Re: [AMBER] Amber14 using triple GTX 1080

From: Elka Firmanda <firmand46.gmail.com>
Date: Wed, 5 Oct 2016 20:04:25 +0700

Have you ever tried CUDA 8.0 ? I think you have to use CUDA 8.0 for Pascal GPU

*maaf pak, setau saya Pascal harus pakai CUDA 8.0, kalau kami di IPB masih menggunakan Maxwell GTX 980, jadi masih bisa pakai CUDA 7.5*

-----Original Message-----
From: "Yohanes Gultom" <yohanes.gultom.gmail.com>
Sent: ‎05/‎10/‎2016 18:59
To: "amber.ambermd.org" <amber.ambermd.org>
Subject: [AMBER] Amber14 using triple GTX 1080

Hi all,

​I have just started using Amber14 on GPU and have managed to run some
experiments on several GPU (GTX 780, 970, 980, Titan X). Now I'm currently
trying to run same experiments on our new triple GTX 1080.

The problem is I can't run any experiment on GTX 1080 (I can run exact same
experiments on other machines with different GPU). I kept getting this
error whenever I run pmemd.cuda.MPI (with any number of GPU):

"cudaMemcpyToSymbol: SetSim copy to cSim failed invalid device symbol"

Could anyone help me to understand the problem?
Is it because GTX 1080 supported by Amber14? Because I only find mentions
on GTX 1080 on Amber16 pages.

Some additional info:
* The triple GTX 1080 running on Debian 8 with CUDA 7.5 and Nvidia
Driver 367.44 (same config for other machines with different GPU)
* I can run most CUDA samples on the GTX 1080 machine. The only anomaly is
a message "MapSMtoCores for SM 6.1 is undefined. Default to use 128
Cores/SM" when running deviceQuery


​Regards
,

Yohanes Gultom
​Faculty of Computer Science​

​University of Indonesia​
_______________________________________________
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 Wed Oct 05 2016 - 06:30:03 PDT
Custom Search