From owner-chemistry@ccl.net Wed Jul 27 08:52:00 2011 From: "David A Mannock dmannock]![ualberta.ca" To: CCL Subject: CCL: Reg: Configuration of a group of computers for Speeding up calculations Message-Id: <-45155-110726182913-15991-zd5Ei9k+Niz3MufhMLxLzg,server.ccl.net> X-Original-From: David A Mannock Content-Type: multipart/alternative; boundary=20cf307f307e72bf9404a9007544 Date: Tue, 26 Jul 2011 16:29:02 -0600 MIME-Version: 1.0 Sent to CCL by: David A Mannock [dmannock(-)ualberta.ca] --20cf307f307e72bf9404a9007544 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable Rocky, Yes, you would be better off either to assemble a dual CPU server (Socket 1207, CPU 2400; C32, cpu 4100; G34, cpu 6100 8-core), or or get several identical small boards capable of running 64-bit OS linux (like Ubuntu 10.04) and link them together. With the former check that bios upgrades for the board will support 6, 8, 12, 16 core processors to give yo= u an upgrade path. Newer boards still require ECC memory, but can take reg or unreg 240 pin SDRAM. Try to pick memory from a well known supplier with the correct timing. Some manufacturers have recommended list of tested configs. Do not buy individual memory sticks, but sets of 2,3,4,6 by the same manufacturer of the same type. No mixing and matching of brands, etc. By doing this, you should be able to gradually fill out the mobo in time. For the cluster approach you can buy several cheap mobos with the same architecture, CPU and memory. Motherboards in the AM2+/AM3 now support 6 core processors, but use non-ECC, unreg DDR2 memory (cheap). Gigabit ethernet is often available even on cheap boards now being sold off at $50-70. I have no idea how to set up this system. It might require working out how to run and connect multiple parallel systems. I have not gone this route myself. The main node would need to have a slightly faster cpu and more memory than the nodes. There is one other option, a single new fast, upgradable mobo. Motherboards with Intel cpus do not have a great capacity for upgrading, but the new AM3+ boards will take 2-8 cores in one socket. I= t might be better to invest in a new board and new fast DDR3 memory and a cheaper cpu (4 or 6-core), while you wait for the new 8-core chips to arriv= e in the fall. With more on chip memory, on chip video, more cores and faster onboard memory, you may find that a single board will give you all the spee= d that you need. You may find that the majority of Windows compatible molecular modeling software runs on a single core only. The linux versions will work with multiple cores, but may have to be compiled to run in that way. Often both C&Fortran compilers are required. Some are free, others wor= k out of the box but cost $$ and a yearly licence. CPUs at Starmicro.com are well priced. Vuugo is good, as is Newegg. Also, expect to make some mistakes along the way. Try the free version of Macrium reflect for images of HDDs, set up regular back ups and for Gods-sake, make a start up CD to spare you the headache of rebuilding a drive. Lastly, if you have a very large molecule or a complex supramolecular architecture to study. Learn the basics of the software on a small destop machine with small molecules first, then get time on a bigger server cluste= r at your university's computing centre. Many centres can install the softwar= e that you need and will help you fix problems with your scripts. All you nee= d to do is SSH in from your lab or home computer using VNC software. I hope that helps. Dave On Tue, Jul 26, 2011 at 12:44 PM, Jussi Lehtola jussi.lehtola*o*helsinki.fi wrote: > > Sent to CCL by: Jussi Lehtola [jussi.lehtola * helsinki.fi] > On Tue, 26 Jul 2011 23:19:46 +0530 > "rocky walden rocky.walden19-.-gmail.com" > wrote: > > Dear CCL folks, > > > > I have 3 Laptop machines which operate 24X7 , so i am planning to > > group them together and try to make a small group so that i can use > > them for exhaustive calculations. > > > > My lappy configuration. > > Intel Dualcore 1.86 GHz,160GB HDD,1GB RAM > > Intel Corei7, (quad core) 1.60Ghz, 6GB RAM, 650 GBHDD, 1 GB Graphics > > ATI Mobil radon > > Intel Corei5 4GB RAM 350GB,HDD, 512 ATI Radon. > > > > All these lappys have windows 7 Ultimate OS > > > > I have license versions of various computational chemistry > > (DESMOND,GROMACS, ABINTIO, )suites, can i set up a home cluster ? > > Can any one give me some ideas > > You are running Windows 7, which is AFAIK unusable for cluster > operation. Basically you could use some of the ready-made cluster linux > distros that run from a CD, but there's really no sense in trying to set > up a cluster with this kind of heterogeneous hardware, especially with a > very slow interconnect (probably 100 MB ethernet). > > You'll be any way restricted to running trivially parallel > calculations. The best you can do is just run separate calculations on > each computer (and maybe install some OS that is more capable for > scientific computing)... > -- > -------------------------------------------------------- > Mr. Jussi Lehtola, M. Sc. Doctoral Student > jussi.lehtola_._helsinki.fi Department of Physics > http://www.helsinki.fi/~jzlehtol University of Helsinki > Office phone: +358 9 191 50 632 Finland > -------------------------------------------------------- > Jussi Lehtola, FM Tohtorikoulutettava > jussi.lehtola_._helsinki.fi Fysiikan laitos > http://www.helsinki.fi/~jzlehtol Helsingin Yliopisto > Ty=F6puhelin: (0)9 191 50 632 > -------------------------------------------------------- > > > > -=3D This is automatically added to each message by the mailing script = =3D-> > > --20cf307f307e72bf9404a9007544 Content-Type: text/html; charset=ISO-8859-1 Content-Transfer-Encoding: quoted-printable Rocky, Yes, you would be better off either to assemble a dual CPU server (S= ocket 1207, CPU 2400; C32, cpu 4100; G34, cpu 6100 8-core), or or get sever= al identical small boards capable of running 64-bit OS linux (like Ubuntu 1= 0.04) and link them together. With the former check that bios upgrades for = the board will support 6, 8, 12, 16 core processors to give you an upgrade = path. Newer boards still require ECC memory, but can take reg or unreg 240 = pin SDRAM. Try to pick memory from a well known supplier with the correct t= iming. Some manufacturers have recommended list of tested configs. Do not b= uy individual memory sticks, but sets of 2,3,4,6 by the same manufacturer o= f the same type. No mixing and matching of brands, etc. By doing this, you = should be able to gradually fill out the mobo in time.

