Subject: Re: Problems with HDs and VIA Chipsets
To: None <wojtek@3miasto.net>
From: Charles M. Hannum <abuse@spamalicious.com>
List: port-i386
Date: 04/30/2001 00:31:16
On Mon, Apr 30, 2001 at 09:06:04AM +0200, wojtek@3miasto.net wrote:
> > > and non-pc boxes are using more and more of PC-parts.
> > 
> > At this point, the 440BX is pretty old, and the performance bites.  So,
> i cannot agree.  it has best memory IO speed of all motherboards i've
> seen. (including KT133 duron "200Mhz" ones)

Either your test is garbage, or you've never tried a ServerWorks or
high-end Intel chipset.  The ServerWorks BLOWS THE DOORS off the BX,
particularly for I/O bandwidth.  For different applications (and, more
importantly, different I/O peripherals), I get a 2-3x bandwidth
improvement going from a BX to a ServerWorks board.  There's hardly
even a basis for comparison.

And the ServerWorks chipset isn't even much more expensive.

Furthermore, the BX is considered more or less end of life, and -- for
non-portable applications -- Intel actively discourages new designs
based on it.  (I *wish* I could go into more detail on that.)


If you want specifics, I'll just point out that the BX doesn't support
133MHz SDRAM, any form of striping to improve burst performance, or any
combination of 66MHz or 64-bit PCI (except abusing the AGP interface).
It also has particularly bad read latency.  Testing further suggests
that PCI bursts completely tie up the SDRAM interface, and therefore
I/O traffic (especially for devices with limited burst size, like all
of the most common Ethernet controllers) totally screws your CPU-memory
bandwidth.

Note that all of this was determined not only by running benchmarks,
but by watching the traffic with a PCI bus analyzer and a scope at the
same time.

In short, the BX sucks.  I don't know what its design goals were, but
it's clear that performance was not at the top of the list.