Hi Sofia,
You could also use H mass repartitioning. Run the prmtop through parmed witht he HMassRepartition action. Then increase your timestep to 4fs. That will get you about an 85% or so performance boost.
That said the numbers you report for a 260K atom system are on the low side, even with a 10 angstrom cutoff so something else is up. Can you try running the AMBER 16 benchmark on the machines you are using and see if you get similar numbers to here:
http://ambermd.org/gpus/benchmarks.htm <
http://ambermd.org/gpus/benchmarks.htm>
Here's the benchmark suite: http://ambermd.org/Amber16_Benchmark_Suite.tar.bz2 <http://ambermd.org/Amber16_Benchmark_Suite.tar.bz2>
Set edit the run_bench_CPU+GPU.sh, set CPU_COUNT=0 and GPU_COUNT=1 (assuming you have just 1 GPU in each machine). Then run it with:
./run_bench_CPU+GPU.sh >& bench.log &
tail -f bench.log
I'd be interested in the performance you get - if this is substantially below what's on the AMBER website then your installation needs some more investigation.
All the best
Ross
> On Apr 5, 2017, at 07:35, Andreas Tosstorff <andreas.tosstorff.cup.uni-muenchen.de <mailto:andreas.tosstorff.cup.uni-muenchen.de>> wrote:
>
> Try using barostat=2, this is a monte carlo barostat optimized for GPU
> use. You can find more info in the amber16 manual.
>
>
> Best,
>
> Andy
>
>
> On 04/05/2017 04:27 PM, Sofia Vasilakaki wrote:
>> Hi to all,
>>
>> I am running an MD simulation of a protein attached to a lipid membrane in
>> water. The system is 262628 atoms and the input file is:
>>
>> imin=0, ntx=5, irest=1,
>> ntc=2, ntf=2,
>> nstlim=3500000, ntt=3,gamma_ln=1.0,
>> temp0=310.0,
>> ntpr=20000, ntwr=500000,ntwx=20000,
>> dt=0.002,ig=-1,
>> ntb=2, ntp=2, cut=10.0, barostat=1, ioutfm=1, ntxo=2
>>
>>
>> On a Tesla K40m is 7.50ns/day, on GTX980 P2P is 10.35ns/day and on GTX1080
>> is 12.57ns/day. Is there any way I could improve performance?
>>
>> Thank you in advance.
>>
>> Kind Regards,
>> Sofia V.
>>
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org <mailto:AMBER.ambermd.org>
>> http://lists.ambermd.org/mailman/listinfo/amber
>
> --
> M.Sc. Andreas Tosstorff
> Lehrstuhl für Pharmazeutische Technologie und Biopharmazie
> Department Pharmazie
> LMU München
> Butenandtstr. 5-13 ( Haus B)
> 81377 München
> Germany
> Tel.: +49 89 2180 77059
>
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org <mailto: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 Wed Apr 05 2017 - 16:30:03 PDT
