Re: [Freesurfer] CUDA Error in file '/space/freesurfer/build/minerva/stable/dev/include/ctfactory.hpp' on line 92 : invalid texture reference
Hemant,
We dont have any cards here that are compute capability 1.0. Our test cards are all 1.3 or higher.
We do compile with -arch sm_11 (compute capability 1.1). I've recompiled mri_em_register_cuda and mri_ca_register_cuda with -arch sm_10 and put them here:
https://surfer.nmr.mgh.harvard.edu/pub/dist/freesurfer/misc/linux-centos4_x8...
i dont know if they will work for you though.
n.
On Mon, 2011-10-31 at 12:29 -0400, R Edgar wrote:
On Mon, Oct 31, 2011 at 11:32 AM, Hemant Bokil hemantbokil@gmail.com wrote:
There was another thread a while back on the freesurfer website: A user reported a similar problem to mine and another user responded that the problem was with CUDA Compute Capability (See the link below- http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg14906.html)
My card is a Quadro 4600 which has Compute Capability 1.0. Do I need a card with a higher Compute Capability ?
I didn't catch that. You absolutely need a more capable card. You should check with Nick as to the minimum supported, but you don't get the full benefits until you get to Comp. Cap. 2.0.
Richard
Hi Nick,
Thanks for generating those files. The code ran through more steps this time (with the new files) but I now have the following error:
Cuda error: MRImeanX kernel failed in file 'mrimean_cuda.cu' in line 578 : invalid device function. register_mri: find_optimal_transform
Would you be able to recompile and put that on the server ? Alternately, if the source code is online and there are instructions for the recompile, I could do it myself. Many thanks,
Hemant
On Mon, Oct 31, 2011 at 1:40 PM, Nick Schmansky nicks@nmr.mgh.harvard.eduwrote:
Hemant,
We dont have any cards here that are compute capability 1.0. Our test cards are all 1.3 or higher.
We do compile with -arch sm_11 (compute capability 1.1). I've recompiled mri_em_register_cuda and mri_ca_register_cuda with -arch sm_10 and put them here:
https://surfer.nmr.mgh.harvard.edu/pub/dist/freesurfer/misc/linux-centos4_x8...
i dont know if they will work for you though.
n.
On Mon, 2011-10-31 at 12:29 -0400, R Edgar wrote:
On Mon, Oct 31, 2011 at 11:32 AM, Hemant Bokil hemantbokil@gmail.com
wrote:
There was another thread a while back on the freesurfer website: A user reported a similar problem to mine and another user responded that the problem was with CUDA Compute Capability (See the link below-
http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg14906.html)
My card is a Quadro 4600 which has Compute Capability 1.0. Do I need a
card
with a higher Compute Capability ?
I didn't catch that. You absolutely need a more capable card. You should check with Nick as to the minimum supported, but you don't get the full benefits until you get to Comp. Cap. 2.0.
Richard
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.
which binary was running when this error occurred? consult the recon-all.log to find out. if it was mri_em_register_cuda or mri_ca_register_cuda, then it would appear that it is attempting to use a function not supported by compute capability 1.0, which means your card is not supported.
n.
On Mon, 2011-10-31 at 14:01 -0400, Hemant Bokil wrote:
Hi Nick,
Thanks for generating those files. The code ran through more steps this time (with the new files) but I now have the following error:
Cuda error: MRImeanX kernel failed in file 'mrimean_cuda.cu' in line 578 : invalid device function. register_mri: find_optimal_transform
Would you be able to recompile and put that on the server ? Alternately, if the source code is online and there are instructions for the recompile, I could do it myself. Many thanks,
Hemant
On Mon, Oct 31, 2011 at 1:40 PM, Nick Schmansky nicks@nmr.mgh.harvard.edu wrote: Hemant,
We dont have any cards here that are compute capability 1.0. Our test cards are all 1.3 or higher. We do compile with -arch sm_11 (compute capability 1.1). I've recompiled mri_em_register_cuda and mri_ca_register_cuda with -arch sm_10 and put them here: https://surfer.nmr.mgh.harvard.edu/pub/dist/freesurfer/misc/linux-centos4_x86_64/ i dont know if they will work for you though. n. On Mon, 2011-10-31 at 12:29 -0400, R Edgar wrote: > On Mon, Oct 31, 2011 at 11:32 AM, Hemant Bokil <hemantbokil@gmail.com> wrote: > > > There was another thread a while back on the freesurfer website: A user > > reported a similar problem to mine and another user responded that the > > problem was with CUDA Compute Capability (See the link below- > > http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg14906.html) > > > > My card is a Quadro 4600 which has Compute Capability 1.0. Do I need a card > > with a higher Compute Capability ? > > I didn't catch that. You absolutely need a more capable card. You > should check with Nick as to the minimum supported, but you don't get > the full benefits until you get to Comp. Cap. 2.0. > > Richard > > 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.
