Increase simulation speed

[Mahta Mousavi]

Hello, 

I have a couple of questions regarding ways to increase the simulation speed especially during parameter sweeps. 

1. Is there a gpu-compatible version of TVB available? 

2. Is there any other ways to speed up each round of simulation? I know that in the gui, one can assign the amount of memory accessible for the simulations, but I'm not sure if I can do something similar when I run the codes in jupyter notebook? 

Thanks very much for your help. 

Best Regards,

Mahta

[Marmaduke Woodman]

Hi 

TVB has support for code generation for faster CPU or GPU parameter sweeps, but it is not yet exposed in a user friendly fashion.  We expect to remedy this during the next few months.   If you want to test it earlier, we can help you try it out. 

cheers,

Marmaduke

--
You received this message because you are subscribed to the Google Groups "TVB Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tvb-users+...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/e4016b4d-9449-44de-b14f-b73f915d1373n%40googlegroups.com.

[Wayne Wong]

Hi Dr. Woodman, 

Greeting! I have a similar question about the simulation speed. Now I am trying to simulate bold signals (with Gamma kernel) in 10min with Stefanescu - Jirsa 3D model in 16 regions. Parameter space exploration was performed but I only used one parameter range (global coupling, [0.01, 0.06, 0.001]) during my simulation.  Even I ran my jobs on HPC system, it cost me around 17 hours to run a single case. I am wondering if it is a normal simulation time or not. Does TVB have any optimization method to improve the speed of a long simulation in HPC system? 

Thanks so much! Looking forward to your reply.

Best Regards, 

Yile

my code example: https://github.com/yilewang/Gc_detect_script/blob/master/go_sim_hpc.py 

[Mahta Mousavi]

Dear Marmaduke, 

Thank you very much for your response. I look forward to the updates in the next few months. 

Best Regards,

Mahta

To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/Y9SBSQ.JJ0J0D5SY9HL%40univ-amu.fr.

[WOODMAN Michael]

Hi

On 3 May 2021, at 20:43, Wayne Wong <ylww...@gmail.com> wrote:

 Stefanescu - Jirsa 3D model in 16 regions. Parameter space exploration was performed but I only used one parameter range (global coupling, [0.01, 0.06, 0.001]) during my simulation.  Even I ran my jobs on HPC system, it cost me around 17 hours to run a single case

This model is the slowest one in TVB, and admittedly we have not spent much time optimising it.  That said, 17 hours is too long, as this could still run in real time in principle, and there may be some other issue occurring.  What version of TVB are you using, and what were your job submission parameters?

Cheers,

Marmaduke

[Wayne Wong]

Dear Marmaduke, 

Thanks so much for your reply. I consulted to a professor at CS department of my university then I adopted a new job submission way to solve this problem. In previous session I was running my jobs in series instead of in parallel so it was the reason why it cost me a huge amount of time. Now I am using TACC/launcer (https://github.com/TACC/launcher) to submit my job so everything is great now. 

Best Regards, 

Yile

[WOODMAN Michael]

hi

thanks for the reply, how long do the individual jobs take now?

cheers,

Marmaduke

---

From: tvb-...@googlegroups.com <tvb-...@googlegroups.com> on behalf of Wayne Wong <ylww...@gmail.com>
Sent: Thursday, May 6, 2021 11:56:24 PM
To: TVB Users
Subject: Re: [TVB] Increase simulation speed

 

--

You received this message because you are subscribed to the Google Groups "TVB Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to tvb-users+...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/tvb-users/f4dea2f1-ce40-4e32-a0aa-d4fb6ea0072en%40googlegroups.com.

[Wayne Wong]

Hi Marmaduke, 

Now it cost me 10 hours for a individual job and but since it is all parallelized so I can run 30 cases with parameter range [0.01, 0.06, 0.001] at the same time. The overall time usage is around 20 hours for 30 subjects' simulations.

Best Regards, 

Yile