Thank you sir for your kind response .I have forgotten to ask how many should I take ntask per node for best ??Thanks and Regards,Saikat
On Friday 18 August 2017, 7:43:49 PM IST, Daniel Roe <daniel.r.roe.gmail.com> wrote:
The GPU implementation of AMBER runs almost entirely on the GPU (one
of the reasons it's so fast), so you only need one CPU task per GPU.
Just use:
pmemd.cuda <options>
You can have separate runs on each GPU at the same time by making use
of CUDA_VISIBLE_DEVICES:
cd <directory1>
export CUDA_VISIBLE_DEVICES=0
pmemd.cuda <options> &
cd <directory2>
export CUDA_VISIBLE_DEVICES=1
pmemd.cuda <options> &
-Dan
On Fri, Aug 18, 2017 at 9:50 AM, Saikat Pal <saikatpaliitg.yahoo.com> wrote:
> Thank you Sir for your kind response. If I will run amber in single gpu how many cpu per task should I take for best run (means less time consuming)??I have 24 cpus.
> Thanks and Regards,Saikat
>
>
> On Friday 18 August 2017, 6:39:42 PM IST, Daniel Roe <daniel.r.roe.gmail.com> wrote:
>
> Your motherboard needs to support it and the GPUs need to be arranged
> so that they are on the same PCI bus. See this post for a good
> description of P2P:
>
> http://exxactcorp.com/blog/exploring-the-complexities-of-pcie-connectivity-and-peer-to-peer-communication/
>
> -Dan
>
> On Fri, Aug 18, 2017 at 8:35 AM, Saikat Pal <saikatpaliitg.yahoo.com> wrote:
>> Thank you Sir for your kind response.In mdout file :--------------- GPU PEER TO PEER INFO -----------------
>> |
>> | Peer to Peer support: DISABLED
>> |
>> | (Selected GPUs cannot communicate over P2P)
>> |
>> |--------------------------------------------------------
>>
>>
>> How can I enable Peer to Peer support??
>> Thanks and Regards,Saikat
>> On Friday 18 August 2017, 5:45:37 PM IST, Daniel Roe <daniel.r.roe.gmail.com> wrote:
>>
>> Hi,
>>
>> On Fri, Aug 18, 2017 at 7:33 AM, Saikat Pal <saikatpaliitg.yahoo.com> wrote:
>>> Dear all,How many GPU card is working ??I think it is 10 but not sure.Please help me.Thanks Saikat
>>
>> It's 2. According to your output CUDA_VISIBLE_DEVICES is set to "0,1",
>> and there are 2 CUDA capable devices detected. Right now 3 tasks are
>> running on one GPU and 3 on the other, which is almost certainly not
>> what you want. You may get some speedup running across 2 GPUs if
>> peer-to-peer support is enabled, but you should benchmark. More useful
>> info on running on multiple GPUs can be found here:
>> http://ambermd.org/gpus/index.htm#Running
>>
>> -Dan
>>
>>>
>>>
>>> ------------------- GPU DEVICE INFO --------------------
>>> |
>>> | Task ID: 0
>>> | CUDA_VISIBLE_DEVICES: 0,1
>>> | CUDA Capable Devices Detected: 2
>>> | CUDA Device ID in use: 0
>>> | CUDA Device Name: Tesla K40c
>>> | CUDA Device Global Mem Size: 11519 MB
>>> | CUDA Device Num Multiprocessors: 15
>>> | CUDA Device Core Freq: 0.75 GHz
>>> |
>>> |
>>> | Task ID: 1
>>> | CUDA_VISIBLE_DEVICES: 0,1
>>> | CUDA Capable Devices Detected: 2
>>> | CUDA Device ID in use: 0
>>> | CUDA Device Name: Tesla K40c
>>> | CUDA Device Global Mem Size: 11519 MB
>>> | CUDA Device Num Multiprocessors: 15
>>> | CUDA Device Core Freq: 0.75 GHz
>>> |
>>> |
>>> | Task ID: 2
>>> | CUDA_VISIBLE_DEVICES: 0,1
>>> | CUDA Capable Devices Detected: 2
>>> | CUDA Device ID in use: 1
>>> | CUDA Device Name: Tesla K40c
>>> | CUDA Device Global Mem Size: 11519 MB
>>> | CUDA Device Num Multiprocessors: 15
>>> | CUDA Device Core Freq: 0.75 GHz
>>> |
>>> |
>>> | Task ID: 3
>>> | CUDA_VISIBLE_DEVICES: 0,1
>>> | CUDA Capable Devices Detected: 2
>>> | CUDA Device ID in use: 1
>>> | CUDA Device Name: Tesla K40c
>>> | CUDA Device Global Mem Size: 11519 MB
>>> | CUDA Device Num Multiprocessors: 15
>>> | CUDA Device Core Freq: 0.75 GHz
>>> |
>>> |
>>> | Task ID: 4
>>> | CUDA_VISIBLE_DEVICES: 0,1
>>> | CUDA Capable Devices Detected: 2
>>> | CUDA Device ID in use: 0
>>> | CUDA Device Name: Tesla K40c
>>> | CUDA Device Global Mem Size: 11519 MB
>>> | CUDA Device Num Multiprocessors: 15
>>> | CUDA Device Core Freq: 0.75 GHz
>>> |
>>> |
>>> | Task ID: 5
>>> | CUDA_VISIBLE_DEVICES: 0,1
>>> | CUDA Capable Devices Detected: 2
>>> | CUDA Device ID in use: 0
>>> | CUDA Device Name: Tesla K40c
>>> | CUDA Device Global Mem Size: 11519 MB
>>> | CUDA Device Num Multiprocessors: 15
>>> _______________________________________________
>>> AMBER mailing list
>>> AMBER.ambermd.org
>>> http://lists.ambermd.org/mailman/listinfo/amber
>>
>>
>>
>> --
>> -------------------------
>> Daniel R. Roe
>> Laboratory of Computational Biology
>> National Institutes of Health, NHLBI
>> 5635 Fishers Ln, Rm T900
>> Rockville MD, 20852
>> https://www.lobos.nih.gov/lcb
>>
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>> _______________________________________________
>> AMBER mailing list
>> AMBER.ambermd.org
>> http://lists.ambermd.org/mailman/listinfo/amber
>
>
>
> --
> -------------------------
> Daniel R. Roe
> Laboratory of Computational Biology
> National Institutes of Health, NHLBI
> 5635 Fishers Ln, Rm T900
> Rockville MD, 20852
> https://www.lobos.nih.gov/lcb
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
--
-------------------------
Daniel R. Roe
Laboratory of Computational Biology
National Institutes of Health, NHLBI
5635 Fishers Ln, Rm T900
Rockville MD, 20852
https://www.lobos.nih.gov/lcb
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Aug 18 2017 - 08:00:04 PDT
