VOGONS


Why are 3Com NICs in such high regard?

Topic actions

Reply 40 of 93, by Caluser2000

User metadata
Rank l33t
Rank
l33t

Running a no name RTL8139B with a 2.4 kernal on this P166mmx setup. Smooth as silk on all the older distros from 1999 on I've tried. Non of this silly running around hunting for drivers crap.

Last edited by Caluser2000 on 2019-09-10, 06:00. Edited 1 time in total.

There's a glitch in the matrix.
A founding member of the 286 appreciation society.
Apparently 32-bit is dead and nobody likes P4s.
Of course, as always, I'm open to correction...😉

Reply 41 of 93, by Errius

User metadata
Rank l33t
Rank
l33t
Grzyb wrote:
Errius wrote:

3C905CX-MLP

A NIC for My Little Pony fans? 😁

That appears to be the low profile version of the 3C905CX-TX-M. The two cards appear to be identical except for the height of the mounting bracket.

Is this too much voodoo?

Reply 43 of 93, by dionb

User metadata
Rank l33t++
Rank
l33t++
feipoa wrote:

Have you tested the CPU load and throughput of a gigabit PCI card on these slower systems?

Not yet. Will see if I can dig up some. Most of my spare Gb NICs are 64b. The newer PCI-X cards won't work in a 5V 32b slot, but older ones might. I'll see if I can find time (and NICs) today.

Tbh I consider ISA to be more interesting, but getting those to work under ancient Linux will be a challenge - there's no repositories for software from 15 years ago online, so if it doesn't work out of the box, it's not likely to be made workable. I did quickly look at 3C509B, an SMC card and an Intel EtherExpress/Pro10+ (all PnP cards), but only the 509B module was able to detect its card but it couldn't actually control it, and the other two couldn't even manage that. It's possible the 509B wasn't in PnP mode and had I/O and IRQ set to something exotic, but I tried the obvious combinations I usually use (0x210 IRQ 11 and 0x300 IRQ 10) and they weren't it...

Alternative would be to test under DOS, but problem there is finding a good test of both throughput and CPU load given lack of multitasking ability.

Reply 44 of 93, by Caluser2000

User metadata
Rank l33t
Rank
l33t
dionb wrote:
feipoa wrote:

Have you tested the CPU load and throughput of a gigabit PCI card on these slower systems?

Tbh I consider ISA to be more interesting, but getting those to work under ancient Linux will be a challenge - there's no repositories for software from 15 years ago online,

You obviously are not looking hard enough then. I used a few Debian Sarge repository's just this morning.

snapshot29.png
Filename
snapshot29.png
File size
116.33 KiB
Views
1750 views
File license
Fair use/fair dealing exception
Last edited by Caluser2000 on 2019-09-10, 22:22. Edited 1 time in total.

There's a glitch in the matrix.
A founding member of the 286 appreciation society.
Apparently 32-bit is dead and nobody likes P4s.
Of course, as always, I'm open to correction...😉

Reply 45 of 93, by GigAHerZ

User metadata
Rank Oldbie
Rank
Oldbie

I plan to test some ISA NICs on my own, but haven't gotten to it yet. I have 40MHz 386DX, 100MHz 486DX4 and 266MHz Pentium MMX to do the benchmarks.

But what i did notice is that AMD PCnet for ISA works way faster and more fluently than the 3com 3C509B did on my 386. No benchmarks, but stuff just works under dos and windows for workgroupd 3.11 faster and more fluently.

Last edited by GigAHerZ on 2019-09-10, 06:53. Edited 1 time in total.

"640K ought to be enough for anybody." - And i intend to get every last bit out of it even after loading every damn driver!

Reply 46 of 93, by Errius

User metadata
Rank l33t
Rank
l33t

These are the ISA and PCI members of the EtherLink III family:

  • 3C509
  • 3C509-C / 3C509-COMBO
  • 3C509-TP
  • 3C509-TPO
  • 3C509B
  • 3C509B-C / 3C509B-COMBO
  • 3C509B-TP
  • 3C509B-TPC
  • 3C509B-TPO
  • 3C590-C / 3C590-COMBO [this is a PCI card]
  • 3C590-TPO [this is a PCI card]

The rule seems to be:

  • No suffix - BNC and AUI connectors
  • C or COMBO - BNC, AUI and RJ45
  • TP - AUI and RJ45
  • TPC - BNC and RJ45
  • TPO - RJ45

Clearly you want to steer clear of the no-suffix cards.

