Running with SPFP should be fine for production runs. However, you'll get slightly different rounding results based on GPU architecture. This should be fine.
DPFP has more precision so it will have the same results on all hardware.
Charlie
________________________________________
From: Guqin Shi [shi.293.osu.edu]
Sent: Thursday, March 24, 2016 12:24 PM
To: AMBER Mailing List
Subject: Re: [AMBER] [AmberTools15 & Amber14 Installation] Make Test Possible Failures Can Be Ignored or Not
LOLs, thank you Charlie!!
I did notice that DPFP is for debugging, not for production run. Is still
suggested to build cuda_DPFP even if it will not really used?
-Guqin
2016-03-24 15:19 GMT-04:00 Charles Lin <clin92.ucsd.edu>:
> Compile with -cuda_DPFP and then run the tests and they should all pass.
> Do note this will link pmemd.cuda to the DPFP version (slower) not the SPFP
> version (faster). You're running with single precision which will have
> some round off errors.
>
> ________________________________________
> From: Guqin Shi [shi.293.osu.edu]
> Sent: Thursday, March 24, 2016 12:12 PM
> To: AMBER Mailing List
> Subject: [AMBER] [AmberTools15 & Amber14 Installation] Make Test Possible
> Failures Can Be Ignored or Not
>
> Hi all,
>
> I've just upgraded to AmberTools15 and recompiled everything following a
> tutorial kindly written by Jason Swails. (
> http://jswails.wikidot.com/installing-amber14-and-ambertools14)
>
> I did:
> make test.serial
> make test.parallel
> everything went well. No comparison diff.
>
> then I did:
> make test.cuda
> ./test_amber_cuda_parallel.sh
> Around half of tests showed comparison diffs, which is expected for cuda.
> Most of them are round errors as I could tell.
>
> However, there are some "possible failures" which I am not sure if I can
> safely ignore them or not. Below is part of them:
>
> ======================================================================
>
> possible FAILURE: check irest1_ntt0_igb8_ntc2.out.dif
> /usr/local/amber14/test/cuda/gb_ala3
> 53c53
> < Frozen or restrained atoms:
> > gb_alpha_hnu = 0.53705, gb_beta_hnu = 0.36286, gb_gamma_hnu
> = 0.11670
> 54c54
> < ibelly = 0, ntr = 0
> > gb_alpha_cnu = 0.33167, gb_beta_cnu = 0.19684, gb_gamma_cnu
> = 0.09342
> 55c55
> < Molecular dynamics:
> > gb_alpha_nnu = 0.68631, gb_beta_nnu = 0.46319, gb_gamma_nnu
> = 0.13872
> 56c56
> < nstlim = 20, nscm = 50, nrespa = 1
> > gb_alpha_onu = 0.60634, gb_beta_onu = 0.46301, gb_gamma_onu
> = 0.14226
> 57c57
> < t = 0., dt = 0.00050, vlimit = -1.00000
> > gb_alpha_pnu = 0.41836, gb_beta_pnu = 0.29005, gb_gamma_pnu
> = 0.10642
> 58c58
> < SHAKE:
> > Frozen or restrained atoms:
> 59c59
> < ntc = 2, jfastw = 0
> > ibelly = 0, ntr = 0
> 60c60
> < tol = 0.00001
> > Molecular dynamics:
> 61c61
> <
>
> --------------------------------------------------------------------------------
> > nstlim = 20, nscm = 50, nrespa = 1
> 62c62
> < 3. ATOMIC COORDINATES AND VELOCITIES
> > t = 0., dt = 0.00050, vlimit = -1.00000
> 63c63
> <
>
> --------------------------------------------------------------------------------
> > SHAKE:
> 64c64
> < ACE
> > ntc = 2, jfastw = 0
> 65c65
> < begin time read from input coords = 1050.000 ps
> > tol = 0.00001
> 66c66
> < Number of triangulated 3-point waters found: 0
> >
>
> --------------------------------------------------------------------------------
> 67c67
> <
>
> --------------------------------------------------------------------------------
> > 3. ATOMIC COORDINATES AND VELOCITIES
> 68c68
> < 4. RESULTS
> >
>
> --------------------------------------------------------------------------------
> 69c69
> <
>
> --------------------------------------------------------------------------------
> > ACE
> 70c70
> < NSTEP = 1 TIME(PS) = 1050.001 TEMP(K) = 337.60 PRESS
> = 0.
> > begin time read from input coords = 1050.000 ps
> 71c71
> < Etot = 21.0035 EKtot = 32.8726 EPtot =
> -11.8690
> > Number of triangulated 3-point waters found: 0
> 72c72
> < BOND = 3.0496 ANGLE = 15.6236 DIHED =
> 25.7299
> >
>
> --------------------------------------------------------------------------------
>
> ======================================================================
>
> I didn't really see what the failure is or is it related to round errors or
> not. Anybody can help look at that? Can I safely ignore them? I can also
> upload full .diff file if needed.
>
> Thanks for your time!
> Guqin
>
> --
> Guqin SHI
> The Ohio State University
> College of Pharmacy
> 500 W. 12th Ave.
> Columbus, OH, 43210
> (614)688-3531
> _______________________________________________
> 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
>
--
Guqin SHI
The Ohio State University
College of Pharmacy
500 W. 12th Ave.
Columbus, OH, 43210
(614)688-3531
_______________________________________________
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 Thu Mar 24 2016 - 12:30:10 PDT
