[Wien] wien and hyperthreading

[Stefaan Cottenier]

I tested the effect of hyperthreading one year ago, and noticed no 
improvement at first sight. However, Leo Errico (UNLP) pointed me to the 
at first sight unexpected and curious fact that hyperthreading does have 
a significant (5-10%) improvement if you run your case with 2 subjobs 
per cpu. What I mean is this: if you have 3 cpu's and 12 k-points, use 
the following .machines:

2:machine1
2:machine1
2:machine2
2:machine2
2:machine3
2:machine3

This goes several % faster than using

4:machine1
4:machine2
4:machine3

This is a very cheap way to get some extra speed out of your cpu's (but, 
Michael, it will work only for the lapw1/lapw2 part, not for your 
lapw0). This effect is consistent with Jeff's explanation of the 
mechanism behing hyperthreading (as soon as the 2 processes that run on 
the same cpu do not call exactly the same cpu functionalities at exactly 
the same time, which apparantly they don't).

Stefaan



Michael Gurnett wrote:
> Was just getting desperate to squeeze more speed out of the system. As 
> no one seems to have a got linux mpi version working (or at least no one 
> has answered previous posts). The system I'm working on now takes only 
> 28 minutes in lapw1, but 40 minutes in lapw0 (this is why mpi would be 
> nice) and 1h20 in lapw2
> 
> So obviously any extra speed would have been nice
> 
> Mick
> 
> 
> ----- Original Message ----- From: "Jeff Spirko" <spirko at lehigh.edu>
> To: <wien at zeus.theochem.tuwien.ac.at>
> Sent: Monday, March 14, 2005 4:42 PM
> Subject: Re: [Wien] wien and hyperthreading
> 
> 
>> With hyperthreading, you don't actually get to do twice as many
>> calculations per second on a single processor.  It allows the
>> processor to quickly switch from one process or thread to another
>> when the first process is waiting for something (like memory, disk,
>> etc.)  If both processes are using the floating point unit (FPU)
>> intensively, as Wien2k does through the BLAS library, the net result
>> is that the two threads just take turns calculating.
>>
>> What HT might do is allow Wien2k and "regular" stuff like Netscape
>> and X Windows to run side-by-side a little more efficiently, because
>> they use different resources.  See how much playing a movie slows down
>> Wien2k with and without HT on.  There you should see some improvement.
>>
>> Regards,
>> -Jeff Spirko
>>
>> On Mon, Mar 14, 2005 at 03:31:08PM +0100, Michael Gurnett wrote:
>>
>>>    I  was running some tests to compare speeds with hyperthreading on 
>>> and
>>>    off  using  the latest kernels for redhat 7.3. However, I was not 
>>> able
>>>    to  see any difference in speed between the two cases. I was 
>>> wondering
>>>    if anyone else has seen this and possibly explain why this is.
>>>    Michael
>>
>>
>> -- 
>> Jeff Spirko   spirko at lehigh.edu   spirko at yahoo.com   WD3V   |=>
>>
>> The study of non-linear physics is like the study of non-elephant 
>> biology.
>>
>> All theoretical chemistry is really physics;
>> and all theoretical chemists know it. -- Richard P. Feynman
>>
>> _______________________________________________
>> 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
>