Hi
I am in the process of acquiring a new computer to run freesurfer on. I am currently think about buying a PC with a geforce 4X0 and a Tesla card along with a two screen setup running either Ubuntu or Redhat Linux enterprise 6.0. I think the PC should be a core i7 with 6 core and above 12 gigs byte of ram. What kind of tesla card is recommend for using with Freesurfer? Will Freesurfer 5.1 also support cuda 4.0.
Knut J
Maybe a NVidia GTX590 (it's a dual GPU)
On Mon, Apr 4, 2011 at 08:13, Knut J Bjuland knutjbj@hotmail.com wrote:
Hi
I am in the process of acquiring a new computer to run freesurfer on. I am currently think about buying a PC with a geforce 4X0 and a Tesla card along with a two screen setup running either Ubuntu or Redhat Linux enterprise 6.0. I think the PC should be a core i7 with 6 core and above 12 gigs byte of ram. What kind of tesla card is recommend for using with Freesurfer? Will Freesurfer 5.1 also support cuda 4.0.
Knut J
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.
On Mon, 2011-04-04 at 13:13 +0200, Knut J Bjuland wrote:
I am in the process of acquiring a new computer to run freesurfer on. I am currently think about buying a PC with a geforce 4X0 and a Tesla card along with a two screen setup running either Ubuntu or Redhat Linux enterprise 6.0. I think the PC should be a core i7 with 6 core and above 12 gigs byte of ram. What kind of tesla card is recommend for using with Freesurfer? Will Freesurfer 5.1 also support cuda 4.0.
If your budget can withstand a Tesla C2050, by all means go for it. I've not checked in detail how much GPU RAM FreeSurfer requires, so it might be possible to run on a 'lesser' GeForce card. There's not enough double precision stuff to be significant in the GeForce vs Tesla choice. And I don't do overlapping transfers, which would be another factor. The main thing to do is make sure you have a 'Fermi' GPU - that's a Tesla 20 series, or GeForce GTX 400 or 500 series. Large speed ups depend on the Fermi card.
As for the CPU... just make sure it's Nehalem class. FreeSurfer on both the CPU and GPU likes fast RAM access.
As for CUDA 4.0 support.... that will depend on NVIDIA not breaking backwards compatibility (not an absolute given - they are not IBM). I don't expect any trouble, but until we upgrade to CUDA 4.0 here (and 3.2 just broke binary compatibility), I wouldn't want to promise anything.
HTH,
Richard
Will a Tesla C2050 or another good CPU be able to reduce the running time from 20-24 hr to less time like for instance 8hr or below that time.
Knut J
From: rge21@nmr.mgh.harvard.edu To: freesurfer@nmr.mgh.harvard.edu Date: Mon, 4 Apr 2011 09:23:07 -0400 Subject: Re: [Freesurfer] recommende pc and cuda
On Mon, 2011-04-04 at 13:13 +0200, Knut J Bjuland wrote:
I am in the process of acquiring a new computer to run freesurfer on. I am currently think about buying a PC with a geforce 4X0 and a Tesla card along with a two screen setup running either Ubuntu or Redhat Linux enterprise 6.0. I think the PC should be a core i7 with 6 core and above 12 gigs byte of ram. What kind of tesla card is recommend for using with Freesurfer? Will Freesurfer 5.1 also support cuda 4.0.
If your budget can withstand a Tesla C2050, by all means go for it. I've not checked in detail how much GPU RAM FreeSurfer requires, so it might be possible to run on a 'lesser' GeForce card. There's not enough double precision stuff to be significant in the GeForce vs Tesla choice. And I don't do overlapping transfers, which would be another factor. The main thing to do is make sure you have a 'Fermi' GPU - that's a Tesla 20 series, or GeForce GTX 400 or 500 series. Large speed ups depend on the Fermi card.
As for the CPU... just make sure it's Nehalem class. FreeSurfer on both the CPU and GPU likes fast RAM access.
