Dear Experts,
I am using Intel® Core™ i7-4790 CPU @ 3.60GHz × 8 processor with 15.6 GiB ram, GeForce GT 610/PCIe/SSE2 graphic card.
I have successfully installed cuda runtime version 5.0. After that I performed cuda detect and output result is pasted below.
$ cudadetect Detecting CUDA... There is 1 device supporting CUDA:
Device 0: "GeForce GT 610" CUDA Driver Version: 8.0 CUDA Runtime Version: 5.0 CUDA Capability Major revision number: 2 CUDA Capability Minor revision number: 1 Total amount of global memory: 2080440320 bytes Number of multiprocessors: 1 Number of cores: 32 Total amount of constant memory: 65536 bytes Total amount of shared memory per block: 49152 bytes Total number of registers available per block: 32768 Warp size: 32 Maximum number of threads per block: 1024 Maximum sizes of each dimension of a block: 1024 x 1024 x 64 Maximum sizes of each dimension of a grid: 65535 x 65535 x 65535 Maximum memory pitch: 2147483647 bytes Texture alignment: 512 bytes Clock rate: 1.62 GHz Concurrent copy and execution: Yes Run time limit on kernels: Yes Integrated: No Support host page-locked memory mapping: Yes Compute mode: Default (multiple host threads can use this device simultaneously)
Then statrted recon-all with use-gpu flag. $ recon-all -all -i grp229_t1.nii -s grp229 -sd grp229 -use-gpu
But it was giving some error report continuously and I interrupted the process. Please see the attached log report and help me to use gpu effectively.
Thanks Ramesh
On 20 May 2017 at 00:56, Ramesh Babu mgrameshbabu2013@gmail.com wrote:
I am using Intel® Core™ i7-4790 CPU @ 3.60GHz × 8 processor with 15.6 GiB ram, GeForce GT 610/PCIe/SSE2 graphic card.
I have successfully installed cuda runtime version 5.0. After that I performed cuda detect and output result is pasted below.
$ cudadetect Detecting CUDA... There is 1 device supporting CUDA:
Device 0: "GeForce GT 610" CUDA Driver Version: 8.0 CUDA Runtime Version: 5.0 CUDA Capability Major revision number: 2 CUDA Capability Minor revision number: 1 Total amount of global memory: 2080440320 bytes Number of multiprocessors: 1 Number of cores: 32 Total amount of constant memory: 65536 bytes Total amount of shared memory per block: 49152 bytes Total number of registers available per block: 32768 Warp size: 32 Maximum number of threads per block: 1024 Maximum sizes of each dimension of a block: 1024 x 1024 x 64 Maximum sizes of each dimension of a grid: 65535 x 65535 x 65535 Maximum memory pitch: 2147483647 bytes Texture alignment: 512 bytes Clock rate: 1.62 GHz Concurrent copy and execution: Yes Run time limit on kernels: Yes Integrated: No Support host page-locked memory mapping: Yes Compute mode: Default (multiple host threads can use this device simultaneously)
Then statrted recon-all with use-gpu flag. $ recon-all -all -i grp229_t1.nii -s grp229 -sd grp229 -use-gpu
But it was giving some error report continuously and I interrupted the process. Please see the attached log report and help me to use gpu effectively.
Sorry, can you highlight the error message in the log? I did a quick scan and it looked like normal output to me.
Having said that, I see that you've got a display hooked up to this card, and that it's only 2 GiB of RAM. Both of those *might* give problems.
Richard
Dear Richard,
I have attached two documents. One document contains report with -use-gpu flag which I have interrupted and another document is without -use-gpu flag, completed without any error. Wherever report differ from normal run I have highlighted. Please have a look and give your suggestion.
I am using 16 gb RAM and NIVIDA 2 gb graphic card. What kind of changes should I do to use both 16 gb RAM and graphic card.
Thanks Ramesh
On Sun, May 21, 2017 at 2:12 AM, R Edgar freesurfer.rge@gmail.com wrote:
On 20 May 2017 at 00:56, Ramesh Babu mgrameshbabu2013@gmail.com wrote:
I am using Intel® Core™ i7-4790 CPU @ 3.60GHz × 8 processor with 15.6 GiB ram, GeForce GT 610/PCIe/SSE2 graphic card.
I have successfully installed cuda runtime version 5.0. After that I performed cuda detect and output result is pasted below.
$ cudadetect Detecting CUDA... There is 1 device supporting CUDA:
Device 0: "GeForce GT 610" CUDA Driver Version: 8.0 CUDA Runtime Version: 5.0 CUDA Capability Major revision number: 2 CUDA Capability Minor revision number: 1 Total amount of global memory: 2080440320 bytes Number of multiprocessors: 1 Number of cores: 32 Total amount of constant memory: 65536 bytes Total amount of shared memory per block: 49152 bytes Total number of registers available per block: 32768 Warp size: 32 Maximum number of threads per block: 1024 Maximum sizes of each dimension of a block: 1024 x 1024 x 64 Maximum sizes of each dimension of a grid: 65535 x 65535 x 65535 Maximum memory pitch: 2147483647 bytes Texture alignment: 512 bytes Clock rate: 1.62 GHz Concurrent copy and execution: Yes Run time limit on kernels: Yes Integrated: No Support host page-locked memory mapping: Yes Compute mode: Default (multiple host threads can use this device simultaneously)
Then statrted recon-all with use-gpu flag. $ recon-all -all -i grp229_t1.nii -s grp229 -sd grp229 -use-gpu
But it was giving some error report continuously and I interrupted the process. Please see the attached log report and help me to use gpu effectively.
