Re: [gmx-users] Different Results with Different Platforms
Thank you very much, Justin. I was take note of this in subsequent runs. Thank you so, so much for your assistance. It is greatly appreciated! -- Best wishes, Kelechi -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
On 10/6/16 7:13 AM, Kelechi Okoroafor wrote: Thank you for the tip, Justin. I incorrectly thought that the coordinates and box dimensions were sufficient to preserve the pressure of the system. I shall make a habit of restarting simulations with the checkpoint file rather than .gro files. My sequence of commands were: - gmx grompp -f npt.mdp -c EM.gro -p sys.top -o NPT1.tpr -n index.ndx -maxwarn -1 - gmx mdrun -v -deffnm NPT1 -nb gpu -tunepme - gmx editconf -f NPT1.gro -o NPT1.pdb -c -d 0.00 - gmx editconf -f NPT1.pdb -o NPT1.gro -c -d 0.00 - If you just want to convert file formats, omit -c -d in your editconf command and nothing will change. By using -d 0.00, you will be adjusting the system size (you will see this by comparing the box vectors in the input and output files). You're ensuring a bad discontinuity by doing this. My recommendation is to always use unmodified files when continuing runs; it's basically bulletproof. If you want a different format for visualization, that's fine, but just use it for visualization. Your original NPT1.gro should have been fine, but having transformed to .pdb while manipulating the box, then back, you're just invalidating anything that was done previously. -Justin gmx grompp -f pme.mdp -c NPT1.gro -p sys.top -o NVT1.tpr -n index.ndx -maxwarn -1 - gmx mdrun -v -deffnm NVT1 -nb gpu -tunepme -- == Justin A. Lemkul, Ph.D. Ruth L. Kirschstein NRSA Postdoctoral Fellow Department of Pharmaceutical Sciences School of Pharmacy Health Sciences Facility II, Room 629 University of Maryland, Baltimore 20 Penn St. Baltimore, MD 21201 jalem...@outerbanks.umaryland.edu | (410) 706-7441 http://mackerell.umaryland.edu/~jalemkul == -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
Thank you for the tip, Justin. I incorrectly thought that the coordinates and box dimensions were sufficient to preserve the pressure of the system. I shall make a habit of restarting simulations with the checkpoint file rather than .gro files. My sequence of commands were: - gmx grompp -f npt.mdp -c EM.gro -p sys.top -o NPT1.tpr -n index.ndx -maxwarn -1 - gmx mdrun -v -deffnm NPT1 -nb gpu -tunepme - gmx editconf -f NPT1.gro -o NPT1.pdb -c -d 0.00 - gmx editconf -f NPT1.pdb -o NPT1.gro -c -d 0.00 - gmx grompp -f pme.mdp -c NPT1.gro -p sys.top -o NVT1.tpr -n index.ndx -maxwarn -1 - gmx mdrun -v -deffnm NVT1 -nb gpu -tunepme -- Best wishes, Kelechi -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
On 10/5/16 9:46 PM, Kelechi Okoroafor wrote: Thank you very much, Justin. So I did not restart the simulation directly. One of my desktops had a hardware malfunction so I had to move my files between desktops. Before moving my files, I converted my final npt.gro trajectory into a .pdb file (for ease of vmd visualisation as vmd often crashes while viewing .gro files), then converted the files back to .gro to restart the simulation in a different desktop. Somewhere in between my file conversion, the trajectories must have been edited to cause such drastic pressure increase. The difference between .pdb and .gro is negligible. Not preserving the state by using the checkpoint basically negates any previous simulation as you will be restarting from a random state. But without an exact sequence of your commands, it's pure guesswork and is unproductive for all involved. -Justin -- == Justin A. Lemkul, Ph.D. Ruth L. Kirschstein NRSA Postdoctoral Fellow Department of Pharmaceutical Sciences School of Pharmacy Health Sciences Facility II, Room 629 University of Maryland, Baltimore 20 Penn St. Baltimore, MD 21201 jalem...@outerbanks.umaryland.edu | (410) 706-7441 http://mackerell.umaryland.edu/~jalemkul == -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
Thank you very much, Justin. So I did not restart the simulation directly. One of my desktops had a hardware malfunction so I had to move my files between desktops. Before moving my files, I converted my final npt.gro trajectory into a .pdb file (for ease of vmd visualisation as vmd often crashes while viewing .gro files), then converted the files back to .gro to restart the simulation in a different desktop. Somewhere in between my file conversion, the trajectories must have been edited to cause such drastic pressure increase. -- Best wishes, Kelechi -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
