Are you able to test it in Linux using iperf? Seems that when I test stuff in Linux it approaches the transfer spec, e.g. my gigabit ethernet benches around 950 mbit.
Did 3Com ever make a gigabit product?
They used Marvel/Sysconnect chips for some.
The 3c940 is an example of one of those . It was used as a LOM (LAN on motherboard) solution on boards such as the Asus P4P800 (I had one).
3Com cards are active ones (with own CPU, RAM; doing protocoll handling etc.) while most other cards are passive, meaning that the CPU must do most of the work. Active cards are better especially for slower PCs.
no no and maybe yes but no. ISA 3Com cards are dumb as a rock, PCI ones got checksum offloading, but so do bottom of the barrel Realteks RTL8139C+ from 2001.
1Do not purchase this card, even as a joke. It's performance is horrible, and it breaks in many ways.
This turd has enough buffer memory for one packet (max Ethernet MTU 1500 bytes). Once you received one packet all others are lost while you transfer current one to ram.
3C503 is 3Coms spin on NE2000, same controller + some proprietary addons just like RLT8019AS.
The ISA 3Coms such as the 3C509 use programmed I/O, which tended to be slower than shared memory or DMA designs used by contemporary SMC or AMD designs respectively.
ISA DMA is really slow unless you do busmastering. Afaik AMD did busmastering only in VLB/PCI implementations. Using ISA DMA controller is really bad idea.
non B 509 has only 4KB of buffer, enough for one incoming and one outgoing packet.
dionbwrote on 2019-09-10, 10:57:These things are a pain, but starting to get results:
Artisoft AE2/C ISA NatSemi DP83902V 8.27Mbps 79.5% (NE2000-compatible)
Net […] Show full quote
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.
I've recenty done some tests with a 3C509B (ISA) versus a RLT8019AS (ISA) comparing the troughput on a 486.
The 3C outperformed the Realtek easily.
Tested it with mTCP's FTPSERV tool.
The 3C501 was pretty bad, as the "love letter" referenced above notes .. I'd love to find one for my collection just to see how horrible they are.
I'm pretty happy with the 3C503 and I consider them desirable for retro computing. There are clearly better cards out there for performance, but at this point I think widespread software support, reliability and compatibility matter far more than performance. An obscure card that I can't find drivers or configuration software for is useless to me.
For the same reasons I also recommend the NE1000, NE2000, and Intel EtherExpress 8/16 cards. These are also not great performers, but they are widely supported.
The Western Digital WD8003 series (later acquired by SMC) are *great* performers, but you need to be careful on the early ones - some of them have RJ45 jacks but they are *not* compatible with standard Ethernet switches and hubs. I use the AUI port with an external transceiver (Centrecom 210) for those cards.
Intel EtherExpress 8/16 cards. These are also not great performers, but they are widely supported.
I've got a few of those Intel EE 8/16 ISA cards from around 1993-94, so I use those in my 386/486 PCs, mostly to upload/download stuff from my NAS via FTP with mTCP. I have noticed that upload speed is nice but when it comes to download it's abysmal, I don't have the actual numbers at hand but it was definitely under 100KB/s. Is that to be expected with these cards? Also, while the upload process is "continuous", downloading has small "pauses" after a certain amount of data has been written, as if a buffer was being flushed or something like that. The cards pass all the checks on the Intel configuration utility by the way.
i once configured my apple quicksilver with 3com pci x gigabit LAN and an 8 port sata pci x. the disks were external sata to power up independantly. Throughput was glorious and with built in lan and disk controllers disabled there was no heat issues. Best for streaming around the home for me but 3com is terrific. No PSU mod needed. Also served as internet gateway to put an internet threat filter up. These days wifi is the go even though not the best its cheaper than wiring up the house.
1 * The RealTek 8139 PCI NIC redefines the meaning of 'low end.' This is 2 * probably the worst PCI ethernet controller ever made, with the possible 3 * exception of the FEAST chip made by SMC. The 8139 supports bus-master 4 * DMA, but it has a terrible interface that nullifies any performance 5 * gains that bus-master DMA usually offers. 6 * 7 * For transmission, the chip offers a series of four TX descriptor 8 * registers. Each transmit frame must be in a contiguous buffer, aligned 9 * on a longword (32-bit) boundary. This means we almost always have to 10 * do mbuf copies in order to transmit a frame, except in the unlikely 11 * case where a) the packet fits into a single mbuf, and b) the packet 12 * is 32-bit aligned within the mbuf's data area. The presence of only 13 * four descriptor registers means that we can never have more than four 14 * packets queued for transmission at any one time. 15 * 16 * Reception is not much better. The driver has to allocate a single large 17 * buffer area (up to 64K in size) into which the chip will DMA received 18 * frames. Because we don't know where within this region received packets 19 * will begin or end, we have no choice but to copy data from the buffer 20 * area into mbufs in order to pass the packets up to the higher protocol 21 * levels. 22 * 23 * It's impossible given this rotten design to really achieve decent 24 * performance at 100Mbps, unless you happen to have a 400Mhz PII or 25 * some equally overmuscled CPU to drive it.