As for CUDA 4.0 support.... that will depend on NVIDIA not breaking backwards compatibility (not an absolute given - they are not IBM). I don't expect any trouble, but until we upgrade to CUDA 4.0 here (and 3.2 just broke binary compatibility), I wouldn't want to promise anything.
HTH,
Richard
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.
On Tue, 2011-04-05 at 11:23 +0200, Knut J Bjuland wrote:
Will a Tesla C2050 or another good CPU be able to reduce the running time from 20-24 hr to less time like for instance 8hr or below that time.
On the standard test case we use here, a full recon-all run takes 8 hours on a 3.2 GHz Nehalem core, and about 4 hours 20 mins when using the Tesla C2050.
Richard
Richard G. Edgar wrote:
On Tue, 2011-04-05 at 11:23 +0200, Knut J Bjuland wrote:
Will a Tesla C2050 or another good CPU be able to reduce the running time from 20-24 hr to less time like for instance 8hr or below that time.
On the standard test case we use here, a full recon-all run takes 8 hours on a 3.2 GHz Nehalem core, and about 4 hours 20 mins when using the Tesla C2050.
Would I be right in concluding that a 4-core Nehalem (e.g. i7) has more throughput than the C2050 then?
On Tue, 2011-04-05 at 16:30 +0100, Ian Malone wrote:
Richard G. Edgar wrote:
On Tue, 2011-04-05 at 11:23 +0200, Knut J Bjuland wrote:
Will a Tesla C2050 or another good CPU be able to reduce the running time from 20-24 hr to less time like for instance 8hr or below that time.
On the standard test case we use here, a full recon-all run takes 8 hours on a 3.2 GHz Nehalem core, and about 4 hours 20 mins when using the Tesla C2050.
Would I be right in concluding that a 4-core Nehalem (e.g. i7) has more throughput than the C2050 then?
Yes, but less than having 3 CPU jobs, and one GPU one. I did test once, and there isn't much penalty to running one recon-all job per core on a Nehalem system.
Right now, the CPU still does most of the work in the recon-all stream - it's something of a game of Amdahl's Law Wac-A-Mole. You could always try starting 4 GPU jobs at once.... I've not done the testing, but a C2050 would probably have enough RAM, and in any given recon-all run, the GPU does spend a lot of time idle. Hence, it would end up being divvied up between the four jobs.
Richard
On 05/04/11 16:55, Richard G. Edgar wrote:
On Tue, 2011-04-05 at 16:30 +0100, Ian Malone wrote:
Richard G. Edgar wrote:
On the standard test case we use here, a full recon-all run takes 8 hours on a 3.2 GHz Nehalem core, and about 4 hours 20 mins when using the Tesla C2050.
Would I be right in concluding that a 4-core Nehalem (e.g. i7) has more throughput than the C2050 then?
Yes, but less than having 3 CPU jobs, and one GPU one. I did test once, and there isn't much penalty to running one recon-all job per core on a Nehalem system.
Right now, the CPU still does most of the work in the recon-all stream - it's something of a game of Amdahl's Law Wac-A-Mole. You could always try starting 4 GPU jobs at once.... I've not done the testing, but a C2050 would probably have enough RAM, and in any given recon-all run, the GPU does spend a lot of time idle. Hence, it would end up being divvied up between the four jobs.
Thanks, that's interesting to know.
Hi Freesurfers,
What Operating System are you using?
I'm using CentOS 5.5 but with CUDA 3.2 and I have some problems with the nvidia.ko kernel module. Every time that I reboot the computer I have to make a modprobe and test with some CUDA SDK example... Other problems appear when I run multiple recon-all, often the processes crash with a kernel panic or simply with an abort.
I don't know if the OS is interfering in the execution...
Any idea will be apreciated...
Thank you!
