Re: [AMBER] REMD_just_1.10ns/day

From: Carlos Simmerling <carlos.simmerling.gmail.com>
Date: Tue, 24 Jan 2017 19:37:40 -0500

It's best to determine the timing using a certain number of cores in MD,
then use the same number for each replica in remd. I can't tell from your
message if your 128 CPUs is per replica or for the entire run. If for all,
then you are using 128/8=16 for each, which will probably not be as fast as
using 108 in your MD run.

Determine optimal scaling in md first and compare those results to the
published benchmarks.

On Jan 24, 2017 6:12 PM, "Thakur, Abhishek" <axt651.miami.edu> wrote:

I am sorry

I have use 108 processor for MD and aMD and have got around 7-8 and 6-7
ns/day respectively.


Here for REMD I am using 8 replicas on 128 processors.

________________________________
From: Adrian Roitberg <roitberg.ufl.edu>
Sent: Tuesday, January 24, 2017 10:09:35 AM
To: AMBER Mailing List
Subject: Re: [AMBER] REMD_just_1.10ns/day

Thakur, you need togive us much better data if we are to help youy.

What does "somewhat less processor" means?

You said you used 128 cores for REMD. How many replicas ?

How many cores did you use to run without REMD?

adrian

On 1/24/17 3:58 PM, Thakur, Abhishek wrote:
> Hi Dr.Simmerling,
>
> I have ran with same mdind file MD with some what less processor but ot
gave me around 7-8ns/day.
>
> than is why I am more shocked that REMD is giving so slow.
>
> With aMD I have got 6ns/day.
>
> ________________________________
> From: Carlos Simmerling <carlos.simmerling.gmail.com>
> Sent: Tuesday, January 24, 2017 9:50:28 AM
> To: AMBER Mailing List
> Subject: Re: [AMBER] REMD_just_1.10ns/day
>
> I would suggest doing the benchmarking without REMD. Do your MD speeds
> match published benchmarks for similar size systems? REMD should not add
> much overhead but can add complications if your computer system is not
> configured well. It's best to try to isolate the factors and test MD
alone.
>
> On Tue, Jan 24, 2017 at 1:53 PM, Thakur, Abhishek <axt651.miami.edu>
wrote:
>
>> Hi every one,
>>
>> I am running REMD on 128 processors with 10706 atoms implicitly and just
>> getting 1.1ns/day.
>>
>> Is it normal that such calculation run so slow?
>>
>> Or can there be a possibility of some fault in my script?
>>
>> Here is what I am using
>>
>>
>> &cntrl
>> irest=1, ntx=5,
>> nstlim=100, dt=0.004,
>> ntt=3, gamma_ln=50.0,
>> temp0=XXXXX, ig=RANDOM_NUMBER,
>> ntc=2, ntf=2, nscm=1000,
>> ntb=0, igb=5,
>> cut=999.0, rgbmax=15.0,
>> ntpr=100, ntwx=1000, ntwr=100000,
>> nmropt=1,
>> numexchg=100000,
>> saltcon=0.2,
>> /
>> &wt TYPE='END'
>> /
>> DISANG=WT_chir.dat
>>
>>
>> Thanking you,
>>
>> -AT
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.
ambermd.org_mailman_listinfo_amber&d=DwICAg&c=y2w-uYmhgFWijp_IQN0DhA&r=
0Hah93XWYYBgmROOVcaccg&m=apxSRhIqQPlMirdn8KNpkauo4szznBBPRUM29yIzThY&s=
uFr0oJ0PFsqhiHJK2FGkMKPXGwboVmLKsWA5gFLRylA&e=
>>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.
ambermd.org_mailman_listinfo_amber&d=DwICAg&c=y2w-uYmhgFWijp_IQN0DhA&r=
0Hah93XWYYBgmROOVcaccg&m=apxSRhIqQPlMirdn8KNpkauo4szznBBPRUM29yIzThY&s=
uFr0oJ0PFsqhiHJK2FGkMKPXGwboVmLKsWA5gFLRylA&e=
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.
ambermd.org_mailman_listinfo_amber&d=DwICAg&c=y2w-uYmhgFWijp_IQN0DhA&r=
0Hah93XWYYBgmROOVcaccg&m=3BttAiSNFBYh7c-zhD-OLUDh2n7o4VQKgc_f0St-eXA&s=
h2k0ACL47cfcuGJ5adD1biunFJg8Nt3zo8AwmEHa3M4&e=

--
Dr. Adrian E. Roitberg
University of Florida Research Foundation Professor.
Department of Chemistry
University of Florida
roitberg.ufl.edu
352-392-6972
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
https://urldefense.proofpoint.com/v2/url?u=http-3A__lists.
ambermd.org_mailman_listinfo_amber&d=DwICAg&c=y2w-uYmhgFWijp_IQN0DhA&r=
0Hah93XWYYBgmROOVcaccg&m=3BttAiSNFBYh7c-zhD-OLUDh2n7o4VQKgc_f0St-eXA&s=
h2k0ACL47cfcuGJ5adD1biunFJg8Nt3zo8AwmEHa3M4&e=
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Tue Jan 24 2017 - 17:00:02 PST
Custom Search