For the cluster approach you can buy several cheap mobos with the same = architecture, CPU and memory. Motherboards in the AM2+/AM3 now support 6 co= re processors, but use non-ECC, unreg DDR2 memory (cheap). Gigabit ethernet= is often available even on cheap boards now being sold off at $50-70. I ha= ve no idea how to set up this system. It might require working out how to r= un and connect multiple parallel systems. I have not gone this route myself= . The main node would need to have a slightly faster cpu and more memory th= an the nodes. There is one other option, a single new fast, upgradable mobo= . Motherboards with Intel cpus do not have a great capacity for upgrading, = but the new AM3+ boards will take 2-8 cores in one socket. It might be bett= er to invest in a new board and new fast DDR3 memory and a cheaper cpu (4 o= r 6-core), while you wait for the new 8-core chips to arrive in the fall. W= ith more on chip memory, on chip video, more cores and faster onboard memor= y, you may find that a single board will give you all the speed that you ne= ed. You may find that the majority of Windows compatible molecular modeling= software runs on a single core only. The linux versions will work with mul= tiple cores, but may have to be compiled to run in that way. Often both C&a= mp;Fortran compilers are required. Some are free, others work out of the bo= x but cost $$ and a yearly licence.

CPUs at Starmicro.com are well priced. Vuugo is good, as is Newegg. Als= o, expect to make some mistakes along the way.

Try the free version = of Macrium reflect for images of HDDs, set up regular back ups and for Gods= -sake, make a start up CD to spare you the headache of rebuilding a drive.<= br>
Lastly, if you have a very large molecule or a complex supramolecular a= rchitecture to study. Learn the basics of the software on a small destop ma= chine with small molecules first, then get time on a bigger server cluster = at your university's computing centre. Many centres can install the sof= tware that you need and will help you fix problems with your scripts. All y= ou need to do is SSH in from your lab or home computer using VNC software. = I hope that helps.
Dave

On Tue, Jul 26, 2011 at 12:44 PM, Ju= ssi Lehtola jussi.lehtola*o*helsinki.fi = <owner-chem= istry%ccl.net> wrote:

Sent to CCL by: Jussi Lehtola [jussi.lehtola * helsinki.fi]
On Tue, 26 Jul 2011 23:19:46 +0530
"rocky walden rocky.walden19-.-gmail.com" <owner-chemistry_._ccl.net>
wrote:
> Dear CCL folks,
>
> =A0 =A0 I have 3 Laptop machines which operate 24X7 , so i am planning= to
> group them together and try to make a small group so that i can use > them for exhaustive calculations.
>
> My lappy configuration.
> Intel Dualcore 1.86 GHz,160GB HDD,1GB RAM
> Intel Corei7, (quad core) 1.60Ghz, 6GB RAM, 650 GBHDD, 1 GB =A0Graphic= s
> ATI Mobil radon
> Intel Corei5 =A04GB RAM 350GB,HDD, 512 ATI Radon.
>
> All these lappys have windows 7 Ultimate OS
>
> I have license versions of various computational chemistry
> (DESMOND,GROMACS, ABINTIO, )suites, can i set up a home cluster ?
> Can any one give me some ideas

You are running Windows 7, which is AFAIK unusable for cluster
operation. Basically you could use some of the ready-made cluster linux
distros that run from a CD, but there's really no sense in trying to se= t
up a cluster with this kind of heterogeneous hardware, especially with a very slow interconnect (probably 100 MB ethernet).