On 10/5/16 8:45 PM, Kelechi Okoroafor wrote: Thank you very much, James. My apologies, I miscommunicated myself. I meant to say that I used NPT simulation to equilibrate the pressure to 1bar as desired. I had planned to do NVT simulation after the NPT simulation, expecting the pressure to fluctuate reasonably about the mean value of 1bar. Unfortunately, editconf edited the trajectory such that during the NVT simulation, the pressure fluctuated about the mean value of 900bar causing the observed unusual behaviour. I don't see how that's possible. If you move from NPT -> NVT, the ensemble that is preserved is the one that is present in the final snapshot; even if the average over the NPT run is 1 bar, there is no guarantee that instantaneous value is 1 bar, though the density is likely correct. Where was editconf involved in this process? If you're continuing a run and changing ensemble, you should be invoking grompp -t with the previous checkpoint. Any other use of GROMACS tools introduces discontinuity. -Justin -- == Justin A. Lemkul, Ph.D. Ruth L. Kirschstein NRSA Postdoctoral Fellow Department of Pharmaceutical Sciences School of Pharmacy Health Sciences Facility II, Room 629 University of Maryland, Baltimore 20 Penn St. Baltimore, MD 21201 jalem...@outerbanks.umaryland.edu | (410) 706-7441 http://mackerell.umaryland.edu/~jalemkul == -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
Thank you very much, James. My apologies, I miscommunicated myself. I meant to say that I used NPT simulation to equilibrate the pressure to 1bar as desired. I had planned to do NVT simulation after the NPT simulation, expecting the pressure to fluctuate reasonably about the mean value of 1bar. Unfortunately, editconf edited the trajectory such that during the NVT simulation, the pressure fluctuated about the mean value of 900bar causing the observed unusual behaviour. -- Best wishes, Kelechi -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
That is not a bug. The idea of NVT is it keeps your volume constant not your pressure. In theory the pressure should stay fairly constant for a given temperature as the solvent pushes against the boundaries or itself to a fairly constant extent that depends on its velocities. I have not however heard of anyone usually pressure coupling (NPT) for equilibration only but I imagine it should work. You'd probably be best to keep the velocities to be sure rather than generating them again. Did you not use the same mdp files in the same order on all machines? Best wishes James > As advised, I compared the systems between the different platforms and > found the bug. > I had initially equilibrated the pressure of my system in an NPT run > before > doing an NVT run on my desktop. > Unbeknownst to me, editconf had adjusted the coordinates of my molecules > such that the pressure during the NVT run was almost 3 orders of magnitude > higher than the initially equilibrated pressure, hence the unusual > behaviour. > I just did an NPT run on my desktop and it is loooking good. > This bug cost me some frustrating weeks of my life but on the bright side, > I may just get a publication from studying it. :D > > Thank you so, so much Mark, Justin and Szilárd. > I really appreciate your assistance! > -- > Gromacs Users mailing list > > * Please search the archive at > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before > posting! > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists > > * For (un)subscribe requests visit > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send > a mail to gmx-users-requ...@gromacs.org. -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
As advised, I compared the systems between the different platforms and found the bug. I had initially equilibrated the pressure of my system in an NPT run before doing an NVT run on my desktop. Unbeknownst to me, editconf had adjusted the coordinates of my molecules such that the pressure during the NVT run was almost 3 orders of magnitude higher than the initially equilibrated pressure, hence the unusual behaviour. I just did an NPT run on my desktop and it is loooking good. This bug cost me some frustrating weeks of my life but on the bright side, I may just get a publication from studying it. :D Thank you so, so much Mark, Justin and Szilárd. I really appreciate your assistance! -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
