Hi Christopher,
I am not an expert, but I can tell you this:
In order to get a speed up when running parallel GPUs, they need to be on
the same PCI-E bus, which enables P2P communication.
See the pmemd.cuda.mpi documentation here:
http://ambermd.org/gpus/
-----Original Message-----
From: Neale, Christopher Andrew [mailto:cneale.lanl.gov]
Sent: Monday, December 19, 2016 7:31 PM
To: amber.ambermd.org
Subject: [AMBER] Amber16 on GPUs and speed differences between
CUDA_VISIBLE_DEVICES=0, 1 or 0, 2
I learned how to get a bit more information, which perhaps will help:
$ nvidia-smi topo -m
GPU0 GPU1 GPU2 GPU3 CPU Affinity
GPU0 X PHB SOC SOC 0-5,12-17
GPU1 PHB X SOC SOC 0-5,12-17
GPU2 SOC SOC X PHB 6-11,18-23
GPU3 SOC SOC PHB X 6-11,18-23
Legend:
X = Self
SOC = Path traverses a socket-level link (e.g. QPI)
PHB = Path traverses a PCIe host bridge
PXB = Path traverses multiple PCIe internal switches
PIX = Path traverses a PCIe internal switch
Thank you,
Chris.
_______________________________________________
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 Mon Dec 19 2016 - 14:00:02 PST
