12.3.1. Motherboards, busses, and chipsets

12.3.1.1. * ISA

12.3.1.2. * EISA

12.3.1.3. * VLB

12.3.1.4. PCI

Contributed by David O'Brien <obrien@FreeBSD.ORG> from postings by Rodney Grimes <rgrimes@FreeBSD.ORG>.
25 April 1995.

Continuing updates by Jordan K. Hubbard <jkh@FreeBSD.ORG>.
Last update on 26 August 1996.

Of the Intel PCI chip sets, the following list describes various types of known-brokenness and the degree of breakage, listed from worst to best.

Mercury:

Cache coherency problems, especially if there are ISA bus masters behind the ISA to PCI bridge chip. Hardware flaw, only known work around is to turn the cache off.

Saturn-I (ie, 82424ZX at rev 0, 1 or 2):

Write back cache coherency problems. Hardware flaw, only known work around is to set the external cache to write-through mode. Upgrade to Saturn-II.

Saturn-II (ie, 82424ZX at rev 3 or 4):

Works fine, but many MB manufactures leave out the external dirty bit SRAM needed for write back operation. Work arounds are either run it in write through mode, or get the dirty bit SRAM installed. (I have these for the ASUS PCI/I-486SP3G rev 1.6 and later boards).

Neptune:

Can not run more than 2 bus master devices. Admitted Intel design flaw. Workarounds include do not run more than 2 bus masters, special hardware design to replace the PCI bus arbiter (appears on Intel Altair board and several other Intel server group MB's). And of course Intel's official answer, move to the Triton chip set, we ``fixed it there''.

Triton (ie, 430FX):

No known cache coherency or bus master problems, chip set does not implement parity checking. Workaround for parity issue. Use Triton-II based motherboards if you have the choice.

Triton-II (ie, 430HX):

All reports on motherboards using this chipset have been favorable so far. No known problems.

Orion:

Early versions of this chipset suffered from a PCI write-posting bug which can cause noticeable performance degradation in applications where large amounts of PCI bus traffic is involved. B0 stepping or later revisions of the chipset fixed this problem.

440FX:

This Pentium Pro support chipset seems to work well, and does not suffer from any of the early Orion chipset problems. It also supports a wider variety of memory, including ECC and parity. The only known problem with it is that the Matrox Meteor frame grabber card doesn't like it.