[Wien] laptop can work faster than a typical good cluster!

[Saeid Jalali]

Thank you for your interest to GPU parallel programming. Simply the GPU is a
processor of the graphic card of our personal computers. In the graphic
cards there can be more than 1000 cores! This extremely huge capability was
used for running games! Now, time is to use these cores of the graphic cards
for scientific purposes as well. You would see the figure in page 9 of the
following presentation:
http://www.cisl.ucar.edu/dir/CAS2K9/Presentations/govett.pdf

Or figure 2 in page 3 of the following file:
http://download.microsoft.com/download/A/6/C/A6C0223B-9460-4346-8DC0-B6BCFD9
269B4/MSFT_GPGPU_whitepaper_FINAL.PDF

Everybody can buy a good graphic card with more than hundred cores. The
graphic cards are too cheap. A scientific program can also use this
capability of the graphic cards, provided that it is changed to be able to
use the GPU instead of only CPU. In this case, it is not necessary to spend
a lot of money and time to build a cluster using several expensive nodes,
and then very soon try to exchange the nodes by the new ones. Indeed,
cluster building is an end loop procedure. Fortunately, time is going
quickly to be over for clusters and its rich builders. 


Sincerely yours,
S. Jalali
/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
Saeid Jalali Asadabadi,
Department of Physics, Faculty of Science,
University of Isfahan (UI), Hezar Gerib Avenue,
81744 Isfahan, Iran.
Phones:
Dep. of Phys.   :+98-0311-793 2435
Office          :+98-0311-793 4176
Fax No.         :+98-0311-793 2409
E-mail          :sjalali at phys.ui.ac.ir
                :sjalali at sci.ui.ac.ir
                :sjalali at mailaps.org
                :saeid.jalali.asadabadi at gmail.com
                :s_jalali_a at yahoo.com
Homepage        :http://sci.ui.ac.ir/~sjalali
www             :http://www.ui.ac.ir
/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/


>-----Original Message-----
>From: wien-bounces at zeus.theochem.tuwien.ac.at [mailto:wien-
>bounces at zeus.theochem.tuwien.ac.at] On Behalf Of Dr. Manish K. Kashyap
>Sent: Saturday, October 30, 2010 8:19 AM
>To: A Mailing list for WIEN2k users
>Subject: Re: [Wien] laptop can work faster than a typical good cluster!
>
>On 10/16/10, Ramkumar Thapa <r.k.thapa at gmail.com> wrote:
>> Dear Dr. Saeid,
>> I really appreciate your statements on using the laptops for ab initio
>> calculations. Please continue feeding such suggestions which is really
>> helpful for people like us.
>> Regards,
>> R.K.Thapa
>> Mizoram University
>> INDIA
>> =======================
>>
>> 2010/10/15 Saeid Jalali <sjalali at phys.ui.ac.ir>
>>
>>>  Dear All,
>>>
>>>
>>>
>>> Let us start with the following question.
>>>
>>> Although this question and its answer may be trivial for authors and
>>> experts, but it may be interesting for many of us.
>>>
>>>
>>>
>>> Is it possible to run the WIEN2k code for a heavy case, using a
>regular
>>> laptop faster than a typical cluster containing for example more than
>100
>>> nodes even if the speed of each nodes of the cluster is higher than
>the
>>> speed of the laptop?
>>>
>>>
>>>
>>> The answer surprisingly can be positive, if the authors enable the
>wien2k
>>> code to use the graphics processing units (GPU) technology.
>>>
>>>
>>>
>>> We know that the maximum speed of silicon-based or even new
>relativistic
>>> graphene-based processors asymptotically approaches to limited
>values, and
>>> researchers are now switching from CPU to  GPU technology.
>>>
>>>
>>>
>>> This will help many researchers with limited facility to perform
>their
>>> heavy ab initio works like the others scientists who access to fast
>super
>>> computers.
>>>
>>>
>>>
>>> For more information see:
>>>
>>> [1] Journal of Computational Physics 227 (2008) 5342.
>>>
>>> [2] Computer Physics Communications 181, (2010) 1549.
>>>
>>>
>>>
>>> DFT calculations:
>>>
>>> [3] Journal of Chemical Physics 131 (2009) 034103.
>>>
>>> [4] Journal of Chemical Theory and Computation 4 (2008) 1230.
>>>
>>>
>>>
>>> Interesting enough, one may also know that:
>>>
>>> CUBLAS is an implementation of BLAS on top of GPU.
>>>
>>> CUFFT  is an implementation of FFT on top of GPU.
>>>
>>>
>>>
>>> You would also have a look at the following link:
>>>
>>> http://pcquest.ciol.com/content/technology/2007/107031003.asp
>>>
>>>
>>>
>>> And do google search on GPU.
>>>
>>>
>>>
>>> Sincerely yours,
>>>
>>> S. Jalali
>>>
>>> /_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
>>>
>>> Saeid Jalali Asadabadi,
>>>
>>> Department of Physics, Faculty of Science,
>>>
>>> University of Isfahan (UI), Hezar Gerib Avenue,
>>>
>>> 81744 Isfahan, Iran.
>>>
>>> Phones:
>>>
>>> Dep. of Phys.   :+98-0311-793 2435
>>>
>>> Office                  :+98-0311-793 4176
>>>
>>> Fax No.                :+98-0311-793 2409
>>>
>>> E-mail                  :sjalali at phys.ui.ac.ir
>>>
>>>                                :sjalali at sci.ui.ac.ir
>>>
>>>                                :sjalali at mailaps.org
>>>
>>>                                 :saeid.jalali.asadabadi at gmail.com
>>>
>>>                                 :s_jalali_a at yahoo.com
>>>
>>> Homepage          :http://sci.ui.ac.ir/~sjalali
>>>
>>> www                      :http://www.ui.ac.ir
>>>
>>> /_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/_/
>>>
>>> _______________________________________________
>>> Wien mailing list
>>> Wien at zeus.theochem.tuwien.ac.at
>>> http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien
>>>
>>>
>>
>_______________________________________________
>Wien mailing list
>Wien at zeus.theochem.tuwien.ac.at
>http://zeus.theochem.tuwien.ac.at/mailman/listinfo/wien