August 27, 2019 at 5:52 pm #1869todflakParticipant
(I see there was a similar question last year, but that was not answered.)
I am putting together a new computer, and I specifically want to improve the speed of svreg processing. Would svreg be able to take advantage of many cores/threads?? I am considering an AMD Ryzen 7 3700X (8 Cores, 16 Threads) versus one of the AMD Threadripper processors (32 cores / 64 threads). My overall question is: will a single svreg process take advantage of all those many cores, to complete the registration much more quickly?
Any other advice on hardware selection is appreciated!
August 29, 2019 at 10:43 am #1879Anand JoshiModerator
SVReg definitely makes use of available cores. So for surface registration, it processes 2 cortical hemispheres on 2 separate threads. During volume registration, whenever possible, it processes 3 coordinates on 3 threads.
Moreover, each of the separate threads use multiple CPUs during various operations such as conjugate gradient minimization and interpolations.
So yes, the more cores, the better it is for SVReg processing.
Additionally, you can also process multiple subjects simultaneously if you have computing power.
The system you mention is definitely great (both).
August 29, 2019 at 10:53 am #1880todflakParticipant
Thanks for that!
January 10, 2020 at 5:35 am #2114stevemarkhamParticipant
Your post is one of precious few on this topic, which I expected to generate many more opinions than it did. Did increasing core-count improve your svreg times significantly? Did you put in a beefy GPU in your build, and can you tell if it is being used at all? How much RAM did you put in, and can you tell the utilization on that?
I’m in the same boat, building a new system specifically to run BrainSuite as fast as possible. I’ve got budget, but no matter how much budget I have, I still need to figure out whether to spend it on cores, RAM, or GPU, so I’d love to learn from your experience if possible.
- You must be logged in to reply to this topic.