Thank you so much, Szilárd. Yes, it passed the regression tests. Let me explore system characteristics and see how they differ between the platforms. -- Best wishes, Kelechi -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
Thank you, Justin. As suggested, I re-ran the regression tests and got the result: All 16 simple tests PASSED All 45 complex tests PASSED All 142 kernel tests PASSED All 10 freeenergy tests PASSED All 12 rotation tests PASSED All 0 extra tests PASSED All 42 pdb2gmx tests PASSED I shall keep the simulation running and see what happens. -- Best wishes, Kelechi -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
If the regressiontests pass, I'd recommend trying to identify differences measurable/reproducible between those runs on shorter and preferably known time-scales. -- Szilárd On Tue, Oct 4, 2016 at 5:30 PM, Kelechi Okoroafor wrote: > Thank you very much, Mark. > > You are absolutely right in that I do not have enough statistical data > to definitely say that the structures ought to form at 200ns. > > However, I have repeated the experimental set in total 4 times on > Stampede and twice on my school's cluster. > In all cases, without exception, all the systems have formed the > expected self-assembled structures within the 200ns time frame. > > But on my desktop, the structure did not happen even after 523ns. > > Do you suggest that I keep running the simulation for observation? > > > > -- > Best wishes, > Kelechi > -- > Gromacs Users mailing list > > * Please search the archive at > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists > > * For (un)subscribe requests visit > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a > mail to gmx-users-requ...@gromacs.org. -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
Thank you very much, Mark. You are absolutely right in that I do not have enough statistical data to definitely say that the structures ought to form at 200ns. However, I have repeated the experimental set in total 4 times on Stampede and twice on my school's cluster. In all cases, without exception, all the systems have formed the expected self-assembled structures within the 200ns time frame. But on my desktop, the structure did not happen even after 523ns. Do you suggest that I keep running the simulation for observation? -- Best wishes, Kelechi -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
On 10/4/16 11:12 AM, Kelechi Okoroafor wrote: Thank you very much, Justin, for your response. It is highly appreciated. When I built Gromacs on both desktops, I recall doing the regression tests, though I do not recall the results of the test, my bad. Do you mean I should re-run the regression tests? Yes. Or are there other tests I can run? In the absence of any demonstration that there's actually something wrong with the software in the tests, I concur in full with Mark's reply. You're seeing variability in your results. This is par for the course in MD simulations. -Justin -- == Justin A. Lemkul, Ph.D. Ruth L. Kirschstein NRSA Postdoctoral Fellow Department of Pharmaceutical Sciences School of Pharmacy Health Sciences Facility II, Room 629 University of Maryland, Baltimore 20 Penn St. Baltimore, MD 21201 jalem...@outerbanks.umaryland.edu | (410) 706-7441 http://mackerell.umaryland.edu/~jalemkul == -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
Thank you very much, Justin, for your response. It is highly appreciated. When I built Gromacs on both desktops, I recall doing the regression tests, though I do not recall the results of the test, my bad. Do you mean I should re-run the regression tests? Or are there other tests I can run? -- Best wishes, Kelechi -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
