Let me add two more things. Note that we *always* compare performance and acceleration to our super-tuned state-of-the-art CPU code, which I can confidently say that is among the fastest if not the fastest to date, and never to some slow (CPU) implementation. Therefore, while other codes might be able to get 10-20x with GPUs and many x-es even with pre-Fermi (CC <2.0) cards, for obvious reasons we simply can't.
Still you'll get high *absolute performance* regardless whether you can use CPU only or CPU+GPU, so this should be a decent deal, right? If anybody volunteers to add software-based floating point atomic operations for CC <2.0 and try running the current kernels on earlier hardware, I'm willing to give pointers, but I simply had no time to try it myself. This would enable using earlier cards, with a considerable performance penalty of emulated atomic ops (plus these cards are anyway slower), but it *might* be worth a try! Feel free to drop me a mail if anybody is interested. Cheers, -- Szilárd On Thu, Jan 24, 2013 at 3:48 PM, Szilárd Páll <szilard.p...@cbr.su.se>wrote: > On Thu, Jan 24, 2013 at 3:28 PM, Justin Lemkul <jalem...@vt.edu> wrote: > >> >> >> On 1/24/13 9:23 AM, James Starlight wrote: >> >>> oh that was simply solved by upgrading of G++ :) >>> >>> the only problem which remains is the missing of support of mu GPU :( >>> That time I've tried to start simulation on simply 2 cores CPU + geforce >>> 9500 >>> >>> From md run I've obtained >>>> >>> >>> NOTE: Error occurred during GPU detection: >>> CUDA driver version is insufficient for CUDA runtime version >>> Can not use GPU acceleration, will fall back to CPU kernels. >>> >>> ED sampling will be performed! >>> ED: Reading edi file new_a2a_pca_biased.edi >>> ED: Note: Reference and average structure are composed of the same atom >>> indices. >>> ED: Found 1 ED group. >>> Using 1 MPI thread >>> Using 2 OpenMP threads >>> >>> No GPUs detected >>> >>> >>> I've installed cuda-5.0 with driver >>> NVRM version: NVIDIA UNIX x86_64 Kernel Module 304.54 Sat Sep 29 >>> 00:05:49 PDT 2012 >>> >>> that system works fine with the recent nvidia GPU's but has no support >>> for GeForce 9500. >>> >>> >> GeForce 9500 cards have a compute capability of 1.1. The minimum >> required for Gromacs is 2.0 (noted in the installation instructions). > > > Exactly! Anything below CC 2.0 lacks certain features that make those > devices rather slow for our algorithms and therefore pretty much > impractical. > > Cheers, > -- > Szilárd > > > > > >> >> >> -Justin >> >> -- >> ==============================**========== >> >> Justin A. Lemkul, Ph.D. >> Research Scientist >> Department of Biochemistry >> Virginia Tech >> Blacksburg, VA >> jalemkul[at]vt.edu | (540) 231-9080 >> http://www.bevanlab.biochem.**vt.edu/Pages/Personal/justin<http://www.bevanlab.biochem.vt.edu/Pages/Personal/justin> >> >> ==============================**========== >> >> -- >> gmx-users mailing list gmx-users@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<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 to gmx-users-requ...@gromacs.org. >> * Can't post? Read >> http://www.gromacs.org/**Support/Mailing_Lists<http://www.gromacs.org/Support/Mailing_Lists> >> > > -- gmx-users mailing list gmx-users@gromacs.org 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 to gmx-users-requ...@gromacs.org. * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists
