really well.

We just have to figure out where we can do it the fastest.

Wow thanks, I actually understood part of it too haha.

To the bold, computers were only made to do that one thing in the first place anyway so I'd say we acomplished the goal and then some. Everything else is just gravy, sort of like building a toaster and one day finding out you can play games on it.

[The Guy]

I can't agree anymore. Having some programming experience I know how hard it is to take use of multiple cores; I have only dug up the surface and found it immensely difficult. This is largely related to the requirement of A before B as mentioned before. Its quite sad that we have come such a long way to have only made a glorified calculator and the fact we need to go organic for smaller CPUs almost makes computer science biology class (Oh I dread the day when I am in Varcity).

Originally Posted by bojangles

Kind of. That's what pipelining was for. It was a way to act like parallelism, but also behave like a linear process. It allowed instructions to fetch, execute and store while others were doing the same thing on a different clock cycle. It solved most of the shared memory problems and wait times to store data in a cache or the RAM. This is why GPUs have more processing power than CPUs these days.

The CPU still uses a very large cache array, while the GPU uses a very complicated pipeline architecture. However, CPUs are best for solving static instructions like math calculations and things that are already written in stone (C++ code, etc), which is why the cache plays a vital role when fetching data and instructions. I don't know how the cache is set up in today's chips, but what I know is that they do separate the L1, L2, and L3 cache in such a way that it allows higher priority to either data or instructions, depending on what the processor is designed for (I'm talking the construction of any CPU, not just a Pentium, Phenom, etc.).

The GPU are extremely valuable when it comes to dynamic instructions, etc. because it needs to calculate rapidly changing variables, like in video games. Because of its pipelining architecture, it can solve vector algebra much quicker than a CPU (solving an N x N matric by a CPU is an order of N^2), but parallelism can solve that much more quickly), which is why GPUs are becoming widely used in solving highly complex environments. It is also why we put multiple CPUs in a server. However, with today's technology, we can push a higher amount of numbers out with less GPUs, than an array of CPUs.

A computer is stupid. Seriously. It can only do one thing, but it can do it really well.

We just have to figure out where we can do it the fastest.

[Mr. Cipher]

Having minimal programming experience (pascal, java, VBA - all for finance not gaming uses), I find it hard to understand why portions of games cannot be assigned to one thread, while another thread handles the "wait until A is complete to do B" set of instructions that prevents us from moving very far with multi-thread programming, as you suggested above.

For example, couldn't one thread handle static stuff like weather, time of day while another handles some AI? Or can threads not be interrupted by another thread once started?

[Generic User #2]

Originally Posted by Mr. Cipher

while another thread handles the "wait until A is complete to do B" set of instructions that prevents us from moving very far with multi-thread programming, as you suggested above.

i'm not entirely sure if i get you...but here goes.

you can multi-thread all you want, but if the in-order process takes longer than the out-of-order processes, you're just wasting effort.

as an example, there is an in-order process takes 10 seconds and there are three out-of-order processes that take 5 seconds each.

in a single-threaded application, the program would take 25 seconds to run. in a dual-threaded environment, it would take 15 seconds to run(the IO in one thread and OOOs in another thread). with 3 threads, it would take 10 seconds(IO in one, two OOs in another, and the 3rd OO by itself). now....we COULD enable 4-threads....but that ould be pointless since the run-time would till be 10 seconds.

theres a whole lot of other technicalities and workarounds and stuff, but thats quite a bit beyond me....

ps, there is a difference between processes and threads...i'm just being lazy with my terminology.