On Sun, 30 Jan 2000 00:02:12 +1100 
 Giles Lean <giles@nemeton.com.au> wrote:

 > Netgear
 > 
 >     Recommended enthusiastically by Steve Paul, subject to them
 >     being old stock with Digital chips.  The new Netgear cards use
 >     a different chip and don't work with NetBSD.  (Hey, Jason, you
 >     missed one! :-)

Actually, I didn't really "miss one" :-)  I did include support for it
in my `tlp' driver (which is only in -current for the moment), but
it has some problems.  Most of these are due to the fact that the
Lite-On 82C169 ``PNIC'', which is the chip used on the new NetGear
cards, is the most amazing piece of s**t I've run across in recent
history.  It suffers from no end of hardware bugs, including failure
to load its receive filter information correctly, poor DMA performance
(leading to corrupted packets going out onto the wire), and the tendency
to just "flake out" occasionally.  Also, there is apparently a nasty
receiver bug in them that requires a fairly CPU-intensive work-around,
which basically kills performance.

Based on my experience with these things, I cannot recommend the
new NetGear FA310 PCI cards.  As far as I can tell, Bill Paul, who
wrote the FreeBSD driver for these things, is equally frustrated
with them.  Bill, I feel your pain.

 > Intel 10/100 (EtherExpress Pro?)
 > 
 >     Recommended by Jeff Northon who replaced a bunch of 3com cards
 >     with these.  (On the other hand, Steve Paul likes the 3com
 >     cards mostly and is unsure of the Intel ones! :-)
 > 
 > Jason Thorpe also liked Macronix based cards in -current but noted
 > that he'd not made the 'tlp' driver auto-sense with them yet.

Of all of the DEC Tulip clones that I've actually used, they seem
to work the best, and can be found on many inexpensive boards.  The
Linksys LNE100TX Version 2 also has a variant of the Macronix chips
called the "PNIC-II", which, even though it has the Lite-On name on
it, is actually a Macronix chip (seems to be very close to the 98725).

 > I've had a previous recommendation for PCI NE2000s but I guess they
 > don't do DMA anymore than the ISA ones did, so they're perhaps
 > suited to light duty (like the realtek! :-).

...and are also limited to 10baseT or at best 10baseT-FDX.

 > For myself I went to my usual supplier and found:
 > 
 > - they didn't stock SMC

Yah, this seems to be a problem, and it really annoys me, because
I find the SMC 9432TX to be one of the best cards around, hands-down!

 > - the Netgear cards (AUD$50) were all the new type

Yah, the 21140A-based ones are basically impossible to find.

 > - plenty of stock of Intel EtherExpress (AUD$125)
 > - there was a RealTek card on the pricelist (AUD$32)
 > 
 > Having discovered that the HP branded NIC in a loaner machine is
 > picked up by the 'le' driver I only needed one new card (the 486s
 > and hp300 can stay on coax -- they'll never notice :) and chose
 the Intel one.

You what what I really wish would happen... Some brave company out there
would produce a NEW board based on the now-Intel-badged 21143 ... they
are STILL PRODUCED, and are now generally nice, low power parts (since
they're QUITE common on CardBus Ethernet cards).  As much as I wish I
would never hear the word "Tulip" again (you try maintaining that driver
sometime ... I now understand why Matt Thomas got sick of it :-), I still
think the chip is very good.  The only real `problem' with it is that it's
very flexible, and thus complicated at times :-)

I guess I should also say that if you can find any cards based on
the SiS 900 chip, those work okay, too.  I wrote a driver for that
chip for NCI (it's on-board on some NC), and even though I had early
rev chips that were... buggy :-), I found it to be fairly nice to
program, and it worked pretty well.  There's a newer version of it,
too, called the SiS 7016, which is not yet supported by the `sip'
driver, but adding support for it would not be that hard.  Unfortunately,
I don't know what cards on the market have that chip on them... I'm
suspecting they're currently being relegated to the integrated-on-motherboard
market.

        -- Jason R. Thorpe <thorpej@nas.nasa.gov>