Questions about LC040 and 16 bit bus

[hellslinger]

Two general questions about LCs and systems with LC040 that I've wondered for a long time:

1. Is it possible to replace the LC040 with 68040 in a system with a socket? Are the pinouts even the same?

2. Why did LCs only have 16 bit data busses instead of 32? How did this really lower the cost of the machine?

 

[MJ313]

The answer to #1 is Yes. People replace the LC040 in their LC475's.... need to add a heat sink though.

No idea about #2.

 

[Elfen]

#2 - to keep the size of the board the same, and not use as many chips (RAM, I/O, etc.) kept the costs down for the LC 475.

If you compare the LC 475 board to the Quadra 700 or 800, you can see those other boards are much bigger and prone to Cap Goo Trace Rot death than the 475. It is more data lines that can be killed, and more chips to support the wider bus.

With fewer chips and lesser data lines for a smaller board, the cost of LC 475 was kept to a minimum.

 

[MJ313]

If you compare the LC 475 board to the Quadra 700 or 800, you can see those other boards are much bigger and prone to Cap Goo Trace Rot death than the 475.

TheWhiteFalcon in 3...2...1....

 

[360alaska]

Two general questions about LCs and systems with LC040 that I've wondered for a long time:

1. Is it possible to replace the LC040 with 68040 in a system with a socket? Are the pinouts even the same?

2. Why did LCs only have 16 bit data busses instead of 32? How did this really lower the cost of the machine?

The LC040 is the 040 without a built-in floating point unit. There is no pin differences or anything...

 

[Elfen]

If you compare the LC 475 board to the Quadra 700 or 800, you can see those other boards are much bigger and prone to Cap Goo Trace Rot death than the 475. 

TheWhiteFalcon in 3...2...1....

That should be "more prone to" as any Mac can be killed off by trace rot.

 

[360alaska]

Not really, Quadra 700 and 800 boards have solid state tantalum caps installed from the factory...

 

[Elfen]

Learn something new every day.

 

[Gorgonops]

The LC040 is the 040 without a built-in floating point unit. There is no pin differences or anything...

Yes, there is clearly some confusion here. The "LC" in the 68LC040 CPU simply means it's lacking the onboard floating-point coprocessor of the "full" 68040; all Apple machines with 68040-family CPUs have 32 bit paths to RAM/ROM and at least *some* of the onboard peripherals. (The later LC-family desktops do indeed have a 16 bit bus translator in front of other parts of the board, like the expansion slot, for backwards compatibility with older models.) This has nothing directly to do with "LC" as used by Apple, and of the Apple machines only the original LC and LCII have a 16 bit bus running "all the way to the CPU"; the LCIII and later are fully 32 bit machines, again, so far as RAM and ROM go.

For tasks that don't involve the FPU a 68040 and 68LC040 running at the same CPU clock will benchmark identically; swapping out the CPU does *not* double the width of the bus. As to how using the 16 bit bus "saved money" in the original LC/LCII, well... it's actually sort of debatable whether it really did. Using a narrower bus does let you get away with a little less glue logic and makes for a slightly more compact board but unlike the case of, say, the Intel 386SX, which was bus-compatible with the 80286 and allowed vendors to upgrade old designs to a "386" with just a chip swap, the Mac LCs were a complete new build for Apple and used the same CPU as a full 32 bit implementation. (The CPU itself has built-in support for the narrower bus, in part to facilitate upgrades from 68000/68010 designs, but it requires more work than a 286->386SX swap.) It probably saved a *few* bucks but I'm 80%-ish sure it was mostly a marketing decision to make sure that LCs performed enough slower than a "real" Mac II-family machine to clearly differentiate the two.(*)

(* I do vaguely recall reading that early prototypes of the LC were meant to be powered by the 68000 and this was changed to avoid having to make a 68000-compatible port Color Quickdraw, but I don't know how true that is. The memory map of the LC/LCII looks more like a Mac II family machine than most 68000-powered compacts but the Macintosh Portable, which is also 68000, sort of looks like a Mac II stuck in 24 bit addressing mode so it's certainly *possible*. If it were true and the ASICs for the LC were originally targeted at the 68000 that does provide a technical justification for the narrow bus.)

 

[MJ313]

That should be "more prone to" as any Mac can be killed off by trace rot.

I didn't realize it either, but TheWhiteFalcon educated me (albeit unknowingly).

 

[hellslinger]

Yes, there is clearly some confusion here. The "LC" in the 68LC040 CPU simply means it's lacking the onboard floating-point coprocessor of the "full" 68040; all Apple machines with 68040-family CPUs have 32 bit paths to RAM/ROM and at least *some* of the onboard peripherals. (The later LC-family desktops do indeed have a 16 bit bus translator in front of other parts of the board, like the expansion slot, for backwards compatibility with older models.) This has nothing directly to do with "LC" as used by Apple, and of the Apple machines only the original LC and LCII have a 16 bit bus running "all the way to the CPU"; the LCIII and later are fully 32 bit machines, again, so far as RAM and ROM go.

