Folks,

 

I am a computer administrator trying to put together a proposal for a new University shared resource compute cluster.

Some fMRI image analysts have suggested that GPUs are frequently used in the field and that in particular, double precision GPUs are the hardware of choice.

 

Double precision GPU can be an order of magnitude more expensive to purchase than their single precision version.

 

Google sees many opinions

For fMRI…

 

These articles suggest single precision:

https://blogs.nvidia.com/blog/2010/04/21/the-world-is-parallel-gpus-speed-medical-imaging/

https://www.nvidia.com/content/cuda/spotlights/anders-eklund-vtcr.html

https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3953750/

 

This page from NVIDIA suggests the double precision Tesla but when you click on all of their links the papers were written with single precisions cards!

https://www.nvidia.com/object/medical_imaging.html

 

 

I realize that this appeal for information may not be appropriate for this list but I would appreciate some guidance, if anyone has the time and interest.

The goal is to create a hardware resource that would be useful to a broad user base including fMRI analysts. I am not after a high throughput clinical system but more of a software development capability.

What sort of GPUs should I be pricing?

 

Thanks in advance

 

Starr Hazard, Ph.D.

Computational Biology Resource Center

Medical University of South Carolina

Charleston, SC