Sorry, can you highlight the error message in the log? I did a quick scan and it looked like normal output to me.
Having said that, I see that you've got a display hooked up to this card, and that it's only 2 GiB of RAM. Both of those *might* give problems.
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 22 May 2017 at 03:10, Ramesh Babu mgrameshbabu2013@gmail.com wrote:
I have attached two documents. One document contains report with -use-gpu flag which I have interrupted and another document is without -use-gpu flag, completed without any error. Wherever report differ from normal run I have highlighted. Please have a look and give your suggestion.
I can see that the "with GPU" trace ends in a run of mri_ca_register rather suddenly, but I do not see any errors in that trace. What happened? Did mri_ca_register_cuda crash? It didn't reach the same step count as the CPU trace.
Richard
Dear Richard,
The following last line get repeated same way except (%) in the second line pasted below. On the third line "Did not call MRIsetResolution", is it error or normal process? Since repeated many times I thought there is some error.
PutOnGPU: Did not reset gca in dst 0270: dt=9.072001, rms=0.557 (0.021%), neg=0, invalid=762 LabelTerm: Did not call MRIsetResolution
Shall I repeat the process again with -use-gpu flag?
I hope mri_ca_register_cuda is working fine.
$ mri_ca_register_cuda nvcc: NVIDIA (R) Cuda compiler driver Copyright (c) 2005-2012 NVIDIA Corporation Built on Fri_Sep_21_17:28:58_PDT_2012 Cuda compilation tools, release 5.0, V0.2.1221
Driver : 8.0 Runtime : 5.0
Acquiring CUDA device Using default device CUDA device: GeForce GT 610 Help
NAME mri_ca_register
SYNOPSIS mri_ca_register <invol> <template> <outvol> .... ....(didn't past complete output)
On Mon, May 22, 2017 at 2:55 PM, R Edgar freesurfer.rge@gmail.com wrote:
On 22 May 2017 at 03:10, Ramesh Babu mgrameshbabu2013@gmail.com wrote:
I have attached two documents. One document contains report with -use-gpu flag which I have interrupted and another document is without -use-gpu
flag,
completed without any error. Wherever report differ from normal run I
have
highlighted. Please have a look and give your suggestion.
I can see that the "with GPU" trace ends in a run of mri_ca_register rather suddenly, but I do not see any errors in that trace. What happened? Did mri_ca_register_cuda crash? It didn't reach the same step count as the CPU trace.
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 22 May 2017 at 07:39, Ramesh Babu mgrameshbabu2013@gmail.com wrote:
The following last line get repeated same way except (%) in the second line pasted below. On the third line "Did not call MRIsetResolution", is it error or normal process? Since repeated many times I thought there is some error.
PutOnGPU: Did not reset gca in dst 0270: dt=9.072001, rms=0.557 (0.021%), neg=0, invalid=762 LabelTerm: Did not call MRIsetResolution
That's completely normal - it was sort of an aide-memoire for myself. Should things start messing up in interesting ways, that might be something to go and check (at the time I did the port, it was a no-op for the command I was running). So long as the nnn: dt=.... lines are popping out OK, it's running.
Richard
Dear Richard,
Yesterday I performed single subject analysis with -use-gpu flag and allowed to take its own time to complete. It's get completed within 5.8 hrs. Yes you are correct. I should not interrupt before a final output. And I have realized that gpu and cpu outputs on the terminal are not similar. Thank you for your feedback.
I have one question. When use -use-gpu flag, can I use -itkthreads 8 or -use openmp 8 flag?
Thanks Ramesh
On Tue, May 23, 2017 at 1:51 AM, R Edgar freesurfer.rge@gmail.com wrote:
On 22 May 2017 at 07:39, Ramesh Babu mgrameshbabu2013@gmail.com wrote:
The following last line get repeated same way except (%) in the second
line
pasted below. On the third line "Did not call MRIsetResolution", is it
error
or normal process? Since repeated many times I thought there is some
error.
PutOnGPU: Did not reset gca in dst 0270: dt=9.072001, rms=0.557 (0.021%), neg=0, invalid=762 LabelTerm: Did not call MRIsetResolution
That's completely normal - it was sort of an aide-memoire for myself. Should things start messing up in interesting ways, that might be something to go and check (at the time I did the port, it was a no-op for the command I was running). So long as the nnn: dt=.... lines are popping out OK, it's running.
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 23 May 2017 at 01:10, Ramesh Babu mgrameshbabu2013@gmail.com wrote:
I have one question. When use -use-gpu flag, can I use -itkthreads 8 or -use openmp 8 flag?
That's not something I've tested. I think that OpenMP at least is OK (things will run in parallel on the CPU or the GPU.
