# [gmx-users] REMD speed calculation compared to classical MD

Chris Neale chris.neale at utoronto.ca
Wed Oct 13 20:38:54 CEST 2010

```Stefane,

If you are asking for a comparison of simulation time required, then
that is not something that can be answered in the abstract. There are
even cases in which REMD will take more total simulation time to
converge the 300K ensemble than will N simulations all at 300K. The
efficiency of REMD will be highly dependent on the size of the important
energy barriers. REMD becomes more efficient than MD as the barriers get
higher. To take that to an extreme, if you wanted to know the
distribution of x-coordinate values of one specific water molecule in a
box of pure water, REMD with 100 replicas would be 100 times less
efficient than simple MD -- there are no energy barriers and so the 99
replicas at temperatures above 300K are a total waste of time. At the
other extreme, if your energy barrier is 30 Kcal/mol at 300K, then you
will essentially never cross it at 300K and REMD, if it works, will be
infinitely more "efficient" because the replica at 300K will essentially
never sample the other side of that barrier.

There have been a few comparisons, but I don;t know of any general rule
that you can apply here.

On 10/13/10 7:42 PM, XAvier Periole wrote:

>/
/>/  Well the exchanges do not cost time ... so basically each replica runs
/>/  at the speed it would run it was a regular simulation on the number of
/>/  cpu you give each replica.
/>/
/>/  I do not think there any significant difference in NPT/NVT simulations.
/>/  Note NVT is often used for REMD
/It us, but that gives high pressure at high T, which is typically not
what you want.

If you use NPT REMD than the run time between exchanges is in general
equal to that of the coldest replica, since that will be at the highest
density (unless you go below the density maximum of water :)).

>/
/>/  XAvier.
/>/
/>/  On Oct 13, 2010, at 11:36 AM, ABEL Stephane 175950 wrote:
/>/
/>>/  Dear All,
/>>/
/>>/  For a futur project, I would like to perform the REMD calculations
/>>/  with GROMACS4.5.X. To have an estimation of CPU time required, I have
/>>/  a naive question: What is the speed of REMD compared to a classic NPT
/>>/  MD ? I am aware that the response depends a lot of factor (number of
/>>/  replica, etc.), but since i have no idea and experience with this type
/>>/  of MD, I ask the question ;).
/>>/
/>>/  The system will have approximatively 25000 atoms. Previously for this
/>>/  system, I got 11 ns/day on 32 CPU for a NPT MD with GROMOS53A6 and
/>>/  GMX4.0.5.
/>>/
/>>/
/>>/  Stefane
/>>/
/>>/  --
/>>/  gmx-users mailing listgmx-users at gromacs.org  <http://lists.gromacs.org/mailman/listinfo/gmx-users>
/>>/  http://lists.gromacs.org/mailman/listinfo/gmx-users
/>>/  Please search the archive at
/>>/  http://www.gromacs.org/Support/Mailing_Lists/Search  before posting!
/>>/  Please don't post (un)subscribe requests to the list. Use the
/>>/  www interface or send it togmx-users-request at gromacs.org.  <http://lists.gromacs.org/mailman/listinfo/gmx-users>