Take the 2-minute tour ×
Super User is a question and answer site for computer enthusiasts and power users. It's 100% free, no registration required.

I am doing a research about the difference between CPU and GPU parallel processing.

So in bref, nvidia GPU card is able to do a massive parallel processing, by using threads. each GPU can contains at least 1024 threads or plus. the GPU architecture is composed by grid and each grid contains 16 clusters or blocs of threads. the thread is able to treat one instruction at a time.

the maximum CPU core now, is 8 core.

my question is : why today, we dont have a card who hthing as the same architecture as the GPU nvidia but by using CPU and not threads? i am asking this because i need to know why it is difficult to do a card like that, is there any technical who made doing the same architecture as the GPU but with CPU?? (transistor, synchronisation,...)

thanks in advance

share|improve this question

closed as unclear what you're asking by Nifle, Tog, ncdownpat, Simon Sheehan, Moses Oct 25 '13 at 1:49

Please clarify your specific problem or add additional details to highlight exactly what you need. As it's currently written, it’s hard to tell exactly what you're asking. See the How to Ask page for help clarifying this question.If this question can be reworded to fit the rules in the help center, please edit the question.

2  
I don't understand your question. Are you asking why CPUs aren't using massively parallel architectures? –  gronostaj Oct 23 '13 at 12:08
    
@ALi - There are processors with the capability to run 16 threads ( 8 physical cores and 8 HT cores ) currently on the market. You do understand that the main purpose of a GPU is to do floating point operations right? Your question doesn't make sense. –  Ramhound Oct 23 '13 at 12:23
    
sorry for my bad english, I wanted to say that seeing the evolution of graphics cards in front of the CPU, we find that the graphics card is much more improved. My question was, what difference in architecture between the two that allows GPU to evolve more rapidly technical standpoint? for example, apparently it is easier to add a GPU threads than adding a CPU? –  ALi Oct 23 '13 at 12:51
    

1 Answer 1

GPU and CPU serve different purposes, and different purposes usually mean different solutions.

GPUs are massively parallelized, but they are used for completely different operations than CPUs. Those operations can be processed in parallel with good results. But not every algorithm can be parallelized. Parallel processing brings a lot of new problems that have to be solved in some way. Sometimes the overhead from those solutions is so big that it exceeds the performance gain. And some algorithms just can't be parallelized.

Multithreading isn't an ultimate solution for gaining better performance. It works in situations where multiple parallel operations have to be performed, like on database servers etc. But at home you couldn't make use of massively parallelized system. Your everyday tasks work well on few cores, a powerful single-core CPU would probably handle it too.

Now let's say you have replaced your 4-core CPU with 256-core one. More cores come at a price: each core requires additional power and more sophisticated cache mechanisms have to be implemented, thus increasing CPU's complexity. You have 64 times more cores, so each of those will be 64 times slower. Instead of 2.4 GHz per core, now you have 37.5 MHz per core.

What happens? Your system is consuming way more power and giving much more heat. If you try to use Windows 8, it's crawling. Booting takes forever, despite hybrid boot being turned on. You try Windows 98 and it's still too slow to use it. Most of your cores aren't used at all. Few are working at 100%, but it's still just 37.5 MHz (versus 2400 MHz on your previous CPU).

Of course that's an extreme example, but it's a real issue. Actually, there's some interesting discussion on this topic going on currently: Samsung has presented Exynos Octa, an eight-core mobile CPU family based on big.LITTLE architecture, and here's the Qualcomm's response.

share|improve this answer

Not the answer you're looking for? Browse other questions tagged or ask your own question.