You'll be any way restricted to running trivially parallel
calculations. The best you can do is just run separate calculations on
each computer (and maybe install some OS that is more capable for
scientific computing)...
--
--------------------------------------------------------
Mr. Jussi Lehtola, M. Sc. =A0 =A0 =A0 =A0 Doctoral Student
jussi.lehtola_._helsinki.f= i =A0 =A0 =A0 =A0 Department of Physics
http://www= .helsinki.fi/~jzlehtol =A0University of Helsinki
Office phone: +358 9 191 50 632 =A0 Finland
--------------------------------------------------------
Jussi Lehtola, FM =A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 Tohtorikoulutettava
jussi.lehtola_._helsinki.f= i =A0 =A0 =A0 =A0 Fysiikan laitos
http://www= .helsinki.fi/~jzlehtol =A0Helsingin Yliopisto
Ty=F6puhelin: (0)9 191 50 632
--------------------------------------------------------



-=3D This is automatically added to each message by the mailing script =3D-=
E-mail to subscribers: CHEMISTRY%ccl.n= et or use:
=A0 =A0 =A0http://www.ccl.net/cgi-bin/ccl/send_ccl_message

E-mail to administrators: CHEM= ISTRY-REQUEST%ccl.net or use
=A0 =A0 =A0http://www.ccl.net/cgi-bin/ccl/send_ccl_message

Subscribe/Unsubscribe:
=A0 =A0 =A0http://www.ccl.net/chemistry/sub_unsub.shtml

Before posting, check wait time at: http://www.ccl.net

Job: http://www.ccl.n= et/jobs
Conferences: http://server.ccl.net/chemistry/announcements/co= nferences/

Search Messages: http://www.ccl.net/chemistry/searchccl/index.shtml
=A0 =A0 =A0h= ttp://www.ccl.net/spammers.txt