Last edited by Errius on 2019-09-10, 23:49. Edited 1 time in total.

Is this too much voodoo?

Reply 47 of 93, by dionb

User metadata
Rank l33t++
Rank
l33t++

Right, managed to get some 3C509s to play nice with my Linux install (somehow can't do DHCP on anything that wasn't autodetected at boot - but with manal IP it works fine).

3Com 3C509-TP (1992 model) ISA 3Com 8350-01 6.06Mbps 1.0%
3Com 3C509B-TPO ISA 3Com 40-0130-002 8.41Mbps 4.6% (note: no duplex option in config, maybe old firmware)
3Com 3C509B-C ISA 3Com 40-0130-002 8.08Mbps 4.3% (note: duplex option present in config)
3Com 3C509B-C ISA 3Com 40-0130-004 8.60Mbps 4.6% (note: duplex option present in config)

Big difference between the -B and old non-B cards, with the -B being much faster, but the non-B seeming to load the CPU much less (even when taking lower speed into consideration), differences between the B cards seem within margin of error, it's all good stuff. Will now try some other ISA cards...

Edit:
These things are a pain, but starting to get results:
Artisoft AE2/C ISA NatSemi DP83902V 8.27Mbps 79.5% (NE2000-compatible)
Networth UTB16B ISA NatSemi DP83905AVQB 8.03Mbps 98.6% (NE2000-compatible)
Noname NE-12 ISA UMC9008F 8.09Mbps 95.9% (NE2000-compatible)

Well, looks like in case of ISA the 'noname' NE2000 cards are definitely NOT the ones to go for, at least not under Linux... managing to pull a P3-500E to its knees pushing through 8Mbps. Just wow.
Note that I was just using the regular NE2000 (module: ne) for the cards with that behaviour. Some offered alternatieve config (the Artisoft could use DMA), and other cards (3c503, SMC EtherEZ) offer shared memory, but I can't get those to run at all.

Reply 48 of 93, by Caluser2000

User metadata
Rank l33t
Rank
l33t

In a hobby setting I'd imagine it really doesn't matter what you use as long as it works.

There's a glitch in the matrix.
A founding member of the 286 appreciation society.
Apparently 32-bit is dead and nobody likes P4s.
Of course, as always, I'm open to correction...😉

Reply 49 of 93, by Sedrosken

User metadata
Rank Member
Rank
Member

My reasons are much the same as everyone else's. They're honking everywhere, I took home, like, six of the 3C-905B cards from my high school one day a few years back because they were otherwise going to the dump. The 905s in particular have baked-in drivers for Windows on everything from 95 on up, they keep CPU utilization down while still being reasonably quick. 509s have a really nice tendency to Just Work™ no matter what you chuck them in. If I had the dongle that goes to my 3C-589 I'd honestly use that on my T2130CS over my Intel card, just because it's so convenient to have the baked in drivers. I got lucky in that my Intel card came with the driver disks (which now that I think about it I really ought to get to imaging and mirroring... hm).

Nanto: H61H2-AM3, 4GB, GTS250 1GB, SB0730, 512GB SSD, XP USP4
Rithwic: EP-61BXM-A, Celeron 300A@450, 768MB, GF2MX400/V2, YMF744, 128GB SD2IDE, 98SE (Kex)
Cragstone: Alaris Cougar, 486BL2-66, 16MB, GD5428 VLB, CT2800, 16GB SD2IDE, 95CNOIE

Reply 51 of 93, by dionb

User metadata
Rank l33t++
Rank
l33t++
Caluser2000 wrote:

In a hobby setting I'd imagine it really doesn't matter what you use as long as it works.

If Linux CPU usage is also seen in DOS/Win9x, I'd say it does, at least with ISA NICs. If your CPU is tied up with servicing programmed I/O NIC it's not available for say a game, which could cost you a lot of FPS.

