10-100Mbps ethenet cards. Differnce in CPU usage?

[ferix]

It's much simpler than that.
10Mb cards are usually based on early ethernet designs like the ne2000. Those designs are quite old, and they only act as transmiter and receiver devices. All the packet cheking and buffer control relays on the system's cpu. It implies checking the destination MAC address, calculating the CRC checksum of received packets, discarting bad ones, set and control of the reception buffer, and all this stuff. So, on slow cpus, It's a very high load.
Those are very bad designs, but they were very common long time ago.
100mb design are based on new designs, with dedicated MAC controllers that hide all that things to the cpu. The CPU only have to get the packets from the card's buffer ready to be used.

[ferix]

Under NetBSD, at least, a 100 Mbps card has less overhead in an Amiga 1200 than a 10 Mbps, but the difference isn't huge - the top transfer speed is just over 700k/sec in a mostly idle system as compared with the 10 Mbps card which was about 600k/sec.

I'm using a Dlink DFE-670TXD which is an NE2000 card.

There's a diference between a ne2000 based card, and a enhaced ne2000 card.
First one runs at same speed of an original ne2000 card.
The second one is a ethernet card that can run in ne2000 compatible mode, but with the proper driver, it can run faster than the original. Even without this driver, those cards can be slightly faster becouse the improved design.

Believe me, I spent almost a year writting ethernet drivers for embedded applications.

[ferix]

If it gets you on the internet, I'd say it's good.

I agree with that. Nothing more to say.