Richard
Dear Richard, This command recon-all -all -i grp237_t1.nii -s grp237 -sd grp237 -use-gpu -itkthreat 4 took 5.18 hrs and recon-all -all -i grp237_t1.nii -s grp237 -sd grp237 -use-gpu -openmp 4 took 4.69 hrs without any error.
Thank you very much for your suggestions.
Ramesh
On Tue, May 23, 2017 at 3:37 PM, R Edgar freesurfer.rge@gmail.com wrote:
On 23 May 2017 at 01:10, Ramesh Babu mgrameshbabu2013@gmail.com wrote:
I have one question. When use -use-gpu flag, can I use -itkthreads 8 or
-use
openmp 8 flag?
That's not something I've tested. I think that OpenMP at least is OK (things will run in parallel on the CPU or the GPU.
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.
Hi Ramesh,
I do not know if you have considered this at all but I at least wanted to bring it to your attention. If you have please disregard.
I've been doing I.T. for over 20 years and the Nvidia card you have listed is the Geforce 610 concerns me that all of the effort you're going through won't be worth it in the end. It is several years old, released in May of 2012, and is a very basic card. I would be surprised if the CUDA code path runs quicker than your CPU because, even when it was released, it was the lowest end card of the 600 series. Unless you have a higher end Nvidia card, such as the 970, 980, 1060 (or higher) your CPU will very likely be faster in the recon-all process.
Best,
Peter ________________________________ From: freesurfer-bounces@nmr.mgh.harvard.edu [freesurfer-bounces@nmr.mgh.harvard.edu] on behalf of Ramesh Babu [mgrameshbabu2013@gmail.com] Sent: Wednesday, May 24, 2017 3:51 AM To: Freesurfer support list Subject: Re: [Freesurfer] cuda-gpu problem
Dear Richard, This command recon-all -all -i grp237_t1.nii -s grp237 -sd grp237 -use-gpu -itkthreat 4 took 5.18 hrs and recon-all -all -i grp237_t1.nii -s grp237 -sd grp237 -use-gpu -openmp 4 took 4.69 hrs without any error.
Thank you very much for your suggestions.
Ramesh
On Tue, May 23, 2017 at 3:37 PM, R Edgar <freesurfer.rge@gmail.commailto:freesurfer.rge@gmail.com> wrote: On 23 May 2017 at 01:10, Ramesh Babu <mgrameshbabu2013@gmail.commailto:mgrameshbabu2013@gmail.com> wrote:
I have one question. When use -use-gpu flag, can I use -itkthreads 8 or -use openmp 8 flag?
That's not something I've tested. I think that OpenMP at least is OK (things will run in parallel on the CPU or the GPU.
Richard _______________________________________________ Freesurfer mailing list Freesurfer@nmr.mgh.harvard.edumailto: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.
Dear Peter,
Thank you very much for your suggestion. You mean to say, 4.6 hrs duration for recon-all process is because of cpu not gpu? If I run this process without gpu card then I should get the same speed. Right? Ok I will remove my graphic card and run directly from mother board connection which will use only cpu. I just want to know what will be difference?
If the process will be completed in ~4.6 hrs, then installing higher end graphic must reduce the duration further. If so then I will go for higher end graphic card.
Thanks Ramesh
On Wed, May 24, 2017 at 9:38 PM, Senften, Peter peter.senften@ubc.ca wrote:
Hi Ramesh,
I do not know if you have considered this at all but I at least wanted to bring it to your attention. If you have please disregard.
I've been doing I.T. for over 20 years and the Nvidia card you have listed is the Geforce 610 concerns me that all of the effort you're going through won't be worth it in the end. It is several years old, released in May of 2012, and is a very basic card. I would be surprised if the CUDA code path runs quicker than your CPU because, even when it was released, it was the lowest end card of the 600 series. Unless you have a higher end Nvidia card, such as the 970, 980, 1060 (or higher) your CPU will very likely be faster in the recon-all process.
Best,
Peter
*From:* freesurfer-bounces@nmr.mgh.harvard.edu [ freesurfer-bounces@nmr.mgh.harvard.edu] on behalf of Ramesh Babu [ mgrameshbabu2013@gmail.com] *Sent:* Wednesday, May 24, 2017 3:51 AM *To:* Freesurfer support list *Subject:* Re: [Freesurfer] cuda-gpu problem
Dear Richard, This command recon-all -all -i grp237_t1.nii -s grp237 -sd grp237 -use-gpu -itkthreat 4 took 5.18 hrs and recon-all -all -i grp237_t1.nii -s grp237 -sd grp237 -use-gpu -openmp 4 took 4.69 hrs without any error.
Thank you very much for your suggestions.
Ramesh
On Tue, May 23, 2017 at 3:37 PM, R Edgar freesurfer.rge@gmail.com wrote:
On 23 May 2017 at 01:10, Ramesh Babu mgrameshbabu2013@gmail.com wrote:
I have one question. When use -use-gpu flag, can I use -itkthreads 8 or
-use
openmp 8 flag?
That's not something I've tested. I think that OpenMP at least is OK (things will run in parallel on the CPU or the GPU.
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.
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