For tasks that don't involve the FPU a 68040 and 68LC040 running at the same CPU clock will benchmark identically; swapping out the CPU does *not* double the width of the bus. As to how using the 16 bit bus "saved money" in the original LC/LCII, well... it's actually sort of debatable whether it really did. Using a narrower bus does let you get away with a little less glue logic and makes for a slightly more compact board but unlike the case of, say, the Intel 386SX, which was bus-compatible with the 80286 and allowed vendors to upgrade old designs to a "386" with just a chip swap, the Mac LCs were a complete new build for Apple and used the same CPU as a full 32 bit implementation. (The CPU itself has built-in support for the narrower bus, in part to facilitate upgrades from 68000/68010 designs, but it requires more work than a 286->386SX swap.) It probably saved a *few* bucks but I'm 80%-ish sure it was mostly a marketing decision to make sure that LCs performed enough slower than a "real" Mac II-family machine to clearly differentiate the two.(*)

(* I do vaguely recall reading that early prototypes of the LC were meant to be powered by the 68000 and this was changed to avoid having to make a 68000-compatible port Color Quickdraw, but I don't know how true that is. The memory map of the LC/LCII looks more like a Mac II family machine than most 68000-powered compacts but the Macintosh Portable, which is also 68000, sort of looks like a Mac II stuck in 24 bit addressing mode so it's certainly *possible*. If it were true and the ASICs for the LC were originally targeted at the 68000 that does provide a technical justification for the narrow bus.)

Great response, thanks!

I can't imagine that a few bus decoders/mux chips would save enough money to make an appreciable difference in cost relative to a multi-thousand dollar machine, unless the cost of adding layers to PCBs was enormous in ~1990 (doubtful). I was always suspicious that they were intentionally crippled too, because I remember hating LCs and Performas and how much it hurt the Mac reputation.

Now for Steve Jobs' second absence, Apple is finding new ways to cripple their machines in the hopes of squeezing a few more dollars out of their customers.

 

[Charlieman]

The 16 bit memory bus and 10MB RAM limit (LC and Classic) were almost certainly marketing decisions. Here are some UK education prices for Macs in their lowest spec:

LC I 2/40: £1065;  IIsi 2/40: £1606.50;  IIci 5/40: £2726.50;  IIfx 4/floppy: £2927

Classic I 1/floppy: £490; SE/30 2/40: £1556.75

Third party developers had previously used bits from the Plus and SE (even just ROMs) for creations such as MacColby, Dynamac, Intellitec MX Plus, Outbound (and a few more). The LC and Classic families therefore needed to be functionally limited to discourage third party developers.

Let's imagine how an LC I with a 32 bit RAM bus and no RAM limit might have been developed. Buy an LC I for £1065. Remove logic board and floppy disk from case. Throw away case and sell RAM, PSU and hard disk as spares. Install logic board into new large case with better PSU and cooling. Insert 68030 accelerator board with 8/16/24 bit graphics card (different VRAM sizes and ROMs for different resolutions) into PDS slot. Provide a PDS pass through slot for a network adapter (Ethernet or Token Ring) or for any other purpose. Encourage other manufacturers to make PDS cards with a pass through slot. Install RAM and hard disk at what ever size you like.

This monster could not use NuBus cards, but it would have hurt IIsi and IIci sales if it had been permitted to exist. The price difference between LC I and IIci was £1661.50 (say, £1800 after selling off the surplus LC I parts). There would have been so many opportunities to build something weird and wacky when the 68040 went into production.

 

[Elfen]

If you think a about it, when you're making millions of units like Apple did with the LC, the more you can save on the board by using less chips, the cheaper it will be in the long run. Deleting a $1 part/chip that would give 32bit data bus and forcing everyone with a 16bit data bus, means saving a million dollars a for every 1,000,000th unit they made.

 

[hellslinger]

If you think a about it, when you're making millions of units like Apple did with the LC, the more you can save on the board by using less chips, the cheaper it will be in the long run. Deleting a $1 part/chip that would give 32bit data bus and forcing everyone with a 16bit data bus, means saving a million dollars a for every 1,000,000th unit they made.

And it took apple 10 years to recover from the damage that penny-pinching did to their reputation.

 

[Raptor007]

People replace the LC040 in their LC475's.... need to add a heat sink though.

Does the full '040 run hotter than the LC040?  I just did the swap in a Performa 575 without adding a heat sink and it seems fine, but I didn't really stress test it or check the temperatures.

 

[MJ313]

It does run hotter... but Uniserver's kind of been-there, done-that, so hopefully he will chime in.  It seems the 25Mhz 040's all have passive heatsinks when installed by Apple (like in the Q700). I guess they are there for a reason.

https://68kmla.org/forums/index.php?/topic/20913-full-040-in-a-575-need-a-heat-sink/?hl=heat+sink

 

[Raptor007]

It does run hotter... but Uniserver's kind of been-there, done-that, so hopefully he will chime in.  It seems the 25Mhz 040's all have passive heatsinks when installed by Apple (like in the Q700). I guess they are there for a reason.

https://68kmla.org/forums/index.php?/topic/20913-full-040-in-a-575-need-a-heat-sink/?hl=heat+sink

Fortunately mine's an MC68040RC40A marked 02L88M, which apparently means it's a more recent production with a smaller die:

http://eab.abime.net/showthread.php?p=647148#post647148

But I should probably pop a heat sink on there just in case.  I bet I've got something in my basement parts dungeon that would work.

Oh, and is there a way to monitor the temperature?