Differences between revisions 31 and 32
Deletions are marked like this. Additions are marked like this.
Line 6: Line 6:
A 2GHz or faster processor, at least 4GB of RAM, and a 3D graphics card (with a GPU and its own graphics memory) with accelerated OpenGL drivers, is recommended. Freesurfer is highly CPU and memory intensive (and moderately disk intensive), so concentrate on boosting those performance aspects (more memory is better...'''8GB is highly recommended!'''). Most modern video graphics card will perform fine, but be aware that graphics cards that use CPU memory as video memory will have a noticeably slow redraw rate. CUDA v3.2, compute capability 1.3 or higher, and the current NVIDIA driver are recommended to run GPU-optimized functions in FreeSurfer (although GPU-optimized functions are optional). A 17" (or greater!) monitor is also recommended. A 2GHz or faster processor, at least 4GB of RAM, and a 3D graphics card (with a GPU and its own graphics memory) with accelerated OpenGL drivers, is recommended. Freesurfer is highly CPU and memory intensive (and moderately disk intensive), so concentrate on boosting those performance aspects (more memory is better...'''8GB is highly recommended!'''). With the inclusion on new OpenMP-enabled code in freesurfer, multiple-cores can now be accessed by recon-all. This means you should also consider purchasing a multi-core machine if you intend to process multiple subjects at the same time. If you do this, be sure to purchase 4GB per subject, that is, if you would like to process four subjects at once, your machine must have four cores and 16GB of memory.

Most modern video graphics card will perform fine, but be aware that graphics cards that use CPU memory as video memory will have a noticeably slow redraw rate. NVidia graphics cards have been found to work best with freesurfer. While freesurfer includes some GPU support, it is not actively supported anymore, so purchasing a GPU card explicitly for freesurfer is not recommended, in lieu of purchasing multiple cores and using the OpenMP functionality. A 19" (or greater!) monitor is also recommended.
Line 14: Line 16:
From post to Freesurfer list by '''Nick Schmansky, 2013-04-24''', regarding '''suggested specs for a Linux box''' Mailing list notes regarding '''suggested specs for a Linux box''':
Line 16: Line 18:
 * CPU: Intel vs AMD: no preference. also, CPU speed is not critical. Its better to have at least 4GB, and budget 4GB per subject you want to<<BR>> process simultaneously. Try to get a 'Sandy bridge' motherboard architecture or whatever is newer. This allows better addressing of<<BR>> scattered memory which is common in freesurfer, and it alone accounts for a 5% or so speedup. AMD might have a similar memory controller.
 * Video card
  * ATI (AMD) vs nVidia: We still spec nVidia because we don't have problems with its OpenGL-X driver under linux. Perhaps ATI has finally supplied one that works, but I haven't tried in a couple years. ATI cards on the Mac work fine with freesurfer though.
  * GPU: We no longer support [further] CUDA or GPU development because of lack of resources and difficulty, in preference to using OpenMP, which uses CPU cores. We will continue to support the existing GPU CUDA code that runs on recon-all with the -use-gpu switch. By support I mean keeping it running with each new nvidia cuda driver release as best as we can.
 * CPU: Intel vs AMD: no preference. also, CPU speed is not critical. It is better to have at least 4GB, and budget 4GB per subject you want to process simultaneously. Try to get an 'Ivy Bridge' motherboard architecture or whatever is newer. This allows better addressing of scattered memory which is common in freesurfer, and it alone accounts for a 5% or so speedup. AMD might have a similar memory controller.
 * Graphics card:
  * ATI (AMD) vs nVidia: We still spec nVidia because we don't have problems with its OpenGL-X driver under linux. Perhaps ATI has finally supplied one that works, but this hasn't been attempted in a couple years. ATI cards on the Mac work fine with freesurfer though.
  * GPU: We no longer support [further] CUDA or GPU development because of lack of resources and difficulty, in preference to using OpenMP, which uses CPU cores. We will continue to support the existing GPU CUDA code that runs on recon-all with the -use-gpu switch. By support, keeping it running with each new nvidia cuda driver release as best as we can.
Line 21: Line 23:
 * Linux: Both freesurfer and fsl groups use Centos in our Centers. They seem to work well.  * Linux: Both freesurfer and fsl groups use CentOS 6 in our Centers. It seems to work well.

top | next

FreeSurfer System Requirements

Freesurfer is available for the Linux and Mac OS X operating systems. It can also be used on Windows with VirtualBox (but runs a bit slower). The download page lists the supported platforms. Matlab is required to run the FS-FAST component of Freesurfer (but Matlab is not required to run the reconstruction utilities).

A 2GHz or faster processor, at least 4GB of RAM, and a 3D graphics card (with a GPU and its own graphics memory) with accelerated OpenGL drivers, is recommended. Freesurfer is highly CPU and memory intensive (and moderately disk intensive), so concentrate on boosting those performance aspects (more memory is better...8GB is highly recommended!). With the inclusion on new OpenMP-enabled code in freesurfer, multiple-cores can now be accessed by recon-all. This means you should also consider purchasing a multi-core machine if you intend to process multiple subjects at the same time. If you do this, be sure to purchase 4GB per subject, that is, if you would like to process four subjects at once, your machine must have four cores and 16GB of memory.

Most modern video graphics card will perform fine, but be aware that graphics cards that use CPU memory as video memory will have a noticeably slow redraw rate. NVidia graphics cards have been found to work best with freesurfer. While freesurfer includes some GPU support, it is not actively supported anymore, so purchasing a GPU card explicitly for freesurfer is not recommended, in lieu of purchasing multiple cores and using the OpenMP functionality. A 19" (or greater!) monitor is also recommended.

If you're running Freesurfer on a server, then allocate 4GB of memory to your job.

Freesurfer requires about 8.5GB of disk space for the full installation, which includes the Freesurfer binaries, support libraries, the MINC toolkit, and sample MRI data.

The volume and surface data files produced by Freesurfer for a typical subject (assuming two structural MRI scans of source data) consume about 370MB of disk space. The full tutorial dataset is 18GB.

Mailing list notes regarding suggested specs for a Linux box:

  • CPU: Intel vs AMD: no preference. also, CPU speed is not critical. It is better to have at least 4GB, and budget 4GB per subject you want to process simultaneously. Try to get an 'Ivy Bridge' motherboard architecture or whatever is newer. This allows better addressing of scattered memory which is common in freesurfer, and it alone accounts for a 5% or so speedup. AMD might have a similar memory controller.
  • Graphics card:
    • ATI (AMD) vs nVidia: We still spec nVidia because we don't have problems with its OpenGL-X driver under linux. Perhaps ATI has finally supplied one that works, but this hasn't been attempted in a couple years. ATI cards on the Mac work fine with freesurfer though.
    • GPU: We no longer support [further] CUDA or GPU development because of lack of resources and difficulty, in preference to using OpenMP, which uses CPU cores. We will continue to support the existing GPU CUDA code that runs on recon-all with the -use-gpu switch. By support, keeping it running with each new nvidia cuda driver release as best as we can.
  • Disk: Freesurfer is not disk intensive, so SSD is not a benefit. Budget 400MB per subject for storage.
  • Linux: Both freesurfer and fsl groups use CentOS 6 in our Centers. It seems to work well.

SystemRequirements (last edited 2018-01-16 12:35:17 by AndrewHoopes)