Wow very good to know.

Marco wrote on 2025-10-27, 18:08:

Thanks for the hint. There are unfortunately no packet drivers available for the 515 😒

Bummer... I've run into very few cards where there isn't a packet driver (or one for a compatible card).

To be completely honest however, although I had dozens of NICs back in the day (and still have a few, even a couple
ISA ones) which I use on the 4-5 DOS systems I still have, I don't tend to try and run DOS on my newer systems, and
haven't needed to go looking for newer packet drivers. In the very rare cases where I've booted DOS on something
newish for testing - I can usually just stuff in a PCI card if I need networking.

I'm sure most newer hardware doesn't have need for packet drivers as DOS has been almost eliminated
from "modern" consideration factors in picking an OS.

So Tests conducted:

SMB Large file Transfer (15MB):
DOS: 428kb/s
WIN311: 788kb/s
WIN95: 750kb/s

FTP Large File Transfer (from Slackware)
WIN31: 837kbps
WIN95: 1120kbps

Not to NUL but fast HDD. Couldnt get wsftp to download to NUL

Further remarks:
- switch port config is in line with NIC config 100MB HD

So at the end I couldn’t get closer to a possible root cause.

I might test another router. But from the current I do easily get 6MB and more for SMB download speed from a modern PC.

My ISA Bus achieves up to 4400kb/s for linear HDD reads so even without busmastering these results are disappointing and more or less identical or slower then the ones of a 3c509.

Next steps:

1. Use another router (for another ftp test)
2. Test the card in my 486dx2 80 as proposed.

Anyway maybe these results do already give someone some more hints?!

Thank you

Marco wrote on 2025-10-27, 21:59:

Not the NUL but fast HDD. Couldnt get wsftp to download to NUL

You mean you're using the WS_FTP client?

My preferred setup is:

• FTP server on a modern PC
• FTP client on the machine being benchmarked
  • DOS - FTP.EXE from mTCP, with "MTU 1500"
  • Windows 3.x - FTP.EXE from TCP/IP-32, even when using some other TCP/IP stack
  • Windows 9x/NT - FTP.EXE shipped with the system

All three clients can download to NUL.

Anyway maybe these results do already give someone some more hints?!

They look OK for a CPU with 16-bit bus.

Will try now with local FTP on the router itself.

Results:
- Download performance from Modern PC (wifi, to HDD):
--- FTP: 3,35 MB/s
--- SMB: 3,45 MB/s

- Download performance from Testsystem TI486SXLC2 (LAN):
--- WIN95 - FTP.EXE: 467kb/s
--- WIN95 - WSFTP: 520kb/s
--- WIN311 - FTP.EXE: 760kb/s
--- WIN311 - WSFTP: 551kb/s

So all benches now conducted with the testsystem.
Very disappointing.

Again - my conclusion was that 100 Mbps NIC doesn't make sense in a 386, and I was testing 386DX, with 32-bit bus!
Re: Expected network throughput on ISA network cards

Seems like I simply have to accept that fact as well 😀

I was at least hoping for CPU usage reduction by DMA Busmaster. But as always mentioned here in the forum: seems to be black magic and no one really knows if it's working / used.

Again - thanks to everyone

It remembers me on my tests with an EISA 3Com 3C597 (100 MBit/s) in a 486DX50 with an Adaptec 2740W EISA SCSI controller:
EISA: Win95C --> Windows 10: 14.5 MBit/s (from cache)
EISA: WinNT 4.0 --> Windows 10: 25 MBit/s (from cache), 22 MBit/s (from disk)
Re: Why are 3Com NICs regarded in such high regard?

Great results.

I could extend my testings:

1. make NDIS message „Dma busmaster disabled“ disappear by doing auto Dma tuning in config
- thanks for that hint upfront here which I really missed earlier
- results could be improved slightly but marginally

2. tested now on my system 2 (isa at 13,3MHz):
- DOS large file copy from SMB: 1152kb/s
- Win311 large file copy from SMB: 937kb/s
- Win311 large file copy FTP to NUL: 1520kb/s

There is indeed an increase but nothing close to even 50mbit. Again 100MB HD media type setting.

I get more and more the impression that a comment a read here somewhere was right: 515 is capable of allowing 100MB switch port configurations in large setups (auto config, homogen config, homogen switch systems, …) but cannot provide notably (!!) more bandwidth than the 509b.

So, 3C515 performance is *NOT* worse than 3C509B... MYTH BUSTED!

But the window where it's actually better is pretty narrow...
- doesn't make sense in a 386, even on steroids
- doesn't make sense in anything with EISA or PCI
- OK for ISA-only and VLB machines with fast CPU, primarily 486, possibly also NexGen and other oddballs

Anyway, don't expect miracles - though I suspect the results would be somewhat better with protocols lighter than TCP/IP - IPX? NetBEUI? EtherDFS?

I hope to do some benchmarking on my 486DX2 later this week, but only with TCP/IP - nowadays, setting up a server with other protocols is too much work...

Yes I agree.

Some remarks:

1. the myth is not (fully) busted when we talk about operations in „low end“ systems. Maybe at a certain time I will re do tests with the 509b.
2. Etherdfs didn’t show improvements in my previous tests on the 509b. The 515 lacks packet driver Herefore

Anyway thanks to all of your help.

So I've got the following machine:

FTP server is a modern PC.
FTP client downloading in BINARY mode to NUL.
mTCP set to MTU 1500.
WfW 3.11 using TCP/IP-32.
All results in KBytes/s.