RTFI: http://www.ccl.net/chemistry/aboutccl/instructions/



--20cf307f307e72bf9404a9007544-- From owner-chemistry@ccl.net Wed Jul 27 10:28:00 2011 From: "Joaquin Calbo himphoenixCCL]_[gmail.com" To: CCL Subject: CCL: ONIOM problem Message-Id: <-45156-110727102406-1460-hfL8UCFGZFgUqMphes/hXA|a|server.ccl.net> X-Original-From: "Joaquin Calbo" Date: Wed, 27 Jul 2011 10:24:02 -0400 Sent to CCL by: "Joaquin Calbo" [himphoenixCCL_._gmail.com] I am new in ONIOM calculations and I do not know very well how it works. I tried with an example and I cannot fix a problem that it says: -------------- Recover reactant geometry. Energy = -0.31453937D+04 Leave Link 107 at Wed Jul 27 16:03:01 2011, MaxMem= 1476395008 cpu: 1.1 (Enter /soft/g09.a02/g09/l101.exe) ONIOM: restoring gridpoint 3 on chk file. Symbolic Z-matrix: Charge = 0 Multiplicity = 1 for low level calculation on real system. Charge = 0 Multiplicity = 1 for high level calculation on model system. Charge = 0 Multiplicity = 1 for low level calculation on model system. Atoms in 0 layers were given but there should be 2. Error termination via Lnk1e in /soft/g09.a02/g09/l101.exe at Wed Jul 27 16:03:01 2011. --------------- My input file is as follows: --------------- #p opt oniom(MP2/6-31G:UFF) geom=connectivity opt=readfreeze title 0 1 O 0 7.370852 -0.386699 1.064329 L H 161 O 0 8.429162 1.586417 -0.393442 L (...) 1 3 1.0 15 1.0 88 1.0 2 3 1.0 14 1.0 31 1.0 (...) atoms=O Ti notatoms=161 103 162 ---------------- Any suggestions? Thanks in advance From owner-chemistry@ccl.net Wed Jul 27 11:02:00 2011 From: "Tymofii N tim_mail ~~ ukr.net" To: CCL Subject: CCL:G: MP2 binding energy Message-Id: <-45157-110727105810-22175-VGIoWEyQAZjpBy9opOqbJw ~ server.ccl.net> X-Original-From: "Tymofii N" Date: Wed, 27 Jul 2011 10:58:07 -0400 Sent to CCL by: "Tymofii N" [tim_mail%%ukr.net] Dear Adel! You may try to use a freeware tool EGView from http://timn.ho.ua/ccs But you have to remember that it uses fragments energy at their geometries as they are in complex, so the deformation energy is not included. Best, Tymofii Nikolaienko >Sent to CCL by: "Adel El-Azhary" [azhary:_:ksu.edu.sa] >Dear All: >I am confused as to how calculate the binding energy at the MP2 level from the Gaussian output file. Where I can find the MP2 counterpoise corrected energy of the complex and the two fragments constituting the complex in the Gaussian output file. >Thanking you for your cooperation >Best regards, >Adel El-Azhary From owner-chemistry@ccl.net Wed Jul 27 13:13:00 2011 From: "Jerome Kieffer Jerome.Kieffer/a\terre-adelie.org" To: CCL Subject: CCL: Reg: Configuration of a group of computers for Speeding up calculations Message-Id: <-45158-110727104125-7848-zaV6vzaiPKgXAPzcPTRMGg%server.ccl.net> X-Original-From: Jerome Kieffer Content-Transfer-Encoding: 8bit Content-Type: text/plain; charset=UTF-8 Date: Wed, 27 Jul 2011 16:41:04 +0200 Mime-Version: 1.0 Sent to CCL by: Jerome Kieffer [Jerome.Kieffer#%#terre-adelie.org] Hi all, I would like to feed back some experience I had 3 years ago from such "idea" which I implemented and run for a year: desktop computer are not designed to do calculation (and it is probably even worth with laptop computer): Computation chemistry software make heavy use of CPU (and RAM resources). Unlike most desktop program, they run for ages (hours to days or even month) at full power causing a lot of heat to be generated and subsequently exhausted by the cooling system. Surprisingly most of the failure I noticed were from premature death of hard-drives because the whole case became too hot (whereas the CPUs, generating this heat were fine). Later on I bought a small cluster (20 dual CPU, 4core each) that surprised the admin of the "datacenter" as they were consuming 8x more than "standard servers" (they were kicked out of the UPS). This was not due to the hardware but to the type of processing done on the type of programs running: Computation chemistry software are not like web-servers. Conclusion: Buy dedicated hardware !!! -- Jerome Kieffer . PS: I tested the intel "sandy-bridge" (latest i7) and it rocks really for not that expensive: 500€ mobo+proc, RAM is cheap at the moment, take a lot!!!. The server version which is arriving is likely to be great for computational chemistry. From owner-chemistry@ccl.net Wed Jul 27 15:57:00 2011 From: "bonoit dahmani bonoit_10~!~yahoo.fr" To: CCL Subject: CCL: Re : CCL: ONIOM problem Message-Id: <-45159-110727155349-28057-7/2Bp5XoFy4in1SXN2ckIg(~)server.ccl.net> X-Original-From: bonoit dahmani Content-Type: multipart/alternative; boundary="0-637484418-1311796418=:49368" Date: Wed, 27 Jul 2011 20:53:38 +0100 (BST) MIME-Version: 1.0 Sent to CCL by: bonoit dahmani [bonoit_10__yahoo.fr] --0-637484418-1311796418=:49368 Content-Type: text/plain; charset=iso-8859-1 Content-Transfer-Encoding: quoted-printable Dear Joaquin, I think you have just missed an to drop hydrogen atom, try to do that besau= se the problem you have faced is about the multiplicity.=20 To do an ONIOM calculation, you have to get the same multiplicity for all t= he selected layers , in other word: 1 1 1 or 2 2 2 ect. I hope that i've helped you a bit Best wishes Bonoit=20 --- En date de=A0: Mer 27.7.11, Joaquin Calbo himphoenixCCL]_[gmail.com a =E9crit=A0: De: Joaquin Calbo himphoenixCCL]_[gmail.com Objet: CCL: ONIOM problem =C0: "Bonoir, Bonoir " Date: Mercredi 27 juillet 2011, 16h24 Sent to CCL by: "Joaquin=A0 Calbo" [himphoenixCCL_._gmail.com] I am new in ONIOM calculations and I do not know very well how it works. I = tried with an example and I cannot fix a problem that it says: -------------- Recover reactant geometry.=A0 Energy =3D -0.31453937D+04 Leave Link=A0 107 at Wed Jul 27 16:03:01 2011, MaxMem=3D 1476395008 cpu:= =A0 =A0 =A0=A0=A01.1 (Enter /soft/g09.a02/g09/l101.exe) ONIOM: restoring gridpoint=A0 3 on chk file. Symbolic Z-matrix: Charge =3D=A0 0 Multiplicity =3D 1 for low=A0=A0=A0level calculation on re= al=A0 system. Charge =3D=A0 0 Multiplicity =3D 1 for high=A0 level calculation on model = system. Charge =3D=A0 0 Multiplicity =3D 1 for low=A0=A0=A0level calculation on mo= del system. Atoms in 0 layers were given but there should be 2. Error termination via Lnk1e in /soft/g09.a02/g09/l101.exe at Wed Jul 27 16= :03:01 2011. --------------- My input file is as follows: --------------- #p opt oniom(MP2/6-31G:UFF) geom=3Dconnectivity opt=3Dreadfreeze title 0 1=20 O=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 0=A0 =A0 7.370852=A0=A0=A0-0.386699=A0 = =A0 1.064329 L H=A0 =A0 161 O=A0 =A0 =A0 =A0 =A0 =A0 =A0 =A0 0=A0 =A0 8.429162=A0 =A0 1.586417=A0=A0= =A0-0.393442 L (...) 1 3 1.0 15 1.0 88 1.0 2 3 1.0 14 1.0 31 1.0 (...) atoms=3DO Ti notatoms=3D161 103 162 ---------------- Any suggestions? Thanks in advance -=3D This is automatically added to each message by the mailing script =3D-=A0 =A0 =A0=A0 =A0 =A0Subscribe/Unsubscribe:=20 =A0 =A0 =A0Job: http://www.ccl.net/jobs=20=A0 =A0 =A0--0-637484418-1311796418=:49368 Content-Type: text/html; charset=iso-8859-1 Content-Transfer-Encoding: quoted-printable
Dear Joaquin,
I think you have just missed= an to drop hydrogen atom, try to do that besause the problem you have face= d is about the multiplicity.
To do an ONIOM calculation, you have to ge= t the same multiplicity for all the selected layers , in other word: 1 1 1 = or 2 2 2 ect.
I hope that i've helped you a bit
Best wishes
Bonoit=

