Re: Improve performance

From: David Hardy (dhardy_at_ks.uiuc.edu)
Date: Mon Feb 15 2021 - 18:25:41 CST

Hi Ming,

I see that AWS p2 uses K80 GPUs, which are now over 6 years old. The GPU-resident integrator in NAMD 3 has been optimized for V100 or later GPUs. It would be interesting for us to benchmark the most recent codes on K80, but I don’t know that anyone has done that yet.

As to your issue with running NAMD 2.14 on a 4-GPU machine versus an 8-GPU machine, NAMD 2.14 on modern GPUs is CPU bound, so it is possible that you saw no improvement in performance if you didn’t also increase the CPU core count per GPU. In fact, this is our motivation for developing a GPU-resident version of NAMD.

Best regards,
Dave

--
David J. Hardy, Ph.D.
Beckman Institute
University of Illinois at Urbana-Champaign
405 N. Mathews Ave., Urbana, IL 61801
dhardy_at_ks.uiuc.edu, http://www.ks.uiuc.edu/~dhardy/
> On Feb 15, 2021, at 7:15 AM, Ming Sun <msun_at_beamtx.com> wrote:
> 
> Hi All,
>  
> A follow-up of previous discussion on the NAMD performed performances. I’m wondering if anyone has done benchmark on NAMD3 on AWS p2 instances or another GPU workstation? What would be an optimal / okay number of CPU per GPU ?  I am using NAMD2.14 these days. I did not see improvements switching from 4-GPU machine to an 8-GPU machine while the number of CPU does not increase, at least at my hand
>  
> Any thoughts on NAMD3/ 2.14 are greatly appreciated.
>  
> Thanks,
> Ming  
>  
>  
> From: "Vermaas, Josh" <vermaasj_at_msu.edu <mailto:vermaasj_at_msu.edu>>
> Date: Thursday, January 28, 2021 at 7:45 PM
> To: "namd-l_at_ks.uiuc.edu <mailto:namd-l_at_ks.uiuc.edu>" <namd-l_at_ks.uiuc.edu <mailto:namd-l_at_ks.uiuc.edu>>, "Ropon-Palacios G." <biodano.geo_at_gmail.com <mailto:biodano.geo_at_gmail.com>>
> Subject: Re: namd-l: Improve performance
>  
> For the best NAMD performance on a GPU, use NAMD 3. This blog from NVIDIA has been my goto for explaining the differences to folks that haven’t used NAMD 3 yet. https://developer.nvidia.com/blog/delivering-up-to-9x-throughput-with-namd-v3-and-a100-gpu/ <https://urldefense.com/v3/__https://developer.nvidia.com/blog/delivering-up-to-9x-throughput-with-namd-v3-and-a100-gpu/__;!!DZ3fjg!tjYMkz_rYz0-mcygjKo9GeKdRm_0FZD8S4mIIgeJcpPHpSrKAvJF-wFdVk0GayNq_w$>
>  
> -Josh
>  
> From: <owner-namd-l_at_ks.uiuc.edu <mailto:owner-namd-l_at_ks.uiuc.edu>> on behalf of "Ropon-Palacios G." <biodano.geo_at_gmail.com <mailto:biodano.geo_at_gmail.com>>
> Reply-To: "namd-l_at_ks.uiuc.edu <mailto:namd-l_at_ks.uiuc.edu>" <namd-l_at_ks.uiuc.edu <mailto:namd-l_at_ks.uiuc.edu>>, "Ropon-Palacios G." <biodano.geo_at_gmail.com <mailto:biodano.geo_at_gmail.com>>
> Date: Thursday, January 28, 2021 at 7:18 PM
> To: "namd-l_at_ks.uiuc.edu <mailto:namd-l_at_ks.uiuc.edu>" <namd-l_at_ks.uiuc.edu <mailto:namd-l_at_ks.uiuc.edu>>
> Subject: namd-l: Improve performance
>  
> Dear namd users, 
>  
> How can i improve namd2.14 performance on GPU?
>  
> I`m using namd2.14 for run membrane protein simulation, but namd only use nor more the 50% of GPU, as can I make to improve this performance. 
>  
> For other hand, if during equilibration of protein-membrane is necessary keep water out membrane? 
>  
> Thanks, 
>  
> Best, 
>  
> Geo. 
>  
> 
> 
> Disclaimer
> 
> The information contained in this communication from the sender is confidential. It is intended solely for use by the recipient and others authorized to receive it. If you are not the recipient, you are hereby notified that any disclosure, copying, distribution or taking action in relation of the contents of this information is strictly prohibited and may be unlawful.
> 

This archive was generated by hypermail 2.1.6 : Fri Dec 31 2021 - 23:17:10 CST