A story you hear all the time about the Apple IIGS is that Apple purposefully underclocked or limited its processor in some way to protect the nascent Macintosh, and ensure the IIGS, which could build upon the vast installed base of Apple II computers, would not outcompete the Macintosh. I, too, have always assumed this was a real story – or at least, a story with a solid kernel of truth – but Dan Vincent decided to actually properly research this claim, and his findings tell an entirely different story.
His research is excellent – and must have been incredibly time-consuming – and his findings paint a much different story than Apple intentionally holding the IIGS back. The actual issue lied with the production of the 65816 processor that formed the beating heart of the IIGS. It turns out that the 65816 had serious problems with yields, was incredibly difficult to scale, and had a ton of bugs and issues when running at higher speeds.
What a ride, huh? Thanks for making it this far down a fifty-plus minute rabbit hole. I can’t claim that this is the final take on the subject—so many of the players aren’t on the record, but I’m pretty confident in saying that Apple did not artificially limit the IIGS’ clock speed during its development for marketing purposes. Now, I’m not a fool—I know Apple didn’t push the IIGS as hard as it could, and it was very much neglected towards the end of its run. If the REP/SEP flaws hadn’t existed and GTE could’ve shipped stable 4MHz chips in volume, I’m sure Apple would’ve clocked them as fast as possible in 1986.
↫ Dan Vincent
Promise me you’ll read this article before the weekend’s over. It’s a long one, but it’s well-written and a joy to read. You’ll also run into Tony Fadell – the creator of the iPod – somewhere in the story, as well as a public shouting match, and an almost fistfight, between the creator of the 65816 and Jean-Louis Gassée during San Francisco AppleFest in September 1989, right after Gassée placed the blame for the lack of a faster IIGS on the 65816’s design.
This is an evergreen article.
All this about the 65816 being faster clock for clock than a 68000, and motorola not having a 16mhz 68000 in 1985… Motorola already had the 68020 by then.
Woz’s proclamations about the 68000 in the interviews always felt a bit selective or meant to paint his project in the best light; and I hope that I conveyed that adequately. From his POV the 68020 is not a 68000; it’s a 68020—a much more complex (and more expensive) CPU in 1985. 😉 Not unlike how a 65816 is not a 6502, even though they’re compatible.
Also, the interview with Woz where he’s waxing poetic about the ‘816’s efficiency likely took place in the fall of 1984 if it was published in the Dec 84/Jan 85 issues of BYTE. Even with the ‘020’s June ’84 launch, it was supply constrained through much of 1985. Samples were inside Apple in 1984 because it was used to develop the Big Mac, I would think Woz would know about it. But as I’ve found with many Woz things, he likes to selectively quote or leave out statements or facts that might be inconvenient to him. I wasn’t able to determine when the interview took place, but considering the Phoenix development timeline and when WDC was actually shipping real silicon the fall is the most logical time. Even if the interviewer countered with the 68020, Woz probably would’ve countered back with price / target market.
An ‘816 can definitely compete with a plain 68000 in clock efficiency in some areas because of how it accesses memory versus a 68000. But the 68000 has many advantages over an ‘816 even if the ‘816 can sometimes beat it at the same clock speed. Woz was likely under the assumption that RAM would keep getting faster at the same rate as CPUs, something which was already proving to be wrong by 1984. The ‘816 is held back by its 8-bit databus, and the fact that to really wring out all that performance you have to write hand-tuned assembler.
The 68000 was inefficient only in terms of it having to use 2x the cycles to perform 32bit ALU ops. Since the internal bus was 16 bits. Whereas doing 16bit ops, would yield similar results to the 65k parts, since the data paths were pretty similar for those ops. Assuming the operands were already in the registers.
But at least you got a flat 32bit ISA with the 68000. Whereas the 65k was a 16 bit dead end.
Thom,
Thanks for the kind words and for enjoying this. I’ve been following OSNews for a while now and I was very surprised to see myself quoted on it. I appreciate the spotlight and hope you continue your excellent work.
And all in the name of selling the newest and most shiny thing…. Guess it is old Apple practice.
Thanks for this — I really enjoyed it. I have long suspected there must be more to this story.
Has anyone dug into the Apple Möbius? I’d _love_ to know more about that!