GPU or CPU?

[anishtain4]

First I didn't know where should I post this topic and I didn't managed to find a search option for the forum to see if there has been a similar discussion before or no?

As you see GPU for engineering and scientific computation is becoming so common in discussions but I wanted to know if that is really going to be a part of the computational physics for next decade? Please answer if you have direct contact and expriments with GPU instead of quoting some reviews here.

Thanks

[anishtain4]

Any idea?
No one hasn't touched GPU here?

[Bernard]

Only as a huge co-processor to offload main CPU from SIMD intensive floating-point and integer calculation.Still CPU is more adept than GPU in some fields.For example Cache and memorymanagement,branch prediction and clever out-of-order execution and also spotting non-dependent instructions in the code stream and exploit paralelization of serial code.

[gaston-hillar]

Have you tried Intel OpenCL compiler and Intel OpenCL development environment + tools.

IMHO, if you're interested in CPU vs GPU, you should definitely take a look at Intel OpenCL. :)

http://software.intel.com/en-us/articles/vcsource-tools-opencl-sdk/

[Bernard]

Have you tried Intel OpenCL compiler and Intel OpenCL

@Gaston
What is your opinion on Nvidia Cuda?Because I'am considering to learn GPGPU programming.

[gaston-hillar]

We're in a thread that is part of the Intel Parallel Studio forum. I'm not an Intel employee but I'm a big fan of all the exciting hardware and software that Intel produces. I do believe Intel OpenCL tools are very interesting to use your existing OpenCL knowledge to run the code on powerful Intel CPUs.

If you're interested in GPU programming, I do believe you should at least consider Intel's work on its OpenCL products.

I'm not eager on talking about Intel's competitors on Intel forums. I just provide my thoughts about Intel hardware and software in these forums. :)

Hope it helps.

[gaston-hillar]

If you're interested in Intel OpenCL, there are a few Webinars going to happen in the forthcoming days. :)

Take a look at the additional information here:

http://software.intel.com/en-us/blogs/2012/06/10/webinar-series-opencl-programming-on-3rd-gen-intel-core-processors/

[Bernard]

Thank You for those links.

[Bernard]

@Gaston
I completely agree with you it is not a good place to discuss such a question.
I have already invested in abook on CUDA programming and I'am going to buy high-end
GPU so decision has been made :)
But I will give a try to OpenCL technology.

[gaston-hillar]

@iliyapolak

Cool!

Based on my experience, in many cases, I've been surprised by the performance achieved with Intel OpenCL technology, and therefore, I've decided to move from GPU back to CPU and costs were lower.

However, of course, it is important to dedicate the necessary time to test the algorithms that will consume 80% of your computing time in your applications in order to draw accurate conclusions.

I'm a big fan of Intel OpenCL for certain kinds of applications, and therefore, I'm usually very enthusiastic in recommending its evaluation whenever there is a chance to use OpenCL code. :)

Cheers,

Gaston

[Christos_Amarandos]

GPU is what I would personal recommend (be up with the times)