Reply 52 of 93, by the3dfxdude

User metadata
Rank Member
Rank
Member
Sedrosken wrote:

My reasons are much the same as everyone else's. They're honking everywhere, I took home, like, six of the 3C-905B cards from my high school one day a few years back because they were otherwise going to the dump. The 905s in particular have baked-in drivers for Windows on everything from 95 on up, they keep CPU utilization down while still being reasonably quick. 509s have a really nice tendency to Just Work™ no matter what you chuck them in. If I had the dongle that goes to my 3C-589 I'd honestly use that on my T2130CS over my Intel card, just because it's so convenient to have the baked in drivers. I got lucky in that my Intel card came with the driver disks (which now that I think about it I really ought to get to imaging and mirroring... hm).

Does the 3C-589 also benefit users with low cpu usage versus NE-2000 compatible and other pccards? Can someone benchmark this?

Reply 53 of 93, by Eep386

User metadata
Rank Member
Rank
Member

Earlier I had to retire a dying 3com 3C905B-TX that was doing *very* bad things. Also retired an ancient 82C557-based Intel card for making no attempt whatsoever at trying to work.
I've had to pull a more than just a few Realteks in my time, but I've also deep sixed a bunch of 3coms, particularly their later smaller-chip, cost-reduced designs. I've replaced comparatively few Intels though, and never any ISA Intel NICs to my memory.

On the whole I'd say Intel generally made the better (as in longer-lasting) NICs, beforementioned handful of duds aside. 3com NICs do have a great reputation but I found them to be a bit more mediocre in terms of longevity. (Performance is generally really nice though.)

Life isn't long enough to re-enable every hidden option in every BIOS on every board... 🙁

Reply 55 of 93, by Sedrosken

User metadata
Rank Member
Rank
Member
the3dfxdude wrote:

Does the 3C-589 also benefit users with low cpu usage versus NE-2000 compatible and other pccards? Can someone benchmark this?

Wouldn't know. I have the card but no dongle so obviously I can't connect it to my network to test.

Nanto: H61H2-AM3, 4GB, GTS250 1GB, SB0730, 512GB SSD, XP USP4
Rithwic: EP-61BXM-A, Celeron 300A@450, 768MB, GF2MX400/V2, YMF744, 128GB SD2IDE, 98SE (Kex)
Cragstone: Alaris Cougar, 486BL2-66, 16MB, GD5428 VLB, CT2800, 16GB SD2IDE, 95CNOIE

Reply 57 of 93, by derSammler

User metadata
Rank l33t
Rank
l33t
dionb wrote:

Actual speed results are close enough to be considered a virtual tie (apart from the 10MbE card).

Problem is that a Pentium 3 @ 500 MHz is already way to fast for these cards. You would need to test them in a Pentium 1 or even a 486. The almost non-existing difference is because you really only seeing how much of the CPU is used to handle 100 mbit/s of data. The underlying protocoll handling that the 3Com cards can do without using the CPU is so trivial for a Pentium 3 that you won't gain anything by using such a card.

Reply 58 of 93, by dionb

User metadata
Rank l33t++
Rank
l33t++
derSammler wrote:
dionb wrote:

Actual speed results are close enough to be considered a virtual tie (apart from the 10MbE card).

Problem is that a Pentium 3 @ 500 MHz is already way to fast for these cards. You would need to test them in a Pentium 1 or even a 486. The almost non-existing difference is because you really only seeing how much of the CPU is used to handle 100 mbit/s of data. The underlying protocoll handling that the 3Com cards can do without using the CPU is so trivial for a Pentium 3 that you won't gain anything by using such a card.

Possibly, but given the massive differences in CPU usage I did see in ISA cards I'm not so sure. You gain almost all your CPU back there with 3C509 vs NE2000, even with a completely overkill CPU. It would be odd if minor differences with PCI suddenly ballooned.

That said, I intend to do a DOS/ISA comparison one of these days - I'm working on getting pre-installed drivers archived for all my DOS stuff, including NICs. That would show whether the differences were due to hardware or just (dodgy?) Linux drivers. Hopefully I'd also have more luck with the more diverse ISA cards I couldn't get running under Linux. But that's a long-term low-priority idea. I might even add some PCI cards, although something tells me OS (or lack thereof) would be a bigger bottleneck with DOS when working with 100MbE cards.