First, the 10 Mbps card, with these settings:

Always using the drivers from this package:
83-1024-001
EtherDisk for the Etherlink 10 ISA NICs (3C509B Family)
Version 6.1
Copyright 1999 3Com Corporation. All rights reserved.

mTCP, native PD: 786
mTCP, NDIS+DIS_PKT: 995
mTCP, ODI+ODIPKT: 919
WfW: 760

Now, setting "Network Driver Optimization: DOS Client", which is supposed to improve network throughput at the cost of higher CPU utilization...

mTCP, native PD: 786
mTCP, NDIS+DIS_PKT: 994
mTCP, ODI+ODIPKT: 921
WfW: 769

Next, the 100 Mbps card, with these settings:

Did the "tune the DMA parameters" procedure.

Always using the drivers from this package:
83-0809-001
3Com Fast EtherLink ISA Parallel Tasking 10/100 Base-TX
Version 2.0
Copyright 1997 3Com Corporation. All rights reserved.

mTCP, NDIS+DIS_PKT: 860
mTCP, ODI+ODIPKT: 1236 (driver displays 100 FD)
WfW: 1185

Setting "Network Driver Optimization: Maximized Network Performance"...

mTCP, NDIS+DIS_PKT: 1371
mTCP, ODI+ODIPKT: 1233 (driver displays 100 FD)
WfW: 826

Back to original settings, and forcing 100 HD (can't set 100 FD in 3C515CFG)...

mTCP, NDIS+DIS_PKT: 1371
mTCP, ODI+ODIPKT: 1233 (driver displays 100 HD)
WfW: 1188

Still forced 100 HD, and disabling bus mastering...

mTCP, NDIS+DIS_PKT: 988 (driver displays "Bus mastering disabled by Keyword.")
mTCP, ODI+ODIPKT: 927 (driver displays 100 HD)

Conclusion:

• NWay auto negotiation in this card sucks, and wrong speed/duplex mismatch can indeed make it worse than a 10 Mbps card - it's better to force it manually
• disabling bus mastering also hurts a lot
• even in perfect setup, the Corkscrew is not a 100 Mbps card - more like a 15 Mbps card 🤣

1. Fully agree to your conclusions 😀

2. Hereby my results:

3. Conclusions:
- 515 is indeed faster. Myth busted yes
- some strange numbers in red I dont understand
- Even though an addon the TCP IP Layer on Win311 seems to be most performant overall. The fast HTTP download might be related to file-still-in-IE-cache
- Differences in DOS SMB differ largely between Ti486 and 486DX280.
- Win95 suffers largely in lot-small-files transfers. But this seems to be mainly due to high cpu usage for graphical copy jobs (windows commander). running same tests under NC5 where much faster (170kb/s on 3c09).

Thanks again for the nice testings. I will in short try to re-do exactly your tests.

Can you therefor shortly tell me what is meant by:
..., native PD
..., NDIS+DIS_PKT
..., ODI+ODIPKT

I am just aware of using mTCP with packetdriver provided by 3com v6.1 - nothing elso to choose here. Thanks upfront

Marco wrote on 2025-10-31, 10:16:

Can you therefor shortly tell me what is meant by: ..., native PD ..., NDIS+DIS_PKT ..., ODI+ODIPKT […]
Show full quote

Can you therefor shortly tell me what is meant by:
..., native PD
..., NDIS+DIS_PKT
..., ODI+ODIPKT

mTCP - and many other TCP/IP packages for DOS - need a Packet Driver.
In case of lack - or poor quality - of a PD, it can be substituted by NDIS or ODI driver with appropriate converter/shim.

I'm attaching the PD replacements configured for 3C515...

Thanks a lot. Will test.

Wait: I just realize that there ARE packet drivers then for the 515???

Marco wrote on 2025-10-31, 14:32:

Wait: I just realize that there ARE packet drivers then for the 515???

Not quite - but there are drop-in replacements.
So yes, it's possible to use PD applications with that card.
There may be some limitations - but I only know about the incompatiblity between mTCP NetDrive and ODI+ODIPKT.

Grzyb wrote on 2025-10-31, 16:10:

Not quite - but there are drop-in replacements. So yes, it's possible to use PD applications with that card. There may be some l […]
Show full quote

Marco wrote on 2025-10-31, 14:32:

Wait: I just realize that there ARE packet drivers then for the 515???

Not quite - but there are drop-in replacements.
So yes, it's possible to use PD applications with that card.
There may be some limitations - but I only know about the incompatiblity between mTCP NetDrive and ODI+ODIPKT.

As far as I know there is no incompatibility like this. Are you saying that mTCP Net drive doesn't work with any combination of ODI driver and ODIPKT? Is there a bug report that I missed?

mbbrutman wrote on 2025-10-31, 16:52:

As far as I know there is no incompatibility like this. Are you saying that mTCP Net drive doesn't work with any combination of ODI driver and ODIPKT? Is there a bug report that I missed?

I mean this - Re: mTCP NetDrive: network attached storage for DOS 2.0 or better

I've tried it many times, with various NICs, and the result was always the same - NetDrive works with NDIS+DIS_PKT, but doesn't work with ODI+ODIPKT.

Just noted that the non-busmaster setting results are similar to the ones of the 509b. So you could (!) assume that speed advantage mainly derive from the busmaster capability. Just a theory 😀