Subject: Re: NetBSD-amd64v2.0 as server on Shuttle SN95G5 hardware
To: Alicia da Conceicao <alicia@engine.ca>
From: Richard Rauch <rkr@olib.org>
List: port-amd64
Date: 12/29/2004 21:00:25
For the most part, I have little comment, but I would share two points,
having been using an nVidia nForce 3 based system for over a year (I've
been using this AMD64 box as my primary workstation):

 * -current tends to pause irritatingly for a few seconds (sometimes
   quite a few).  I'm not sure exactly what the issue is.  Usually it
   seems to just affect a single process, so I suspect that when the
   "whole system" feels like it freezes for a few seconds, it's really
   something like the X server process.  (I've never managed to
   verify whether ping works at those times from another system.  The
   pause doesn't last long enough for that to be practical...)

   This did not happen around the middle of 2004, but when I updated
   to a newer -current, I ran into it.  I do not know if 2.0 has
   this, or if it is AMD64 specific.

 * I would *avoid* the nVidia-based motherboard, personally.
   (The machine you mention is described as having an nVidia
    nForce 3 chipset.)

   Mine has *never* (anytime that I've tried) supported IOAPIC
   in the kernel.  The port-master has said that he doesn't
   consider running with IOAPIC to be really supported.  I have
   never heard of the problem being fixed.

   With IOAPIC enabled, some devices do not configure, others stop
   working while the OS is running.  (Some work reliably, though.)

   Building a custom kernel without IOAPIC is the only workaround
   that I know of.  If running without IOAPIC is *required* for
   this chipset, and is "not really supported", then I'd not
   recommend using any system based on this chipset.

   I am not aware of any problems with running without IOAPIC,
   except that IOAPIC is enabled in the GENERIC kernel.

   This does not appear to be a NetBSD-only issue.  About a year
   ago, I heard that GNU/LINUX had similar problems before they
   "fixed" it.  I perused their sources at the time and saw that
   they just (dynamically) disable IOAPIC if an nVidia chipset
   is detected.

-- 
  "I probably don't know what I'm talking about."  http://www.olib.org/~rkr/