I still think that both CPU and GPU have their part to play but CPUs are "slowing down". GPU technology is exploding (read: developing a LOT faster) than CPUs and this I think will have a strong bearing on where things go in the near future.

One thing I am concerned about is being locked into the video technology of your CPU. For example, now with discrete video cards you can always upgrade to a better videocard , provided your CPU is still up to par. What happens when CPUs are shipped with GPUs? Granted their is a boost in speed if they are on the CPU die but can you have discrete cards as well for video, to complement the onboard video? If not, what happens after 2-3 years when newer programs and games demand more resources than the onboard chip can provide?

Then we'll have to do what we've always done........upgrade. LOL!

While it is always good to see decent improvements in the speed of CPU's and GPU's. I'd say that at present, they are (generally speaking) fast enough and what we really need is serious improvements in system memory bandwidth/speed.

Looks like the ultimate marriage of CPU and GPU will happen someday maybe in the next decade, but it's already getting there gradually with the new desktop APU's (Clarkdale, SB, Llano).

But it might not happen with x86 CPU's if ARM (or some other tech) do emerge as the superior. ARM is already superior for ultra low-power devices like smartphones, but when will it happen for desktops? I think Microsoft will really need to support ARM processors for Windows apps (with emulation/virtualization of x86 code for compatibility), without letting Intel bribe them dearly.. I dunno??

One crucial step towards the marriage is when APU's become common enough that all GPU's learn how to cooperate and also run completely idle, letting the APU do the 2D work.

I doubt that GPUs will replace CPUs for supercomputing. They're very powerful at simple and primitive calculations, but lack of flexibility. I think that it would work fine in tandem with CPU, that's why nShitia is desperate of creating a good CPU, a pity that their Arm approach will take a long time to catch up in performance compared to Intel due to its mobile CPU roots characteristics.

Clever, huh? You think intel/amd didnt realize that ppl were upgrading GPUs over and over while stretching their CPU as long as they could. In a lot of ways the GPU/CPU combo is a dirty corporate act. In the end i bet intel will become very very close to nvidia, the company they tried to squeeze out of the market. There is no way they are gonna let AMD gain on them and once they start....nvidia is their quickest solution.



actually 3 of the top 5 supercomputers are hybrid GPU/CPU systems. The cpu cores schedule and sort while their GPUs do most of the crunching. Your kinda way off on this one. It will not be long before the top 10 supercomputers will all be hybrid. These supercomputers are all the talk and the software is already there. You cant build an all CPU system with the anywhere near the efficiency and that is why hybrid systems are taking off. You probably didnt know or think of the power requirements issues for a supercomputers. The top x86 systems need their own power plant to operate.

The most powerful computer in the world is hybrid and it is many times more efficient anything a CPU only system can be. This is why hybrid systems will remain important to supercomputing for years to come.

According to whom? I spoke with some developers over a month ago and pretty much they agree on the same thing, how long nVidia has been promoting GPGPU? And yet I haven't seen lots of applications or implementations of super computer using it. It isn't about TFLOPS or PFLOPS, it is about how much can you untap or unleash from, and GPU's are embarassingly parallel, rendering graphics will work like a champ because they don't use lots of dependant code or branch prediction stuff, but they're not flexible compared to CPU's. Think of GPU's as in the case of nVidia's GF110, 512 train rails with one train on each capable of carrying over 1,000 pounds of coal, but they can't trade coal each other to maximize space utilization, crossover rails or move backwards .

A super computer with x86 for example, think of 128 train rails with one train each other capable to carry over 2,000 pounds and are capable to trade coal each other, move backwards or even crossover to minimize stalls. Remember that current x86 cores are much fatter, flexible and smarter than current GPGPU cores like nVidia or AMD, that's why SP Cells couldn't take off either. Each Nehalem core for example, is much faster than each CUDA core, but nVidia's advantage is being able to fit a lot of these in a good TDP. Hybrid systems are the future indeed, but right now software can't catch up with hardware with smart scheduling, let alone lots of dumb cores. That's why nVidia is looking for Arm core to help and feed the GPU's for supercomputing, the same for AMD which has the advantage of using much more powerful CPU's and use stuff like the command queue processor to help maximize the usage of their stream cores on their line up of videocards.

http://www.techpowerup.com/145914/New-N ... ation.html
http://www.nytimes.com/external/gigaom/ ... ss&emc=rss

Are you being serious? Or joking? i really can't tell and so i will speculate.



Fixed. and i get it now.

Just posted links for news? I expected a smarter response from you but that's ok, it is good to know that Hybrid computing is taking off.

And no, I spoke with open source developers who doesn't have any affilliation to hardware vendors, so stop spreading FUD and bullshit, I don't have any relationship with AMD, if I did, I would gladly place it on my signature. I pay for my hardware and my stuff just like any regular customer, something that you get for free in exchange of forum bias and lack of imparciality.