2011/4/5 Ian Malone ian.malone@drc.ion.ucl.ac.uk
On 05/04/11 16:55, Richard G. Edgar wrote:
On Tue, 2011-04-05 at 16:30 +0100, Ian Malone wrote:
Richard G. Edgar wrote:
On the standard test case we use here, a full recon-all run takes 8 hours on a 3.2 GHz Nehalem core, and about 4 hours 20 mins when using the Tesla C2050.
Would I be right in concluding that a 4-core Nehalem (e.g. i7) has more throughput than the C2050 then?
Yes, but less than having 3 CPU jobs, and one GPU one. I did test once, and there isn't much penalty to running one recon-all job per core on a Nehalem system.
Right now, the CPU still does most of the work in the recon-all stream - it's something of a game of Amdahl's Law Wac-A-Mole. You could always try starting 4 GPU jobs at once.... I've not done the testing, but a C2050 would probably have enough RAM, and in any given recon-all run, the GPU does spend a lot of time idle. Hence, it would end up being divvied up between the four jobs.
Thanks, that's interesting to know.
-- imalone _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.
Hi
The system should load nvidia module automatically. You could try to add modprobe nvidia to rc.load. However there might be a problem in your system and you could try to use nvidia-bug-report.sh and send the out put to either nvnews.net nvidia linux forum or to this list.
Knut J
On 04/06/2011 10:02 AM, Jordi Delgado wrote:
Hi Freesurfers,
What Operating System are you using?
I'm using CentOS 5.5 but with CUDA 3.2 and I have some problems with the nvidia.ko kernel module. Every time that I reboot the computer I have to make a modprobe and test with some CUDA SDK example... Other problems appear when I run multiple recon-all, often the processes crash with a kernel panic or simply with an abort.
I don't know if the OS is interfering in the execution...
Any idea will be apreciated...
Thank you!
2011/4/5 Ian Malone <ian.malone@drc.ion.ucl.ac.uk mailto:ian.malone@drc.ion.ucl.ac.uk>
On 05/04/11 16:55, Richard G. Edgar wrote: > > On Tue, 2011-04-05 at 16:30 +0100, Ian Malone wrote: >> Richard G. Edgar wrote: >>> On the standard test case we use here, a full recon-all run takes 8 >>> hours on a 3.2 GHz Nehalem core, and about 4 hours 20 mins when using >>> the Tesla C2050. >>> >>> >> >> Would I be right in concluding that a 4-core Nehalem (e.g. i7) has >> more throughput than the C2050 then? > > Yes, but less than having 3 CPU jobs, and one GPU one. I did test once, > and there isn't much penalty to running one recon-all job per core on a > Nehalem system. > > Right now, the CPU still does most of the work in the recon-all stream - > it's something of a game of Amdahl's Law Wac-A-Mole. You could always > try starting 4 GPU jobs at once.... I've not done the testing, but a > C2050 would probably have enough RAM, and in any given recon-all run, > the GPU does spend a lot of time idle. Hence, it would end up being > divvied up between the four jobs. > Thanks, that's interesting to know. -- imalone _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu <mailto:Freesurfer@nmr.mgh.harvard.edu> https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.-- Jordi Delgado Mengual PIC (Port d'Informació Científica) Campus UAB, Edifici D E-08193 Bellaterra, Barcelona Tel: +34 93 586 82 32 Fax: +34 93 581 41 10 http://www.pic.es Avis - Aviso - Legal Notice: http://www.ifae.es/legal.html
Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edu https://mail.nmr.mgh.harvard.edu/mailman/listinfo/freesurfer
The information in this e-mail is intended only for the person to whom it is addressed. If you believe this e-mail was sent to you in error and the e-mail contains patient information, please contact the Partners Compliance HelpLine at http://www.partners.org/complianceline . If the e-mail was sent to you in error but does not contain patient information, please contact the sender and properly dispose of the e-mail.
freesurfer@nmr.mgh.harvard.edu
-
Ian Malone -
Jordi Delgado -
Knut J Bjuland -
Pedro Paulo de Magalhães Oliveira Junior -
Richard G. Edgar