Re: [gmx-users] Re: GPU version of Gromacs

2013-08-27 Thread Mark Abraham
No, that is not implemented. Roughly speaking, the CPU does the PME+bondeds while the GPU does the short-ranged interactions. Moving PME would only be worthwhile if there was a very weak CPU and it was cheap to do any necessary communication between GPUs. That day is not yet here :-) Mark On Tue,

[gmx-users] Re: GPU version of Gromacs

2013-08-27 Thread grita
Hi, Is it possible to calc the PME on the CUDA card? If it is possible, how can one do it? Best, grita -- View this message in context: http://gromacs.5086.x6.nabble.com/GPU-version-of-Gromacs-tp5010581p5010758.html Sent from the GROMACS Users Forum mailing list archive at Nabble.com. -- gm

Re: [gmx-users] Re: GPU version of Gromacs

2013-08-20 Thread Justin Lemkul
On 8/20/13 5:52 AM, grita wrote: Dear Dwey & Justin, I've not tried it yet and I've asked because in the 4.5.x GPU version of Gromacs not all features were supported. In my simulations I'm using the SD integrator and the Pull-Code. Since both are supported in the current GPU version, I will

[gmx-users] Re: GPU version of Gromacs

2013-08-20 Thread grita
Dear Dwey & Justin, I've not tried it yet and I've asked because in the 4.5.x GPU version of Gromacs not all features were supported. In my simulations I'm using the SD integrator and the Pull-Code. Since both are supported in the current GPU version, I will try it on a computer with a CUDA grap

[gmx-users] Re: GPU version of Gromacs

2013-08-19 Thread Dwey Kauffman
Hi Grita, Yes it is. You need to re-compile a GPU version of Gromacs from source codes. You also need to use Verlet cutoff-scheme. That is, place a new line like cutoff-scheme = Verlet in your mdp file. Finally, run the GPU version of mdrun, adding a parameter -gpu_id 0if you have o