--- En date de : Mer 27.7.11, Joaquin Calbo himphoenixCCL]= _[gmail.com <owner-chemistry(_)ccl.net> a =E9crit :
=

De: Joaquin Calbo himphoenixCCL]_[gmail.com &l= t;owner-chemistry(_)ccl.net>
Objet: CCL: ONIOM problem
=C0: "Bonoir,= Bonoir " <bonoit_10(_)yahoo.fr>
Date: Mercredi 27 juillet 20= 11, 16h24


Sent to CCL by: "Joaquin = ; Calbo" [himphoenixCCL_._gmail.com]
I am new in ONIOM calculations and I do not= know very well how it works. I tried with an example and I cannot fix a pr= oblem that it says:

--------------
Recover reactant geometry.&nb= sp; Energy =3D -0.31453937D+04
Leave Link  107 at Wed Jul 27 16:03= :01 2011, MaxMem=3D 1476395008 cpu:       1.1
= (Enter /soft/g09.a02/g09/l101.exe)
ONIOM: restoring gridpoint  3 o= n chk file.


Symbolic Z-matrix:
Charge =3D  0 Multiplic= ity =3D 1 for low   level calculation on real  system.<= br> Charge =3D  0 Multiplicity =3D 1 for high  level calculation = on model system.
Charge =3D  0 Multiplicity =3D 1 for low &nb= sp; level calculation on model system.

Atoms in 0 layers were = given but there should be 2.
Error termination via Lnk1e in /soft/g09.a= 02/g09/l101.exe at Wed Jul 27 16:03:01 2011.
---------------

My i= nput file is as follows:

---------------
#p opt oniom(MP2/6-31G:UFF) geom= =3Dconnectivity opt=3Dreadfreeze

title

0 1
O   = ;             0    7.370852 &n= bsp; -0.386699    1.064329 L H    161
O  =               0    8.429162&nb= sp;   1.586417   -0.393442 L
(...)

1 3 1.0 15= 1.0 88 1.0
2 3 1.0 14 1.0 31 1.0
(...)

atoms=3DO Ti notatoms= =3D161 103 162

----------------

Any suggestions?

Thank= s in advance



-=3D This is automatically added to each messag= e by the mailing script =3D-
To recover the email address of the author = of the message, please change
the strange characters on the top line to = the (_) sign. You can also
look up the X-Original-From: line in the mail h= eader.

