[SCore-users] Score 4.2.1 on new Athlon systems.
Shinji Sumimoto
s-sumi at flab.fujitsu.co.jp
Thu Mar 7 14:49:54 JST 2002
Hi.
From: Nick Birkett <nrcb at streamline-computing.com>
Subject: [SCore-users] Score 4.2.1 on new Athlon systems.
Date: Thu, 7 Mar 2002 01:16:58 +0000
Message-ID: <0203070116580I.19905 at pecan.comlab.ox.ac.uk>
nrcb> Hi , we have been running both Score and Myricom GM.
nrcb>
nrcb> We now have some new systems based on Tyan Tiger MPX multi-processor boards
nrcb> (1.6 GHz Athlon with DDR memory).
nrcb>
nrcb> Pallas benchmark PMB-MPI1 compiled using -O4 -fomit-frame-pointer
nrcb> -funroll-all-loops and with -DnoCHECK gives very poor bandwidth result for
nrcb> Score Myrinet2k, whereas GM gives good one.
nrcb>
nrcb> Score 4.2.1 gives 90Mbytes/sec point to point over Myrinet2k.
nrcb>
nrcb> GM/MPICH (GM 1.5, MPICH 1.2..8 and 1.2.7.. ) gives twice this bandwidth
nrcb> over Myrinet.
nrcb>
nrcb> GM is using kernel 2.4.18 and Score uses 2.4.10smp (the one that comes with
nrcb> Score 4.2.1 binary release).
nrcb>
nrcb> We have compared GM and Score on other Athlon systems and do not have this
nrcb> big difference (usually bandwidth is same within a few %).
MPICH/GM uses zero-copy at default, but MPICH/SCore does not use
zero-copy at default.
Could you test the program using mpi_zerocopy=on?
Moreover, there is some hardware problems on AMD MPX, especially 66MHz
PCI bus. The problem seems that PCI DMA performance is degrated to
100MB/s with CPU copy on same memory area.
I am now investigating the problem.
Shinji.
------
Shinji Sumimoto, Fujitsu Labs
More information about the SCore-users
mailing list