AW: GPU accelerated MD

From: Norman Geist (norman.geist_at_uni-greifswald.de)
Date: Fri Oct 23 2015 - 05:00:05 CDT

• Next message: Fidan Sumbul: "Re: Error in Metadynamics Simulation"
• Previous message: Nikhil Maroli: "GPU accelerated MD"
• In reply to: Nikhil Maroli: "GPU accelerated MD"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]

What about just trying? From my experience you will get best performance (ns/day) using all available 16 cpus + the gpu. But you should disable hyperthreading if you don't know how to bind namd to the "real" cpu cores.

Norman Geist

-----Ursprüngliche Nachricht-----
Von: owner-namd-l_at_ks.uiuc.edu [mailto:owner-namd-l_at_ks.uiuc.edu] Im Auftrag von Nikhil Maroli
Gesendet: Freitag, 23. Oktober 2015 11:12
An: namd-l_at_ks.uiuc.edu; Vmd L <vmd-l_at_ks.uiuc.edu>
Betreff: namd-l: GPU accelerated MD

Dear all,
i have 2 x Intel® E5-2650V2 Processor (8Core,2.6GHz,20M L3 Cache) (32
Threads) and One Tesla K20 GPU
which option will be faster for NAMD

1. CPU alone
2.CPU+GPU
3.GPU+CPU(by assigning lower threads)

i read somewhere if we give GPU+CPU it will slow down in case of particular
GPU ,so which is best option for me here? ()

Thanks in advance

PS:My benchmark in YASARA shows 29 out of 32 thread without GPU Runs faster=

• Next message: Fidan Sumbul: "Re: Error in Metadynamics Simulation"
• Previous message: Nikhil Maroli: "GPU accelerated MD"
• In reply to: Nikhil Maroli: "GPU accelerated MD"
• Messages sorted by: [ date ] [ thread ] [ subject ] [ author ] [ attachment ]

This archive was generated by hypermail 2.1.6 : Thu Dec 31 2015 - 23:22:10 CST