Re: [gmx-users] Pme on gpu performance

2011-12-13 Thread Szilárd Páll
Dear Rafael, It really depends on the processors, GPUs, as well as whether you are interested in performance or performance/buck. Also note, that with GPUs you'll need considerably more atoms/core. First of all, the E7-s are 8-core processors, so four of them is only 32 (and from what I can see i

Re: [gmx-users] Pme on gpu performance

2011-11-29 Thread rafapa
Dear Szilárd, now 64 cores machines are available, for instance, using Xeons E7-8837. Do you have any idea about which combination will give more performance? I mean, 64 CPU cores or 16-32 CPU cores + 4 C2075 or similar. The limit of 4 GPUs is imposed by the MB of the machine I have in mind.

Re: [gmx-users] Pme on gpu performance

2011-11-29 Thread Szilárd Páll
Hi Andrzej, > One more question: will a ratio of gpu/cpu units and cores be of importance > in next gromacs releases ? at the moment the code uses one core per gpu > unit, wright ? When the code is gpu parallel how can this change ? Yes, it will. We use both CPU & GPU and load balance between the

Re: [gmx-users] Pme on gpu performance

2011-11-29 Thread Szilárd Páll
Thanks for the info. -- Szilárd On Tue, Nov 29, 2011 at 11:50 AM, Andrzej Rzepiela wrote: > Hey, > > Thank you for the info. The data that I obtained for comparison was > performed  with GTX580, 4 fs timestep and heavy hydrogen atoms instead of > constraints, as you suspected.  For dhfr with PME

Re: [gmx-users] Pme on gpu performance

2011-11-29 Thread Szilárd Páll
Hi, On Mon, Nov 28, 2011 at 8:55 PM, Jones de Andrade wrote: > Hi. > > I'm another one waiting for the OpenCL version here. > >> 4.6 will support only CUDA. >> >> OpenCL GPU implementations have yet match CUDA; on CPU it would hardly >> match the performance of the hand-tuned SSE kernels, moreove

[gmx-users] Pme on gpu performance

2011-11-29 Thread Andrzej Rzepiela
Hey One more question: will a ratio of gpu/cpu units and cores be of importance in next gromacs releases ? at the moment the code uses one core per gpu unit, wright ? When the code is gpu parallel how can this change ? Thank you Andrzej -- gmx-users mailing listgmx-users@gromacs.o

[gmx-users] Pme on gpu performance

2011-11-29 Thread Andrzej Rzepiela
Hey, Thank you for the info. The data that I obtained for comparison was performed with GTX580, 4 fs timestep and heavy hydrogen atoms instead of constraints, as you suspected. For dhfr with PME they were able to get 100 ns a day. I tested the same system on Tesla M2090 with 2fs timeste

Re: [gmx-users] Pme on gpu performance

2011-11-28 Thread Jones de Andrade
Hi. I'm another one waiting for the OpenCL version here. 4.6 will support only CUDA. > > OpenCL GPU implementations have yet match CUDA; on CPU it would hardly > match the performance of the hand-tuned SSE kernels, moreover, CPU & > GPU algorithms are not the same. > If you allow me, the (second

Re: [gmx-users] Pme on gpu performance

2011-11-28 Thread Matt Larson
Szilárd, Version 4.6, GPU without OpenMM sounds fantastic. I am very excited! I was having trouble getting simulations to work equivalently well on the GPU as CPU because of having to make different choices with cutoffs and mainly not being able to use the V-rescale thermostat. -Matt On Sun,

Re: [gmx-users] Pme on gpu performance

2011-11-27 Thread Szilárd Páll
Hi, 4.6 will support only CUDA. OpenCL GPU implementations have yet match CUDA; on CPU it would hardly match the performance of the hand-tuned SSE kernels, moreover, CPU & GPU algorithms are not the same. -- Szilárd On Sun, Nov 27, 2011 at 11:25 PM, Alexey Shvetsov wrote: > Hi! > > Will it u

Re: [gmx-users] Pme on gpu performance

2011-11-27 Thread Alexey Shvetsov
Hi! Will it use CUDA or OpenCL? Second one will be more common since it will work with wider range of platfroms (cpu, gpu, fcpga) Szilárd Páll писал 27.11.2011 23:50: Native acceleration = not relying on external libraries. ;) -- Szilárd On Sun, Nov 27, 2011 at 9:33 PM, Peter C. Lai wrot

Re: [gmx-users] Pme on gpu performance

2011-11-27 Thread Peter C. Lai
Oh cool. Looking forward to it then! On 2011-11-27 02:50:43PM -0600, Szilárd Páll wrote: > Native acceleration = not relying on external libraries. ;) > > -- > Szilárd > > > > On Sun, Nov 27, 2011 at 9:33 PM, Peter C. Lai wrote: > > Random hijack: > > > > Will the CUDA acceleration support CM

Re: [gmx-users] Pme on gpu performance

2011-11-27 Thread Szilárd Páll
Native acceleration = not relying on external libraries. ;) -- Szilárd On Sun, Nov 27, 2011 at 9:33 PM, Peter C. Lai wrote: > Random hijack: > > Will the CUDA acceleration support CMAP or will it still be limited to > the limitations of OpenMM that it currently has? > > On 2011-11-27 12:10:47P

Re: [gmx-users] Pme on gpu performance

2011-11-27 Thread Peter C. Lai
Random hijack: Will the CUDA acceleration support CMAP or will it still be limited to the limitations of OpenMM that it currently has? On 2011-11-27 12:10:47PM -0600, Szilárd Páll wrote: > Hi Andrzej, > > GROMACS 4.6 is work in progress, it will have native CUDA acceleration > with multi-GPU sup

Re: [gmx-users] Pme on gpu performance

2011-11-27 Thread Szilárd Páll
Hi Andrzej, GROMACS 4.6 is work in progress, it will have native CUDA acceleration with multi-GPU support along a few other improvements. You can expect a speedup in the ballpark of 3x. We will soon have the code available for testing. I'm a little skeptical about the 5x of ACEMD. What setting di

[gmx-users] Pme on gpu performance

2011-11-24 Thread Andrzej Rzepiela
Dears, I would like to ask what performance improvement is expected for future releases (e.g. 4.6) for PME calculations on gpu cards. Last week I sent benchmark results for dhfr to the list, which show no performance gain when the newest versions of tesla cards vs intel xeon R are used (m