Re: [AMBER] GPU based TI runs

From: Song, Lin <>
Date: Sat, 9 May 2020 15:41:54 +0000

Hi Debarati,

Which version of AMBER are you using? The newest AMBER has just came out. I recommend you using that, since there were bug-fixes along the way and I’m not sure which version you are using.


On May 9, 2020, at 10:00 AM, Debarati DasGupta <<>> wrote:

Hello Prof Case,
I have used barostat=2 (MC barostat) in my input files but I have not been able to still do the NPT equilibration on the GPUs using pmemd.cuda

I am using the cpu processors for NPT (ns) equilibration, as I could not figure out the technique to do it on GPUs, but my project will probably go on for 6 months more before I can see any data produced. On cpus (16 processors) I am getting 0.6ns/day whereas my minimization/heating on GPUs were so fast at all the 12 lambdas. ( Another interesting thing I discovered was that sometimes when I keep restarting the equil runs on GPUs, some lambda values get output produced and immediately after 2-3mins of initiation of the job; my job crashes...)

50000Steps of SD and 100ps(0-50K), 100ps(50-100K), 100ps(100-200K),100 ps(200-300K), 100ps(300K) -----> these steps all together took like 1.8 hours on the GPU.
Now I m struggling with 40+ hours of NPT equilibration.
I have not heard from Dr Cerutti yet; I am hoping he may have some ideas which can help me push the 2ns of equilibration on gpuland.

Thanks Prof Case for all your inputs on my TI simulations.


From: David A Case<>
Sent: 09 May 2020 09:10
To: AMBER Mailing List<>
Cc: Dave Cerutti<>
Subject: Re: [AMBER] GPU based TI runs

On Sat, May 09, 2020, Debarati DasGupta wrote:

Is there anyone here on the mailing list who has done extensive absolute
free energy TI calculations; specifically :NPT equilibrations on GPUs and
have seen success?

This is in reference to Thermodynamic integration calculations on AMBER18.
Well, I have been working on it last ~5-6 months and it seems and
whatever I try, the 2ns of density equilibration (ntp=1) ALWAYS fails on
the GPUs.

Do you see any different behavior when using the Monte Carlo barostat?

Generally, people doing free energy TI calculations don't post much to
this email list, and I suspect that they don't follow it. Dave Cerutti
was supposed to be working on this (as I remember...could be getting
different threads mixed up here.) I'll cc- him here.


AMBER mailing list<>;!!HXCxUKc!m01x4vnLYTHKWRQwZsNr4Wms2eHkEIpQ_bi_oCIqRbxqMWkb5W-b-1ZvpLFz9NCKpwf-KxlA$

AMBER mailing list<>;!!HXCxUKc!m01x4vnLYTHKWRQwZsNr4Wms2eHkEIpQ_bi_oCIqRbxqMWkb5W-b-1ZvpLFz9NCKpwf-KxlA$

AMBER mailing list
Received on Sat May 09 2020 - 09:00:02 PDT
Custom Search