[KLUG Members] AGP problem (maybe?)

[Bryan-TheBS-Smith]

Adam Tauno Williams wrote:
> I've seen this. (Don't even ask why someone had a *server* with
> an AMD processor and/or a AGP card.  Ick on both, IMHO).

Yeah, at least pair it with a PCI card!

> In that system's BIOS it was called "memory aperature".

Yeah, that's the term I couldn't remember in my previous E-mail.

> The system originally had 128Mb of RAM and worked fine.
> They added 64Mb for a total of 192M,  and the server blew up
> chunks all over the nice XFS tiled floor.  The memory aperature
> was set to "128M",  changing it to "192M" (I was a little
> suprised that setting was available).  It seems to shove the
> address space the card uses up out of the way.  The card was
> using 16Mb (why oh why on a server...) and the OS then found
> 176M and tooled along as happily as crappy PC hardware ever does.

Could be an L2 cache addressing issue.  Several ALi/ViA chipsets would
only cache anywhere from 128-512MB of RAM, depending on chipset, L2
cache size, JTAG chip, etc...

> AGP = A Game Platform.  Cheap tricks to get better video
> performance - IMHO.

A "Gee-wiz" PCI (AGP) bus.  Instead of actually trying to get multiple
PCI busses using 64-bitx66MHz in its chipsets, Intel invented AGP.  AGP
is as unstable as VLB was.  God knows the aperature/DME (direct memory
execution) approach has utterly _failed_, and local buffer is much, much
faster.  But Intel didn't think of this, and their i740/752 AGP cards
were horrendously slow because they used DME over local buffer by
default.

-- TheBS

-- 
Bryan "TheBS" Smith    mailto:b.j.smith@ieee.org   chat:thebs413
Engineer  AbsoluteValue Systems, Inc.  http://www.linux-wlan.org
President    SmithConcepts, Inc.    http://www.SmithConcepts.com
----------------------------------------------------------------
* Pentium stop, Pentium SMP go, Athlon MP go ... go very fast! *