E-mail to subscribers: CHEMISTRY(_)ccl.net or use:<= br>      http://www.ccl.net/cgi-bin/ccl/send_ccl_message<= /a>

E-mail to administrators: CHEMISTRY-REQ= UEST(_)ccl.net or use
      http://www.ccl.net/cgi-b= in/ccl/send_ccl_message
   =   http://www.ccl.net/chemistry/sub_unsub.shtml

Before post= ing, check wait time at: h= ttp://www.ccl.net

Job: http://www.ccl.net/jobs
Conferences: http://server.ccl.net/chemistry/announcements/conferences/<= br>
Search Messages: http://www.ccl.net/chemistry/searchccl/index.= shtml
=       http://www.ccl.net/spammers.txt

RTFI: http://www.= ccl.net/chemistry/aboutccl/instructions/


=
--0-637484418-1311796418=:49368-- From owner-chemistry@ccl.net Wed Jul 27 21:40:00 2011 From: "David A Mannock dmannock^ualberta.ca" To: CCL Subject: CCL: Reg: Configuration of a group of computers for Speeding up calculations Message-Id: <-45160-110727185220-16935-I6aHNZ6lIH93sGcOo4F+Dw**server.ccl.net> X-Original-From: David A Mannock Content-Type: multipart/alternative; boundary=bcaec5014df71199ab04a914e621 Date: Wed, 27 Jul 2011 16:52:11 -0600 MIME-Version: 1.0 Sent to CCL by: David A Mannock [dmannock**ualberta.ca] --bcaec5014df71199ab04a914e621 Content-Type: text/plain; charset=windows-1252 Content-Transfer-Encoding: quoted-printable There you have the answer to your problem. The choices are several older dual cpu mobos with 64 bit capability in a cluster ($2-300 each fully equipped) or a dedicated server capable of running 24/7. I do not know what you can afford, but for example a Supermicro 4042G ($1800) + 64GB Kingston value DRAM (8x8GB ** $1000) and 4x 6128 AMD 8-core CPU ($4x$221) would come in at $4000. Add $150 for 4x 2U Dynatron fans. If you go for a dual processor board the costs would be roughly halved. If you buy components from several sources to minimize your costs, you will need someone to assemble this for you and to set up the OS and the software= . Most computer workshops and IT personnel can do this for you. You could buy everything assembled by the store, but get price matching for memory and CPUs. On arrival, have someone check that everything is properly in place before you fire it up. This is a high-end 32 core server system. It will give you fast service for years, plus when the Bulldozer chips are released and come down in price (there will be up to 20 cores/AMD cpu within 2 years), you can upgrade to 64 cores. This spreads the investment over time and if your software can deal with 32/64 cores you will have a good low maintenance system. Buy a good UPS with adequate capacity for the system! An OS like 64-bit Ubuntu 10.04 is free. QM software like Siesta and various interfaces, like Facio or Gabedit, are also free. Your university computer centre might be willing to install precompiled versions of G09/GV5, etc. fo= r you by SSH. Bear in mind that it requires expertise to run the software and take care of the units basic mechanical and OS needs. You may also have to learn some command line operations. The latter is something that I hate, hence my distaste of RHEL 5.3 and my preference for the GUI in Ubuntu, whic= h gives you both options. Miscellaneous ramblings. As has been mentioned in the previous post the older boards and cpus generate more heat and draw more power. I have both old and newer 1207 boards and one is nice and toasty when working hard, so if you need the extra heat on a cold night! The Shanghai CPU is now old technology, which o= n the face of it may now seem cheaper, but the ECC reg memory costs a lot mor= e and is often an older, slower DDR2 design. Also, upgrading some of these boards to be 4 or 6-core capable, requires a new bios chip. In the end the savings made by buying an older server technology are not worthwhile, as upgrading is not cost effective. It is better to start with a more recent technology to ensure that you have an upgrade path. See who is available to help you with this construction project and with software support. Bear in mind that a non-university computer will not be eligible for either software installation or tech support and this may mean that you are on your own. What you want to accomplish is doable and can be done at a surprisingly low cost relative to the multiblade servers of 5 years ago. I would encourage you to go for the G34 boards or the Intel equivalent. The latter would be 20-30% more expensive. My present server has onboard video. It's not great, but it does the job. I SSH into the lab via 127.0.0.1 using Real VNC viewer by tunneling and can see the server desktop (must be logged in on the server). In the lab, I hav= e a USB KVM switch, so my desktop and server are connected to the same keyboard, mouse and screen. You will need Putty to connect by SSH and WinSC= P to transfer files between computers across the network or internet. I do most of my viewing using a medium quality ATI card on the desktop machine. For most small molecule work, that is all you need. 2GB DDR5 gaming video cards are not needed, unless you want 3D movies of big molecules. Setting up scripts to run calculations can be tricky. Cheat by looking at the examples in the tutorials and any online tutorials. Sign up for the relevant forums. I use GV5 and Gabedit on WinXP/Vista64 *and* Ubuntu to calculate in G09, MOPAC, Siesta, etc. I like to write on my desktop and the= n transfer the script file over to the server, changing the path for the calculation in the same viewer. It helps me double check everything before = I start. There are many excellent online manuals which teach you how to write these scripts and the meaning of the various commands. I recommend that you bookmark those sites on each computer for easy access and add them to a custom menu bar. In particular, you will need to learn how to restart some calculations if they have been interrupted. Some calculations can be restarted, others cannot. So, there you have it. For a novice like myself 2 years ago, this was a daunting task and I made mistakes with hardware and software. Now I have an operating system that I love and a server that does a good job using free and proprietary software. However, as I said before, if you want to do calculations on complex systems, you may be better off to do calcs with small basis sets locally and to save each local minimum conformation generated and then use a bigger basis set for a calculation on a large server cluster at your universities computer centre. These are the decision= s that you have to make. I hope that adds to the messages on this topic. Dave On Wed, Jul 27, 2011 at 8:41 AM, Jerome Kieffer Jerome.Kieffer/ aterre-adelie.org wrote: > > Sent to CCL by: Jerome Kieffer [Jerome.Kieffer#%#terre-adelie.org] > Hi all, > > I would like to feed back some experience I had 3 years ago from such > "idea" which I implemented and run for a year: desktop computer are not > designed to do calculation (and it is probably even worth with laptop > computer): > > Computation chemistry software make heavy use of CPU (and RAM resources). > Unlike most desktop program, they run for ages (hours to days or even mon= th) > at full power causing a lot of heat to be generated and subsequently > exhausted by the cooling system. Surprisingly most of the failure I notic= ed > were from premature death of hard-drives because the whole case became to= o > hot (whereas the CPUs, generating this heat were fine). > > Later on I bought a small cluster (20 dual CPU, 4core each) that surprise= d > the admin of the "datacenter" as they were consuming 8x more than "standa= rd > servers" (they were kicked out of the UPS). This was not due to the hardw= are > but to the type of processing done on the type of programs running: > Computation chemistry software are not like web-servers. > > Conclusion: Buy dedicated hardware !!! > -- > Jerome Kieffer . > > PS: > I tested the intel "sandy-bridge" (latest i7) and it rocks really for not > that expensive: 500=80 mobo+proc, RAM is cheap at the moment, take a lot!= !!. > The server version which is arriving is likely to be great for > computational chemistry. > > > > -=3D This is automatically added to each message by the mailing script = =3D-> > > --bcaec5014df71199ab04a914e621 Content-Type: text/html; charset=windows-1252 Content-Transfer-Encoding: quoted-printable There you have the answer to your problem. The choices are several older du= al cpu mobos with 64 bit capability in a cluster ($2-300 each fully equippe= d) or a dedicated server capable of running 24/7. I do not know what you ca= n afford, but for example a Supermicro 4042G ($1800) + 64GB Kingston value = DRAM (8x8GB ** $1000) and 4x 6128 AMD 8-core CPU ($4x$221) would come in at = $4000. Add $150 for 4x 2U Dynatron fans. If you go for a dual processor boa= rd the costs would be roughly halved.