Hi, On Tue, Oct 4, 2016 at 4:17 PM Kelechi Okoroafor wrote: > Thank you very much, Mark, for your prompt response; it is highly > appreciated. > My sincere apologies for the miscommunication in my initial post. > > I meant to say that in performing several repetitions of simulations, > after approximately 200ns of total simulation time on (Stampede and my > school's) cluster, I observed some structures formed that have been > corroborated by other MD simulations as well as experiments. > > However, these same simulations were repeated several times on both my > personal desktop computers and I did not observe the expected > structures after 200ns. > It doesn't follow that because you have observed it in under 200ns that it will always happen in that time, or even that that is a characteristic time. You have to observe the process quite a few times before you can express such times with statistical confidence :-) (In isolation, that the sun was up for about 12 hours today doesn't tell me much about how long tomorrow will be...) Experiment is not necessarily oracular, either... there can be other things going on that don't show up with the method being used there. So the issue is not that the simulations are not exactly the same; it > is that the simulations are not giving physically reasonable results. > Of course one cannot rule out that there is a meaningful problem with the code such that your run(s) with 5.0.6 and 5.1.2 are affected by a bug (which could even be because 5.0.6 had a bug that was fixed in 5.1.2...). You should check out the online release notes to see if there's things of concern for you. But so far, I'd suggest that your problem is merely insufficient sampling. A physically reasonable result requires that you have observed reality correctly in an experiment, have used a model capable of reproducing that reality, have described the system correctly within that model, have used software that correctly samples according to that description, and have done enough sampling. Life can be tricky, here :-) Mark > -- > Best wishes, > Kelechi > -- > Gromacs Users mailing list > > * Please search the archive at > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before > posting! > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists > > * For (un)subscribe requests visit > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or > send a mail to gmx-users-requ...@gromacs.org. > -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
Thank you very much, Mark, for your prompt response; it is highly appreciated. My sincere apologies for the miscommunication in my initial post. I meant to say that in performing several repetitions of simulations, after approximately 200ns of total simulation time on (Stampede and my school's) cluster, I observed some structures formed that have been corroborated by other MD simulations as well as experiments. However, these same simulations were repeated several times on both my personal desktop computers and I did not observe the expected structures after 200ns. So the issue is not that the simulations are not exactly the same; it is that the simulations are not giving physically reasonable results. -- Best wishes, Kelechi -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
On 10/4/16 10:16 AM, Kelechi Okoroafor wrote: Thank you very much, Mark, for your prompt response; it is highly appreciated. My sincere apologies for the miscommunication in my initial post. I meant to say that in performing several repetitions of simulations, after approximately 200ns of total simulation time on (Stampede and my school's) cluster, I observed some structures formed that have been corroborated by other MD simulations as well as experiments. However, these same simulations were repeated several times on both my personal desktop computers and I did not observe the expected structures after 200ns. This sounds like normal sampling variability. It's never a given that independent simulations will do the same thing in the same amount of time. So the issue is not that the simulations are not exactly the same; it is that the simulations are not giving physically reasonable results. Does your desktop implementation of GROMACS pass all tests? If it does, then there's nothing wrong with the results. -Justin -- == Justin A. Lemkul, Ph.D. Ruth L. Kirschstein NRSA Postdoctoral Fellow Department of Pharmaceutical Sciences School of Pharmacy Health Sciences Facility II, Room 629 University of Maryland, Baltimore 20 Penn St. Baltimore, MD 21201 jalem...@outerbanks.umaryland.edu | (410) 706-7441 http://mackerell.umaryland.edu/~jalemkul == -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
Re: [gmx-users] Different Results with Different Platforms
