pymol-users

RE: [PyMOL] multi cpu rendering

[EPF \(E. P. Friis\) <EP...@no...>]

Hi Neil
=20
Have you tried the command
=20
set max_threads, 4
=20
That should enforce 4 threads, even on single CPU machines. Then it's up
to the kernel scheduler to distribute them to the defferent CPUs
=20
=20
Cheers,
=20
Esben
=20
=20
=20

	-----Original Message-----
	From: pym...@li...
[mailto:pym...@li...] On Behalf Of Neil
Ranson
	Sent: 19. april 2005 13:34
	To: PyM...@li...
	Subject: [PyMOL] multi cpu rendering
=09
=09

	                       =20

	Dear All,

	I've been preparing figures for a manuscript on a Sun v40z with
4 x Opteron 850 cpu's and 8Gb of memory (and Suse9.0). When I start
pymol:

	=20

	> pymol.com -c script.pml

	=20

	pymol says that it detects 4cpu and enables multithreaded
rendering, but I never see it using more than one cpu.=20

	Especially for producing ray traced movies, I'd really like to
see the multiprocessing benefits.

	=20

	Does anyone have any suggestions?

	=20

	Cheers,

	=20

	Neil

	=20

	=20

[PyMOL] RE: multi cpu rendering

[Neil R. <n.a...@le...>]

Dear All, 

Thanks for public and private replies alike.
The "set max_threads, n" construct works a treat, giving a 3.6x speedup
going from 1 to 4 cpu's.

A quad opteron is now an extremely fast rendering machine indeed!
Best regards,
Neil.

Re: [PyMOL] multi cpu rendering

[Serge C. <ser...@us...>]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Hi there;

Indeed if you read the documentation of Python, you'll see that the=20
python byte-code interpreter is not able to run multiple thread at any=20=

given time... So know what I'm not sure is if the ray tracing code of=20
pymol is in python or another language (C, C++???). If the ray tracing=20=

code is in Python, you will never be able to see a single pymol process=20=

using running on more than a single processor (at a given time).

SergE.

Le 19 avr. 05, =E0 15:57, EPF (Esben Peter Friis) a =E9crit :

> Hi Neil
> =A0
> Have you tried the command
> =A0
> set max_threads, 4
> =A0
> That should enforce 4 threads, even on single CPU machines. Then it's=20=

> up to the kernel scheduler to distribute them to the defferent CPUs
> =A0
> =A0
> Cheers,
> =A0
> Esben
> =A0
> =A0
> =A0
>> -----Original Message-----
>> From: pym...@li...=20
>> [mailto:pym...@li...] On Behalf Of Neil=20
>> Ranson
>> Sent: 19. april 2005 13:34
>> To: PyM...@li...
>> Subject: [PyMOL] multi cpu rendering
>>
>> =A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0=A0
>> Dear All,
>> I=92ve been preparing figures for a manuscript on a Sun v40z with 4 x=20=

>> Opteron 850 cpu=92s and 8Gb of memory (and Suse9.0). When I start=20
>> pymol:
>> =A0
>> > pymol.com =96c script.pml
>> =A0
>> pymol says that it detects 4cpu and enables multithreaded rendering,=20=

>> but I never see it using more than one cpu.
>> Especially for producing ray traced movies, I=92d really like to see=20=

>> the multiprocessing benefits.
>> =A0
>> Does anyone have any suggestions?
>> =A0
>> Cheers,
>> =A0
>> Neil
>> =A0
>> =A0
*******************************************
Serge Cohen

GPG Key ID: 9CBB58FB
*******************************************

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.2.3 (Darwin)

iD8DBQFCZjod5EPeG5y7WPsRAhYdAJ93jPF22nll0YM4ahLiPR9x9CuotgCcC7F6
hc+9FVkGhzUsUjc+B+MG0LU=3D
=3DjSeg
-----END PGP SIGNATURE-----

Re: [PyMOL] multi cpu rendering

[Kristoffer W. <kri...@bi...>]

Dear Serge,

I suggest you simply make a test.

> Indeed if you read the documentation of Python, you'll see that the 
> python byte-code interpreter is not able to run multiple thread at any 
> given time

I don't think the PyMOL raytracer is written in Python. Furtermore, when
I raytrace in PyMOL on my dual CPU Athlon MP both CPU's are utilized
(top shows 93% under linux). So it does multithread.

Regards,

./ Kristoffer

RE: [PyMOL] multi cpu rendering

[Warren D. <wa...@de...>]

You're all right. =20

The Python interpreter can't multiprocess, but Python threads are true
operating system threads.  So all you need to do is call down from =
Python
into C or some other language, free up the global interpreter lock, and
presto, you get *real* multiprocessing without every having to write a
single line of platform-specific threading code.

Virtually every API call in PyMOL goes through this cycle, which means =
that
PyMOL rendering is indeed much faster on multi-cpu systems.  Also,
multithreaded Python programs (such as GUIs) can run concurrently with =
PyMOL
in the same process on multi-cpu systems.

The best speed-up I've seen so far was on a 12-cpu Altix with 12 =
threads,
where PyMOL rendering was about 9X faster.  Of course, for the money, =
you
should just by a Linux cluster and split your multi-frame rendering job =
over
many dual-CPU nodes. =20

Cheers,
Warren

--
Warren L. DeLano, Ph.D.                    =20
Principal Scientist

. DeLano Scientific LLC =20
. 400 Oyster Point Blvd., Suite 213          =20
. South San Francisco, CA 94080   =20
. Biz:(650)-872-0942  Tech:(650)-872-0834    =20
. Fax:(650)-872-0273  Cell:(650)-346-1154
. mailto:wa...@de...     =20
=20

> -----Original Message-----
> From: pym...@li...=20
> [mailto:pym...@li...] On Behalf Of=20
> Kristoffer Winther S=F8rensen
> Sent: Wednesday, April 20, 2005 4:49 AM
> To: Serge Cohen
> Cc: EPF (Esben Peter Friis); Neil Ranson;=20
> PyM...@li...
> Subject: Re: [PyMOL] multi cpu rendering
>=20
> Dear Serge,
>=20
> I suggest you simply make a test.
>=20
> > Indeed if you read the documentation of Python, you'll see that the=20
> > python byte-code interpreter is not able to run multiple=20
> thread at any=20
> > given time
>=20
> I don't think the PyMOL raytracer is written in Python.=20
> Furtermore, when I raytrace in PyMOL on my dual CPU Athlon MP=20
> both CPU's are utilized (top shows 93% under linux). So it=20
> does multithread.
>=20
> Regards,
>=20
> ./ Kristoffer
>=20
>=20
>=20
>=20
>=20
> -------------------------------------------------------
> This SF.Net email is sponsored by: New Crystal Reports XI.
> Version 11 adds new functionality designed to reduce time=20
> involved in creating, integrating, and deploying reporting=20
> solutions. Free runtime info, new features, or free trial,=20
> at: http://www.businessobjects.com/devxi/728
> _______________________________________________
> PyMOL-users mailing list
> PyM...@li...
> https://lists.sourceforge.net/lists/listinfo/pymol-users
>=20