68000 Coprocessor on Apple Macintosh NUBUS Ethernet ?!

[demik]

Was busy cleaning some NuBus cards and noticed something weird on this card. I'm amazed it's there, as usually you will find something more a 68HC05 or 68HC11 performance wise, so it has to do some "heavy" processing.

Pictures attached for reference



Close view:



What it is for ? Something like LRO / TCP Checksum Offload ? Does MacOS even support this ? I guess one can dump the ROM and have a look though...

 

[techknight]

Only conclusion I can draw is the card handled the entire protocol stack (TCPIP, Appletalk, etc)

 

[Gorgonops]

There was a thread about this pretty recently; TL;DR, the 68000 is for running A/ROSE, an experiment with trying to make Nubus cards easier to build by making them fiendishly complicated and overbuilt. (Yes, seriously, not joking.)

While in theory at least this ridiculously powerful blob of hardware would have allowed tricks like offloading whole parts of the protocol stack in practice it was never used anywhere close to its full potential.

 

[Unknown_K]

I think the old Sillicon Express II Nubus SCSI cards had 68000's as well. Somebody figured it would be better to offload all the work to a CPU on the card instead of using the main CPU.

 

[olePigeon]

There is a measurable difference when using A/ROSE and the Apple NuBUS ethernet card.  My 040 accelerated IIci has noticeably faster transfer speeds when using the Apple NB card versus a Farallon.  I suspect the advantages are more pronounced on slower machines like the original Macintosh II.

 

[jeremywork]

Other than speed, you'll find an 030 machine like the IIfx relatively unhindered when acting as a fileserver using one of these. With normal ethernet cards, a large appleshare transfer will slow other processes significantly, but with A/ROSE it's rather seamless.

 

[demik]

There was a thread about this pretty recently; TL;DR, the 68000 is for running A/ROSE, an experiment with trying to make Nubus cards easier to build by making them fiendishly complicated and overbuilt. (Yes, seriously, not joking.)

While in theory at least this ridiculously powerful blob of hardware would have allowed tricks like offloading whole parts of the protocol stack in practice it was never used anywhere close to its full potential.

Oh i didn't know A/ROSE actually ended up in production. Thats interesting, so A/ROSE is the "offloading" part. The hardware part doesn't seems overcomplicated (Looking at the PCB traces, the 68000 talks to the SONIC chip and NuBus circuitry). The software on another hand...

 

[demik]

There is a measurable difference when using A/ROSE and the Apple NuBUS ethernet card.  My 040 accelerated IIci has noticeably faster transfer speeds when using the Apple NB card versus a Farallon.  I suspect the advantages are more pronounced on slower machines like the original Macintosh II.

I need to benchmark this on my IIcx !

 

[Gorgonops]

From reading what few docs there are about to what extent A/ROSE was actually leveraged by that card it sounds like the 68000 mostly acts as a sort of caching substitute for a DMA controller, there isn't really any protocol offloading per se, but considering how some of the structural defects in the classic MacOS can really make it suffer when trying to respond in real time to a peripheral every little bit probably helped.

 

[demik]

From reading what few docs there are about to what extent A/ROSE was actually leveraged by that card it sounds like the 68000 mostly acts as a sort of caching substitute for a DMA controller, there isn't really any protocol offloading per se, but considering how some of the structural defects in the classic MacOS can really make it suffer when trying to respond in real time to a peripheral every little bit probably helped.

If by chance you have thoses docs around, I would gladly take a look at them

 

[demik]

Meh. IIcx doesn't boot anymore, caps leaked...

One more to the "replace fucking caps" list.