If you buy components from several sources to minimize your costs, you= =20 will need someone to assemble this for you and to set up the OS and the=20 software. Most computer workshops and IT personnel can do this for you.=20 You could buy everything assembled by the store, but get price matching=20 for memory and CPUs. On arrival, have someone check that everything is=20 properly in place before you fire it up. This is a high-end 32 core=20 server system. It will give you fast service for years, plus when the=20 Bulldozer chips are released and come down in price (there will be up to 20 cores/AMD cpu within 2 years), you can upgrade to 64 cores. This=20 spreads the investment over time and if your software can deal with=20 32/64 cores you will have a good low maintenance system. Buy a good UPS=20 with adequate capacity for the system!

An OS like 64-bit Ubuntu 10.04 is free. QM software like Siesta and various interfaces= , like Facio or=20 Gabedit, are also free. Your university computer centre might be willing to install precompiled versions of G09/GV5, etc. for you by SSH. Bear in m= ind that it requires=20 expertise to run the software and take care of the units basic=20 mechanical and OS needs. You may also have to learn some command line=20 operations. The latter is something that I hate, hence my distaste of=20 RHEL 5.3 and my preference for the GUI in Ubuntu, which gives you both=20 options.

Miscellaneous ramblings.
As has been mentioned in the pr= evious post the older boards and cpus generate more heat and draw more powe= r. I have both old and newer 1207 boards and one is nice and toasty when wo= rking hard, so if you need the extra heat on a cold night! The Shanghai CPU= is now old technology, which on the face of it may now seem cheaper, but t= he ECC reg memory costs a lot more and is often an older, slower DDR2 desig= n. Also, upgrading some of these boards to be 4 or 6-core capable, requires= a new bios chip. In the end the savings made by buying an older server tec= hnology are not worthwhile, as upgrading is not cost effective. It is bette= r to start with a more recent technology to ensure that you have an upgrade= path.