Hi, This sounds totally normal. Individual MD trajectories are only reproducible if you take great pains to make them so, and that is typically not a relevant objective (and definitely not implemented for GROMACS on GPUs). If you're trying to do an ergodic sampling, then that's the thing you want to reproduce. But if a single trajectory hasn't done such sampling, wanting to reproduce it doesn't address the real issue that that trajectory was an observation of low significance. You could e.g. run again on your old cluster with a different set of initial velocities, and I expect you will observe divergence. You see to have gotten "lucky" to miss that on the first time you repeated the work. See also http://www.gromacs.org/Documentation/Terminology/Reproducibility Mark On Tue, Oct 4, 2016 at 2:08 PM Kelechi Okoroafor wrote: > Hello, All. > > I had previously run the same coarse-grained MD simulations using the > Martini force field with PME electrostatics on TACC's Stampede GPU cluster > and my school's CPU cluster and gotten the same results. > > I recently built two desktop computers and the results from both my > computers match each other but are completely different from the results > from TACC's Stampede and my school's cluster. > > I have checked through and I have listed the differences between the > architecture of the different platforms I have run the simulations on > below. > > Please has anyone else experienced this? > Does anyone know what may be the issue with my build of Gromacs? > > Stampede TACC: > >- Gromacs Version = 5.0.6 >- FFT library = Intel MKL >- SIMD instruction set = AVX_256 >- CPU = Intel(R) Xeon(R) CPU E5-2680 0 >- > >GPU = NVIDIA Tesla K20m > >- > >CUDA driver = 7.0 > >- > >CUDA runtime = 7.0 > > > > > School's CPU Cluster: > >- Gromacs Version = 5.1.2 >- FFT library = fftw-3.3.4-sse2 >- SIMD instruction set = SSE4.1 >- CPU = Intel(R) Xeon(R) CPU E5-2699 v3 > > > > My Computer 1 > >- Gromacs Version = 5.1.2 >- FFT library = fftw-3.3.4-sse2-avx >- SIMD instruction set = AVX2_256 >- CPU = Intel(R) Core(TM) i7-6800K >- GPU = NVIDIA GeForce GTX 1080 >- CUDA driver = 8.0 >- CUDA runtime = 8.0 > > > My Computer 2 > >- Gromacs Version = 5.1.2 >- FFT library = fftw-3.3.4-sse2-avx >- SIMD instruction set = AVX2_256 >- CPU = Intel(R) Core(TM) i5-6600K >- GPU = NVIDIA GeForce GTX 1070 >- CUDA driver = 8.0 >- CUDA runtime = 8.0 > > > Best wishes, > Kae > -- > Gromacs Users mailing list > > * Please search the archive at > http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before > posting! > > * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists > > * For (un)subscribe requests visit > https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or > send a mail to gmx-users-requ...@gromacs.org. > -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.
[gmx-users] Different Results with Different Platforms
Hello, All. I had previously run the same coarse-grained MD simulations using the Martini force field with PME electrostatics on TACC's Stampede GPU cluster and my school's CPU cluster and gotten the same results. I recently built two desktop computers and the results from both my computers match each other but are completely different from the results from TACC's Stampede and my school's cluster. I have checked through and I have listed the differences between the architecture of the different platforms I have run the simulations on below. Please has anyone else experienced this? Does anyone know what may be the issue with my build of Gromacs? Stampede TACC: - Gromacs Version = 5.0.6 - FFT library = Intel MKL - SIMD instruction set = AVX_256 - CPU = Intel(R) Xeon(R) CPU E5-2680 0 - GPU = NVIDIA Tesla K20m - CUDA driver = 7.0 - CUDA runtime = 7.0 School's CPU Cluster: - Gromacs Version = 5.1.2 - FFT library = fftw-3.3.4-sse2 - SIMD instruction set = SSE4.1 - CPU = Intel(R) Xeon(R) CPU E5-2699 v3 My Computer 1 - Gromacs Version = 5.1.2 - FFT library = fftw-3.3.4-sse2-avx - SIMD instruction set = AVX2_256 - CPU = Intel(R) Core(TM) i7-6800K - GPU = NVIDIA GeForce GTX 1080 - CUDA driver = 8.0 - CUDA runtime = 8.0 My Computer 2 - Gromacs Version = 5.1.2 - FFT library = fftw-3.3.4-sse2-avx - SIMD instruction set = AVX2_256 - CPU = Intel(R) Core(TM) i5-6600K - GPU = NVIDIA GeForce GTX 1070 - CUDA driver = 8.0 - CUDA runtime = 8.0 Best wishes, Kae -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.