The golden gigaflop: Apple’s shrinking performance sweet spot
The gigaflop is a fun name for a unit of measurement. It’s not hard to imagine Back to the Future’s Doc Brown using the word to describe a nightmare scenario of destroyed timelines and endless bad sequels: “This could cause a temporal shockwave of several gigaflops!” (He would pronounce it “jigoflop”, of course.)
In actuality, the flop stands for “floating-point operation”, and a billion of them per second makes a gigaflop. When talking flops, we’re measuring the theoretical maximum number of computations that a hardware design can execute within one second. This is not a very good way to measure CPU performance because it completely ignores real-world considerations such as “what are those operations actually doing?” and “how do we get that much data into the CPU anyway?”… But flops are easy to count, so they’ve stuck around for decades.
At one point, Apple liked gigaflops so much that they ran an ad campaign around it:
You don’t see Apple talking about gigaflops anymore. These days the flop kings are GPUs. NVidia’s current crop of high-end graphics processors can do over 6,000 gigaflops. (Yes, the graphics card is literally thousands of times faster than the late-1990s professional Mac that proudly advertised its flops… But the caveat is that unleashing this performance requires software to be adapted to the GPU model, often a difficult task.)
When gigaflops originally became a hot item and not just a theoretical speculation, ties were in fashion, red leather was cool, and the mythical gigaflop machine was large enough to double as a couch:
The Cray computer cost around $14 million in 1986. Apple bought one of these, for reasons that are not entirely clear (however the rumor that Steve Jobs just walked in at Cray to buy the machine is apparently false).
By the mid-90s, a server computer equipped with multiple DEC Alpha CPUs could manage a gigaflop. A couple of years later in 1999, the Power Mac G4 (shown in the above ad) finally brought Apple’s hardware into gigaflop territory.
But what has happened since? Turns out that the one-gigaflop computer isn’t a relic of history. Instead Apple has released several iterations of computers roughly in this same performance range, but with constantly shrinking physical volume:
The 1999 Power Mac G4 was about 40,000 cm³ (2,440 cubic inches).
The 2015 Apple Watch is about 14 cm³ (5/6 of a cubic inch).
Yet, in some sense, they are the same computer. It’s as if Apple starts out every new product at roughly the same performance level, then lets the product’s performance grow according to Moore’s Law. Because everything needs a brandable name, I’ll call this Apple’s golden gigaflop.
Tech sidenote: I’m being loose with the hardware specs. The high-end G4 models actually could do several gigaflops, while the very first iPhone didn’t reach one. The “golden gigaflop” is not a specific measurement, but more like a performance category and a design target. (Also, “golden” monikers have a history of being somewhat oversold. Despite what popular science would like you to believe, the golden ratio is not really the foundation of all classical art; neither is the golden rule easily applicable in all aspects of one’s life. I’m happy to have the golden gigaflop join this suspicious group.)
Let’s plot out the gigaflop’s volume reduction over time:
The iPad makes a somewhat distracting bump in that chart — but that’s because it was released “out of order”. It actually neatly fills out the slot between the PowerBook and the iPhone, as we can see in this reordered plot:
Looking at those intervals, there’s room for one more future product, possibly even two. Here they have been added to the plot:
“Product 1" would have a volume of about 2.5 cm³. If you built a thin wrist-worn wearable that’s slightly smaller than the 38mm Apple Watch but only 4mm thick, that would be it. An “Apple Watch Air” perhaps?
“Product 2” is more speculative with a volume of .5 cm³. That’s closer in size to a fingernail. Or perhaps a ring? Something sewn into clothing? A skin graft, even?
Curiously, this graph looks substantially similar to those found in Clayton Christensen’s seminal book The Innovator’s Dilemma. In the first part of the book Christensen studies the 1980s disk drive industry, which sounds tremendously boring but was actually a cut-throat high-tech business where companies achieved almost overnight success, IPO’d in a year or two, only to be displaced just as quickly by new competitors. (By the way, this is the book that originally unleashed the abused term “disruption” on us.)
A key element of this competitive landscape was new disk form factors. Companies that built their success on 8" drives generally failed to capture the market when 5" drives emerged. The same happened again at 3.5" and 2.5" transitions. At each step, the smaller drive size started out slower and with less capacity, which made it unappealing to the large disk drive company with existing customers who would generally ask for faster disks and more capacity. But those smaller disks could be sold to new customers, and eventually they would catch up with the previous generation in performance so there wasn’t any reason left to buy the old big thing.
For those disk systems, the sweet spot to kickstart a new generation was somewhere around 20 megabytes. At that capacity the new smaller disk became useful to new applications. (20MB was quite a lot of space in the eighties!)
Apple’s golden gigaflop serves a similar purpose — it’s a “minimum viable disruptive computer”. And of course there’s one big difference from the disk drive market Christensen examined, as Apple intends to do the disrupting themselves rather than waiting for a competitor to come along. Does that mean the Apple Watch is disruptive? Not necessarily. Not all the steps on the shrinkage plot have worked equally well; the Power Mac G4 Cube was an outright failure. But to find out which steps work, Apple may need to build the product rather than let someone else do it.
