Thread: GPU the basis of future supercomputers?

Many people will it hard to believe how can a non graphics computing can be done on GPU. But they can be done & much faster using GPU. This methodology is denoted by famous term called GPGPU i.e. General Purpose Computation on Graphics Processing Unit.
To understand this first to the basic difference between the architecture of GPU and CPU.
As we know that CPU takes one information then process & then take processing of next information. Or in other words it take information serially. It is quite a simplified picture. I do know about Parallel computing. But this is insignificant next to the power of GPU.
Now looking at GPU, it take a number of information all at once & process them. It is because it has many processors.GPU of now are basically designed to support large number of floating point opertaions i.e. operations concerning numbers too small or large to be represent as integer.
Some would like to think that CPU of today also have many processors. They might seem to have many cores but not much. GPU are the king of floating point operations ( flops) and will always be as far as I can see.
Example :
Ati Radeon 5750 processing power is 1.0008 teraflops. To appreciate it more Intel core i7 the fastest CPU for all is just 70 gigaflops. Puny na? Also take into account the power used may be more for Ati card but wen you think of ratio of what core i7 is delivering to consuming it will seem to be trifle.
Of course , many will have their pulses racing now but you will always need CPU for sequential instructions.GPU can't do them. They are designed for graphics & things like that which require floating point computing.
So if you want to programmed you have to represent your problems in terms of graphics. & also not forgetting you will also need to be a champ of directx or opengl to represent them. But people were doing that.
But the things changed with the advent of CUDA by Nvidia in 2007. Nvidia unleashed its GPU based Tesla supercomputers. It was not fastest but its cost & power were a major factor for utilizing it.
Now you can design & compute in programming language as you do not as some graphics problem in term of floating numbers.
But there was still a catch , CUDA was just for Nvidia.
So Intel ( having 49.4% market share) ,IBM,Nvidia (27.8% ) and ATI (20.6% ) decided to joined Opencl developed by Apple under Khronos Compute Working Group.
Now supercomputer using GPU are becoming reality and are already being applied to biology , geology & , other mathematical & scientific fields.

GPU the basis of future supercomputers? by Luke Skywalker of India Broadband is licensed under a Creative Commons Attribution-Noncommercial-Share Alike 2.5 India License.

Originally Posted by Luke Skywalker

Many people will it hard to believe how can a non graphics computing can be done on GPU. But they can be done & much faster using GPU. This methodology is denoted by famous term called GPGPU i.e. General Purpose Computation on Graphics Processing Unit.
To understand this first to the basic difference between the architecture of GPU and CPU.
As we know that CPU takes one information then process & then take processing of next information. Or in other words it take information serially. It is quite a simplified picture. I do know about Parallel computing. But this is insignificant next to the power of GPU.
Now looking at GPU, it take a number of information all at once & process them. It is because it has many processors.GPU of now are basically designed to support large number of floating point opertaions i.e. operations concerning numbers too small or large to be represent as integer.
Some would like to think that CPU of today also have many processors. They might seem to have many cores but not much. GPU are the king of floating point operations ( flops) and will always be as far as I can see.
Example :
Ati Radeon 5750 processing power is 1.0008 teraflops. To appreciate it more Intel core i7 the fastest CPU for all is just 70 gigaflops. Puny na? Also take into account the power used may be more for Ati card but wen you think of ratio of what core i7 is delivering to consuming it will seem to be trifle.
Of course , many will have their pulses racing now but you will always need CPU for sequential instructions.GPU can't do them. They are designed for graphics & things like that which require floating point computing.
So if you want to programmed you have to represent your problems in terms of graphics. & also not forgetting you will also need to be a champ of directx or opengl to represent them. But people were doing that.
But the things changed with the advent of CUDA by Nvidia in 2007. Nvidia unleashed its GPU based Tesla supercomputers. It was not fastest but its cost & power were a major factor for utilizing it.
Now you can design & compute in programming language as you do not as some graphics problem in term of floating numbers.
But there was still a catch , CUDA was just for Nvidia.
So Intel ( having 49.4% market share) ,IBM,Nvidia (27.8% ) and ATI (20.6% ) decided to joined Opencl developed by Apple under Khronos Compute Working Group.
Now supercomputer using GPU are becoming reality and are already being applied to biology , geology & , other mathematical & scientific fields.

Hmm, interesting, so I can buy a gpu & run application on it right? with help of CUDA etc.

Yes on latest graphics cards. but cuda support only Nvidia. Oh also you will have to write the application yourself. Because common application of today are not designed that way.

I heard Adobe CS4 supports gpu acceleration...

CS4 is a gpu application as it needs floating point operations. But to run say your browser on it , you will have to make it.

Originally Posted by Luke Skywalker

CS4 is a gpu application as it needs floating point operations. But to run say your browser on it , you will have to make it.

I mean to say earlier versions of Adobe Photoshop were doesn't supported GPU acceleration, but CS4 does, I also read somewhere new version of flash player supports GPU acceleration...

GPU acceleration is another thing. Earlier games too didn't supported GPU acceleration but modern do. GPU work on CPU was always done on CPU when GPU was not invented.What a CUDA coded program do is that it used GPU instead of CPU. For example doing a search require that google use CPU but you can do using GPU also. That is what is GPGPU all about.

Originally Posted by Luke Skywalker

GPU acceleration is another thing. Earlier games too didn't supported GPU acceleration but modern do. GPU work on CPU was always done on CPU when GPU was not invented.What a CUDA coded program do is that it used GPU instead of CPU. For example doing a search require that google use CPU but you can do using GPU also. That is what is GPGPU all about.

Ok, now I got it.

The GPU, yes, your humble graphics card. More and more, these highly specialised processors - paired with very fast memory - can perform the complex calculations needed for fast rendering of physically realistic scenes in games, for example. So, not only the actual graphics, but also the physics of the world that is being shown can be processed here. And because these units are produced in large numbers for the mass market they are unbeatably cheap. A few hundred dollars give you computing power that could only be had for tens of millions of dollars just ten years ago.

Originally Posted by lockgold20

And because these units are produced in large numbers for the mass market they are unbeatably cheap. A few hundred dollars give you computing power that could only be had for tens of millions of dollars just ten years ago.

This can be also said about CPU also.

GPU (Graphic Processing Unit ) is already being implemented in Game Console like PlayStation 3.

Originally Posted by Sam Bentley

GPU (Graphic Processing Unit ) is already being implemented in Game Console like PlayStation 3.

All know that, It is implemented there before past couple of years.

There is basis that GPU will be a mainstay for future supercomputers. Modern GPUs are very efficient at manipulating computer graphics, and their highly parallel structure makes them more effective than general-purpose CPUs for a range of complex algorithms. In a personal computer, a GPU can be present on a video card, or it can be on the motherboard. More than 90% of new desktop and notebook computers have integrated GPUs, which are usually far less powerful than those on a dedicated video card

nice one luke, rep+.


ATI has stream, Nvidia has CUDA, but we really need more applications. A ATI 4850 has a LOT more computational power than a core i7 in raw numbers.

Thanks Rajan for rep. Didn't saw that any one had replied to my threads. So didn't said earlier.

A book had been released on gpgpu:Amazon.com: Programming Massively Parallel Processors: A Hands-on Approach (9780123814722): David B. Kirk, Wen-mei W. Hwu: Books

I didn't know that this articles rank on google unless i searched for it.