It was indeed using mri_em_register_cuda. I will check with a different card. Thanks a lot, Hemant
On Mon, Oct 31, 2011 at 3:24 PM, Nick Schmansky nicks@nmr.mgh.harvard.eduwrote:
which binary was running when this error occurred? consult the recon-all.log to find out. if it was mri_em_register_cuda or mri_ca_register_cuda, then it would appear that it is attempting to use a function not supported by compute capability 1.0, which means your card is not supported.
n.
On Mon, 2011-10-31 at 14:01 -0400, Hemant Bokil wrote:
Hi Nick,
Thanks for generating those files. The code ran through more steps this time (with the new files) but I now have the following error:
Cuda error: MRImeanX kernel failed in file 'mrimean_cuda.cu' in line 578 : invalid device function. register_mri: find_optimal_transform
Would you be able to recompile and put that on the server ? Alternately, if the source code is online and there are instructions for the recompile, I could do it myself. Many thanks,
Hemant
On Mon, Oct 31, 2011 at 1:40 PM, Nick Schmansky nicks@nmr.mgh.harvard.edu wrote: Hemant,
We dont have any cards here that are compute capability 1.0. Our test cards are all 1.3 or higher. We do compile with -arch sm_11 (compute capability 1.1). I've recompiled mri_em_register_cuda and mri_ca_register_cuda with -arch sm_10 and put them here:https://surfer.nmr.mgh.harvard.edu/pub/dist/freesurfer/misc/linux-centos4_x8...
i dont know if they will work for you though. n. On Mon, 2011-10-31 at 12:29 -0400, R Edgar wrote: > On Mon, Oct 31, 2011 at 11:32 AM, Hemant Bokil <hemantbokil@gmail.com> wrote: > > > There was another thread a while back on the freesurfer website: A user > > reported a similar problem to mine and another user responded that the > > problem was with CUDA Compute Capability (See the link below- > >http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg14906.html)
> > > > My card is a Quadro 4600 which has Compute Capability 1.0. Do I need a card > > with a higher Compute Capability ? > > I didn't catch that. You absolutely need a more capable card. You > should check with Nick as to the minimum supported, but you don't get > the full benefits until you get to Comp. Cap. 2.0. > > Richard > > 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.
sorry it didnt work. i figured there was a reason we built with 1.1 and not 1.0, but forgot what it was (it appears our gpu code requires some functionality found in 1.1 and newer).
n.
On Mon, 2011-10-31 at 15:36 -0400, Hemant Bokil wrote:
It was indeed using mri_em_register_cuda. I will check with a different card. Thanks a lot, Hemant
On Mon, Oct 31, 2011 at 3:24 PM, Nick Schmansky nicks@nmr.mgh.harvard.edu wrote: which binary was running when this error occurred? consult the recon-all.log to find out. if it was mri_em_register_cuda or mri_ca_register_cuda, then it would appear that it is attempting to use a function not supported by compute capability 1.0, which means your card is not supported.
n. On Mon, 2011-10-31 at 14:01 -0400, Hemant Bokil wrote: > Hi Nick, > > Thanks for generating those files. The code ran through more steps > this time (with the new files) but I now have the following error: > > Cuda error: MRImeanX kernel failed in file 'mrimean_cuda.cu' in line > 578 : invalid device function. > register_mri: find_optimal_transform > > Would you be able to recompile and put that on the server ? > Alternately, if the source code is online and there are instructions > for the recompile, I could do it myself. Many thanks, > > Hemant > > On Mon, Oct 31, 2011 at 1:40 PM, Nick Schmansky > <nicks@nmr.mgh.harvard.edu> wrote: > Hemant, > > We dont have any cards here that are compute capability 1.0. > Our test > cards are all 1.3 or higher. > > We do compile with -arch sm_11 (compute capability 1.1). I've > recompiled mri_em_register_cuda and mri_ca_register_cuda with > -arch > sm_10 and put them here: > > https://surfer.nmr.mgh.harvard.edu/pub/dist/freesurfer/misc/linux-centos4_x86_64/ > > i dont know if they will work for you though. > > n. > > > > On Mon, 2011-10-31 at 12:29 -0400, R Edgar wrote: > > On Mon, Oct 31, 2011 at 11:32 AM, Hemant Bokil > <hemantbokil@gmail.com> wrote: > > > > > There was another thread a while back on the freesurfer > website: A user > > > reported a similar problem to mine and another user > responded that the > > > problem was with CUDA Compute Capability (See the link > below- > > > > http://www.mail-archive.com/freesurfer@nmr.mgh.harvard.edu/msg14906.html) > > > > > > My card is a Quadro 4600 which has Compute Capability 1.0. > Do I need a card > > > with a higher Compute Capability ? > > > > I didn't catch that. You absolutely need a more capable > card. You > > should check with Nick as to the minimum supported, but you > don't get > > the full benefits until you get to Comp. Cap. 2.0. > > > > Richard > > > > > > > > > > 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
-
Hemant Bokil -
Nick Schmansky