See who is available to help you with this construction project and wit= h software support. Bear in mind that a non-university computer will not be= eligible for either software installation or tech support and this may mea= n that you are on your own.

What you want to accomplish is doable and can be done at a surprisingly= low cost relative to the multiblade servers of 5 years ago. I would encour= age you to go for the G34 boards or the Intel equivalent. The latter would = be 20-30% more expensive.

My present server has onboard video. It's not great, but it does th= e job. I SSH into the lab via 127.0.0.1 using Real VNC viewer by tunneling = and can see the server desktop (must be logged in on the server). In the la= b, I have a USB KVM switch, so my desktop and server are connected to the s= ame keyboard, mouse and screen. You will need Putty to connect by SSH and W= inSCP to transfer files between computers across the network or internet. I= do most of my viewing using a medium quality ATI card on the desktop machi= ne. For most small molecule work, that is all you need. 2GB DDR5 gaming vid= eo cards are not needed, unless you want 3D movies of big molecules.

Setting up scripts to run calculations can be tricky. Cheat by looking = at the examples in the tutorials and any online tutorials. Sign up for the = relevant forums. I use GV5 and Gabedit on WinXP/Vista64 and Ubuntu t= o calculate in G09, MOPAC, Siesta, etc. I like to write on my desktop and t= hen transfer the script file over to the server, changing the path for the = calculation in the same viewer. It helps me double check everything before = I start. There are many excellent online manuals which teach you how to wri= te these scripts and the meaning of the various commands. I recommend that = you bookmark those sites on each computer for easy access and add them to a= custom menu bar. In particular, you will need to learn how to restart some= calculations if they have been interrupted. Some calculations can be resta= rted, others cannot.

So, there you have it. For a novice like myself 2 years ago, this was a= daunting task and I made mistakes with hardware and software. Now I have a= n operating system that I love and a server that does a good job using free= and proprietary software. However, as I said before, if you want to do cal= culations on complex systems, you may be better off to do calcs with small = basis sets locally and to save each local minimum conformation generated an= d then use a bigger basis set for a calculation on a large server cluster a= t your universities computer centre. These are the decisions that you have = to make.

I hope that adds to the messages on this topic. Dave

On Wed, Jul 27, 2011 at 8:41 AM, Jerome Kieffer Jerome.Kie= ffer/aterre-adelie.org <owner-chemistry**ccl= .net> wrote:

Sent to CCL by: Jerome Kieffer [Jerome.Kieffer#%#terre-adelie.org]
Hi all,

I would like to feed back some experience I had 3 years ago from such "= ;idea" which I implemented and run for a year: desktop computer are no= t designed to do calculation (and it is probably even worth with laptop com= puter):

Computation chemistry software make heavy use of CPU (and RAM resources). U= nlike most desktop program, they run for ages (hours to days or even month)= at full power causing a lot of heat to be generated and subsequently exhau= sted by the cooling system. Surprisingly most of the failure I noticed were= from premature death of hard-drives because the whole case became too hot = (whereas the CPUs, generating this heat were fine).

Later on I bought a small cluster (20 dual CPU, 4core each) that surprised = the admin of the "datacenter" as they were consuming 8x more than= "standard servers" (they were kicked out of the UPS). This was n= ot due to the hardware but to the type of processing done on the type of pr= ograms running: Computation chemistry software are not like web-servers.
Conclusion: Buy dedicated hardware !!!
--
Jerome Kieffer <Jerome.Kieffer a terre-adelie.org>.

PS:
I tested the intel "sandy-bridge" (latest i7) and it rocks really= for not that expensive: 500=80 mobo+proc, RAM is cheap at the moment, take= a lot!!!.
The server version which is arriving is likely to be great for computationa= l chemistry.



-=3D This is automatically added to each message by the mailing script =3D-=
E-mail to subscribers: CHEMISTRY**ccl.n= et or use:
=A0 =A0 =A0http://www.ccl.net/cgi-bin/ccl/send_ccl_message

E-mail to administrators: CHEM= ISTRY-REQUEST**ccl.net or use
=A0 =A0 =A0http://www.ccl.net/cgi-bin/ccl/send_ccl_message

Subscribe/Unsubscribe:
=A0 =A0 =A0http://www.ccl.net/chemistry/sub_unsub.shtml

Before posting, check wait time at: http://www.ccl.net

Job: http://www.ccl.n= et/jobs
Conferences: http://server.ccl.net/chemistry/announcements/co= nferences/

Search Messages: http://www.ccl.net/chemistry/searchccl/index.shtml
=A0 =A0 =A0h= ttp://www.ccl.net/spammers.txt

RTFI: http://www.ccl.net/chemistry/aboutccl/instructions/



--bcaec5014df71199ab04a914e621--