IPP to use GPU

IPP to use GPU

Ritratto di rohitspandey

HI,

 

Does the IPP libaries can be enabled to use openCL to utilise GPU.

 

Regards

Rohit

19 post / 0 new
Ultimo contenuto
Per informazioni complete sulle ottimizzazioni del compilatore, consultare l'Avviso sull'ottimizzazione
Ritratto di Jeffrey Mcallister (Intel)

Hi Rohit,

Today IPP is CPU only. However, this is a topic that the IPP team is very interested in. If/when there are any updates on GPU utilization we'll broadcast here and elsewhere as soon as possible.

Thanks,
Jeff

Ritratto di Sergey Kostrov

Hi everybody,

>>...Does the IPP libaries can be enabled to use openCL to utilise GPU...

I think this is a very interesting subject and I wonder if you had a chance to review what was done by NVIDIA:

NVIDIA Performance Primitives - https://developer.nvidia.com/npp

Ritratto di iliyapolak

Quote:

Sergey Kostrov wrote:

Hi everybody,

>>...Does the IPP libaries can be enabled to use openCL to utilise GPU...

I think this is a very interesting subject and I wonder if you had a chance to review what was done by NVIDIA:

NVIDIA Performance Primitives - https://developer.nvidia.com/npp

One step closer from creation of powerful CPU-GPU "monster".

Ritratto di iliyapolak

Still CPU has some advantage mainly in areas of memory , cache management,out of order execution , speculative execution and exploiting instruction level parallelism.
I think that in the foreseeable more advanced GPU's will some kind of "auxilliary" CPU used residing on board and which will be used instead of standard CPU.

Ritratto di rohitspandey

Hi,

Thanks for the update!!. Definitely like to see its being used by IPP and MKL libraries. Our applications can heavley leverge from it.

Regards
Rohit

Ritratto di Eric B.

I would also like to point out that IPP is according to this going to be compatible with the Intel PHI. Close to the same context as GPU's
Reference:
IPP Coprosessor

GPU's are still better on the price point but on the high end not by much.
Also the IPP eq for CUDA is NPP -> just google cuda npp
it comes with the SDK just FYI

Ritratto di Jeffrey Mcallister (Intel)

Unfortunately Xeon Phi isn't officially supported by IPP. I'll see what I can do to update the article.

Ritratto di Eric B.

Bummer!
Is it planned to be?

Ritratto di Jeffrey Mcallister (Intel)

At this point there are no official commitments, so we don't want anyone to make plans on it being there. However, knowing about your interest is always helpful for product planning.

Ritratto di Sergey Kostrov

>>At this point there are no official commitments, so we don't want anyone to make plans on it being there. However, knowing about your
>>interest is always helpful for product planning.

There are ideas around on adaptive hybrid high-performance algorithms and APIs. A main point is they should use CPU and GPU ( not at the same time! ) dependending on a state of processing, or algorithm, or function etc.

Ritratto di Sergey Kostrov

>>At this point there are no official commitments, so we don't want anyone to make plans on it being there. However, knowing about your
>>interest is always helpful for product planning.

There are ideas around on adaptive hybrid high-performance algorithms and APIs. A main point is they should use CPU and GPU ( not at the same time! ) dependending on a state of processing, or algorithm, or function etc.

Ritratto di iliyapolak

>>>There are ideas around on adaptive hybrid high-performance algorithms and APIs. A main point is they should use CPU and GPU ( not at the same time! ) dependending on a state of processing, or algorithm, or function etc.>>>

Has not been this already implemented.IIRC some russian company was one of the first to use hybrid approach for cracking WPA2 cipher suite.

Ritratto di Sergey Kostrov

>>...Has not been this already implemented...

I wouldn't say so because we simply don't know what is going on inside of thousands of software companies. Application of adaptive hybrid high-performance algorithms which use CPU and GPU is a logical step in evolution of different kinds of algorithms. Also, I prefer to keep silence on what algorithm(s) I would apply that concept first of all.

Ritratto di iliyapolak

Quote:

Sergey Kostrov wrote:

>>...Has not been this already implemented...

I wouldn't say so because we simply don't know what is going on inside of thousands of software companies. Application of adaptive hybrid high-performance algorithms which use CPU and GPU is a logical step in evolution of different kinds of algorithms. Also, I prefer to keep silence on what algorithm(s) I would apply that concept first of all.

Yes I tend to agree with you, because by siimple logical reasoning it is impossible to know exactly what kind of research is beign done on hybrid CPU-GPU algorithms.
I found this great website dedicated partly to CPU-GPU algorithms and to gpgpu programming :http://hgpu.org/?p=6140

Ritratto di iliyapolak

I also plan to invest my time and effort in learning CUDA programming.I think that today and even more in the foreseeable future it will be one of more needed looked for skills.

Ritratto di rohitspandey

Any time lines for supporting openCL or any other implmenetation to use GPU by IPP function.

Ritratto di Sergey Kostrov

>>...gpgpu programming: http://hgpu.org/?p=6140

Abstract is interesting however it is an example of a highly specialized algorithm. I was talking about general purpose algorithms, that is more widely used by software developers.

Ritratto di iliyapolak

Quote:

Sergey Kostrov wrote:

>>...gpgpu programming: http://hgpu.org/?p=6140

Abstract is interesting however it is an example of a highly specialized algorithm. I was talking about general purpose algorithms, that is more widely used by software developers.

You can browse that website and try to look for more general purpose algorithmic implementation.Do not pay attention to abstract.

Accedere per lasciare un commento.