AW: question about GPU

From: Norman Geist (norman.geist_at_uni-greifswald.de)
Date: Wed Feb 08 2017 - 02:24:53 CST

No, you dont need to compile yourself. Using a pre-compiled binary of NAMD should work just fine. As I already said: You don’t have to change anything to the configuration between CPU and GPU runs normally. Therefore it’s easy to compare the performance of a CPU only NAMD to a GPU enabled NAMD.

 

Norman Geist

 

Von: parinaz bashirbanaem [mailto:parinaz.bashirbanaem_at_gmail.com]
Gesendet: Mittwoch, 8. Februar 2017 08:37
An: Norman Geist <norman.geist_at_uni-greifswald.de>
Betreff: Re: namd-l: question about GPU

 

Dear Professor Geist,

 

Thank you very much for precious guidance, I would like to know this issue that I only copy and paste NAMD-2.12 Win64-CUDA (NVIDIA CUDA acceleration) in drive C, then I specified it's path in Command Prompt and I didn't compile it. Is necessary compile it for using of GPU? I really appreciate your concern.

Best Regards.

 

On Tue, Feb 7, 2017 at 1:15 AM, Norman Geist <norman.geist_at_uni-greifswald.de <mailto:norman.geist_at_uni-greifswald.de> > wrote:

As I already pointed out, a GPU enabled version of NAMD can’t run without a GPU. So if you have in fact a GPU enabled version, there’s not much need to care if the GPU is actually used. You should see lines in the NAMD output mentioning the GPU model.

 

There are only a few things, were users can or should manually tweak a GPU powered simulation. These are:

 

1. Choose values above 50 for “dcdfreq” and “outputenergies”, usually default is about 1000.

2. The optimal CPU-GPU ratio. The best performance can be evaluated from the lines in the NAMD output starting with ”Benchmark:“. The more ns/day the better. Usually up to 8 CPU cores per GPU are ok.

3. One can pass the argument +ignoresharing to namd2 and see if it improves or harms performance. This is different and depends on the composition of the molecular system.

4. One can set “twoawayx yes” in the NAMD config in order to create more distributable subunit of the simulation cell, which can improve GPU utilization. If it helps, try also twoawayy.

 

The easiest way to figure out if your GPU acceleration is appropriate, is by comparing it against a CPU only version.

 

Good luck

 

Norman Geist

 

Von: parinaz bashirbanaem [mailto:parinaz.bashirbanaem_at_gmail.com <mailto:parinaz.bashirbanaem_at_gmail.com> ]
Gesendet: Dienstag, 7. Februar 2017 09:50
An: Norman Geist <norman.geist_at_uni-greifswald.de <mailto:norman.geist_at_uni-greifswald.de> >
Betreff: Re: namd-l: question about GPU

 

Dear Professor Geist,

 

 

Thank you very much for quick reply, But I use" NAMD 2.10, Win64-CUDA" and all core of CPU with GPU, what is your suggestion about command?

 

Best Regards.

 

On Mon, Feb 6, 2017 at 7:47 AM, Norman Geist <norman.geist_at_uni-greifswald.de <mailto:norman.geist_at_uni-greifswald.de> > wrote:

You have to download (or build) a GPU accelerated version of NAMD. E.g.:

 

 <http://www.ks.uiuc.edu/Development/Download/download.cgi?UserID=&AccessCode=&ArchiveID=1499> Linux-x86_64-multicore-CUDA (NVIDIA CUDA acceleration)

 

The binary mentioned above is NAMD-2.12 single node. A GPU version of NAMD can’t run without a GPU. Feel free to answer for further details

 

Norman Geist

 

Von: owner-namd-l_at_ks.uiuc.edu <mailto:owner-namd-l_at_ks.uiuc.edu> [mailto:owner-namd-l_at_ks.uiuc.edu <mailto:owner-namd-l_at_ks.uiuc.edu> ] Im Auftrag von parinaz bashirbanaem
Gesendet: Montag, 6. Februar 2017 16:01
An: namd-l_at_ks.uiuc.edu <mailto:namd-l_at_ks.uiuc.edu>
Betreff: namd-l: question about GPU

 

Hi,

 

I want to use the GPU accelerated version of NAMD 2.10, Which is the correct command to run the NAMD?

 

 We are currently using this command :

 

./namd2 +idlepoll +p10 A.conf > A.log

 

But there isn't any different than using CPU, and does not accelerate by GPU. please help me in this way.

 

With Respect.

 

 

 

This archive was generated by hypermail 2.1.6 : Mon Dec 31 2018 - 23:20:05 CST