Date: Sun, 11 Jan 2004 00:40:03 +0900 (JST) From: Nakata Maho <chat95@mbox.kyoto-inet.or.jp> To: freebsd-amd64@freebsd.org Subject: Opteron 4+0 or 4+2 dimm issue Message-ID: <20040111.004003.28786045.chat95@mbox.kyoto-inet.or.jp> In-Reply-To: <20040108180214.GA79495@dragon.nuxi.com> References: <20040107200042.67A0B16A4D2@hub.freebsd.org> <3FFCD437.6090704@citlink.net> <20040108180214.GA79495@dragon.nuxi.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Hi, David > I personally don't like any 4+0 DIMM configuration motherboard -- unless > it is treated as a 1P Opteron 1xy AGP workstation. I.E. the K8T > Master-FAR. 4+0 SMP configuration has NUMA performance issues that can't > be worked around. Well, the board still has some merit when used as an > SMP -- the memory latency across 1-hop is still better than today's > Pentium-4 SMP systems. I heard that HyperTransport (6.4G/sec?) is fast enough so that NUMA is not necessary to implement. can I ask some question? in FreeBSD/amd64 with SMP, is it always true that for a process runs at processor #1 allocates physical memory belongs also to processor #1? process A -------------------------> accesses DIMM#2 accesses DIMM#1 <-------------------------- process B processor #1 --HyperTransport-- processor #2 | | DIMM #1 DIMM #2 if total memory is 2G attached symmetrically, and if a process allocates near 2G bytes, how physical memory is allocated? Regards, maho
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?20040111.004003.28786045.chat95>