TLDR: 8139 can quickly copy buffers to/from ram no problem ... as long as you put them in special custom format/order, meaning at the end of the day you need to manually copy data around anyway wasting CPU cycles. Great as a cheap ethernet card for single home user plugged into 10-30MBit cable router, or for PC bangs, not so great on the server/router or if you expect to shuffle gigs of data around the LAN in the background while doing something else at the same time.
I didn't see an ISA 3c515 in that list. Is it also junk?
Supports ISA bus mastering, should provide best ISA performance possible (up to ~5MB? on 100Mbit link) with lowest CPU utilization (not that it matters in non multitasking dos).
I've got a few of those Intel EE 8/16 ISA cards from around 1993-94, so I use those in my 386/486 PCs, mostly to upload/download stuff from my NAS via FTP with mTCP. I have noticed that upload speed is nice but when it comes to download it's abysmal, I don't have the actual numbers at hand but it was definitely under 100KB/s. Is that to be expected with these cards? Also, while the upload process is "continuous", downloading has small "pauses" after a certain amount of data has been written, as if a buffer was being flushed or something like that. The cards pass all the checks on the Intel configuration utility by the way.
try initiating transfer from the other end. Iv seen here on Vogons someone posting wild speed differences depending on the initiating party while transferring from modern system to vintage one.
Well, mine is the 312-V model, which is kind of different and appears to have the BC912 chipset.. I can't find much about this chipset online at all, aside from the driver pack for this specific NIC. The logo appears to be something like AT&T US?
Well, mine is the 312-V model, which is kind of different and appears to have the BC912 chipset.. I can't find much about this chipset online at all, aside from the driver pack for this specific NIC. The logo appears to be something like AT&T US?
Interesting. Its NE2000 compatible, and you have 32KB SRAM buffer on the card. All I could find is this brief inconclusive test from 1998 https://www.ixbt.com/comm/nictest.html
Looking at driver strings suggests it supports early receive and transmit interrupts, so more advanced than run of the mill ISA realteks, but no manufacturer of the chip and no follow up designs sure look weird.
The only ISA NIC I have is a Surecom EP-312-V - now I'm afraid to ask about how terrible it is..
Imho it is not terrible at all.
Instead, it is a cute freak card, isn't it?
ISA card built in 1997...
...still with 10Base2...
...with a serial DIP-8 32kx8 Atmel EEPROM...
...and an unidentifiable controller chip with sophisticated improved NE2000 emulation and a fake logo which obviously tries to imitate the AT&T logo.
This is not the common boring NE2000 clone like those of which I have tossed many into the e-trash bin 😀
I didn't see an ISA 3c515 in that list. Is it also junk?
Supports ISA bus mastering, should provide best ISA performance possible (up to ~5MB? on 100Mbit link) with lowest CPU utilization (not that it matters in non multitasking dos).
I couldn't get its bus mastering to work, so the performance wasn't great.
Still, it was more than 10 Mbps, therefore the fastest ISA card I've had... Fast Ethernet on ISA
There's no native packet driver for 3C515, but properly configured ODI driver + ODIPKT work fine, for good performance with mTCP it seems important to specify 16 buffers in NET.CFG - "buffers 16 1600"
Kiełbasa smakuje najlepiej, gdy przysmażysz ją laserem!
Just adding some color to the prior post ... while mTCP allocates enough buffers to ensure a reliable connection, at those speeds a lot of buffers are needed. The default number of buffers that the ODI driver specifies is *not* sufficient for that speed, so ODI driver was dropping packets, causing a lot of retries and error handling for TCP/IP. Increasing the number of allocated packets at the ODI driver level eliminated the dropped packets and fixed things.
The only ISA NIC I have is a Surecom EP-312-V - now I'm afraid to ask about how terrible it is..
Imho it is not terrible at all.
Instead, it is a cute freak card, isn't it?
ISA card built in 1997...
...still with 10Base2...
...with a serial DIP-8 32kx8 Atmel EEPROM...
...and an unidentifiable controller chip with sophisticated improved NE2000 emulation and a fake logo which obviously tries to imitate the AT&T logo.
This is not the common boring NE2000 clone like those of which I have tossed many into the e-trash bin 😀
