Dear Ross and Dow,
thank you!
Actually I'd love to try these scripts for GPU validation if they are
available.
Sincerely,
Nick
Nikolay Kuzmich
Department of Drug Safety,
Research Institute of Influenza,
WHO National Influenza Centre of Russia,
15/17 Professor Popov St.,
Saint-Petersburg
Message: 13
Date: Fri, 29 Jun 2018 14:28:45 -0400
From: Dow Hurst <dphurst.uncg.edu>
Subject: Re: [AMBER] Geforce GTX 1080Ti testing on Amber18/CUDA 9.2
To: AMBER Mailing List <amber.ambermd.org>
Message-ID:
<CAF3ati8Jpjuw-j2PP8HDX_fOR0UGbXjz_7h4Oh2Veb3h+3Vzbg.mail.gmail.com>
Content-Type: text/plain; charset="UTF-8"
You should rerun your test at least 20 times in a row and compare the final
data. Three times isn't enough. If you get a variation, then try a smaller
sized system. GPUs ought to be tested for memory errors every year of use
at minimum. Are you aware of Dr. Ross Walker's gpu validation scripts?
Sincerely,
Dow
?Dow Hurst, Research Scientist
340 Sullivan Science Bldg.
Dept. of Chem. and Biochem.
University of North Carolina at Greensboro
PO Box 26170 Greensboro, NC 27402-6170
On Fri, Jun 29, 2018 at 8:02 AM Nikolay N. Kuzmich <nnkuzmich.gmail.com>
wrote:
Message: 4
Date: Mon, 2 Jul 2018 07:57:26 -0400
From: Ross Walker <ross.rosswalker.co.uk>
Subject: Re: [AMBER] Geforce GTX 1080Ti testing on Amber18/CUDA 9.2
To: AMBER Mailing List <amber.ambermd.org>
Message-ID: <4F4C2AD8-3391-4534-8E6E-791C0A688F01.rosswalker.co.uk>
Content-Type: text/plain; charset=us-ascii
Hi Nick,
The random crashing shortly after starting the calculation is a known bug
in AMBER 18. It has been pointed out on the developers list and is being
worked on.
Until it is fixed I suggest only using Amber 16 to test GPUs.
All the best
Ross
Ps. The IEEE message is innocuous and you can ignore it.
On Fri, Jun 29, 2018 at 8:02 AM Nikolay N. Kuzmich <nnkuzmich.gmail.com>
wrote:
> Dear Amber users and developers,
>
> I would like to ask you if the GPU testing should be considered successful
> or not.
>
> The purchased GPU was tested on 3 systems - 26, 600 and 1600 thousand
> atoms.
> The NPT runs (pmemd.cuda) after heating with the same seed were performed
> 3 times for each system.
> For each system the numerical results in the mdout files were identical,
> but in one case the simulation of the biggest system stopped after 3 steps
> with message
> "cudaMemcpy GpuBuffer::Download failed an illegal memory access was
> encountered".
> The message
> "Note: The following floating-point exceptions are signalling:
> IEEE_UNDERFLOW_FLAG IEEE_DENORMAL" also occured every time.
> The NPT/pmemd.cuda simulations followed NVT heating and NPT initial
> equilibration with pmemd.MPI.
> The mdin file:
>
> Equil_density_NPT
> &cntrl
> imin = 0, irest = 1, ntx = 5,
> ntb = 2, pres0 = 1.0, ntp = 1,
> taup = 0.1, ig=200000,
> cut = 9.0, ntr = 1,
> ntc = 2, ntf = 2,
> tempi = 300.0, temp0 = 300.0,
> ntt = 3, gamma_ln=1.0, nscm=5000,
> nstlim = 600000, dt = 0.002,
> ntpr = 1000, ntwx = 1000, ntwr = 10000,
> ioutfm=1, iwrap=1,
> restraintmask='!:WAT & !.H= & !:NA',
> restraint_wt=100.0,
> /
>
> [nstlim number actually varied for each system].
>
> Kind regards,
> Nick
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Jul 04 2018 - 12:30:03 PDT
