While the iPad 3 graphics are clearly unbalanced, relative to its predecessor, that was hardly a mistake. Rather it was a conscious choice, as Apple’s engineers had very little in the way of alternatives. The A5X is a 165mm2 die, which is absolutely enormous; the GPU alone is a bit over 61mm2. To give some context, the 22nm quad-core, GT2 Ivy Bridge is 160mm2 with around 40mm2 for the GPU, while the 40nm Tegra 3 inside the Surface RT is around 80mm2.
A chip the size of the A5X is rather expensive to produce, especially when it is sold inside a $500-$830 tablet. Doubling the GPU area on the A5X would maintain the same level of visual quality as the previous generation for the larger display, but would result in a cripplingly expensive 220mm2 chip. More importantly, even if die area was not a constraint, the power consumption would be prohibitive. The 42.5WH battery for the iPad 3 is larger than the one found in the MacBook Air, and even with that, the battery life was somewhat of a sore point. Doubling the GPU performance would make matters considerably worse and increase total power consumption by ~2W under heavy loads, perhaps even requiring fans to maintain appropriate skin temperatures.
Conveniently for Apple, the solution to the problem was right around the corner, courtesy of Moore’s Law. The A5X was manufactured in a fairly old process by 2012 standards; Samsung’s 45nm with conventional SiON gate dielectrics. To put this in perspective, AMD and Nvidia were already shipping high-end 28nm GPUs fabbed by TSMC as early as January 2012, although not in particularly high volumes.
Samsung’s 32nm high-k/metal gate process came online shortly after the release of the iPad 3, offering better density and substantially improved power consumption. The A6X takes advantage of the additional density to pack in a higher performance GPU; the SGX554MP4 has roughly twice the FLOPs/cycle as the previous generation SGX543MP4. Apple’s 32nm implementation reaches 28% higher frequencies than its predecessor, delivering a 2.5× overall improvement. That is enough to improve the FLOPs/pixel by around 33% compared to the iPad 2, and restore the historical trend of increasing visual quality.
Not only did the A6X offer substantially better graphics performance than the previous generation, but the die size is also considerably smaller, at around 123mm2. The high cost and power consumption of the A5X was undoubtedly one of the reasons why Apple opted to discontinue the iPad 3 a mere 7 months later (as soon as the A6X-based iPad 4 launched). While many dismiss the importance of process technology, this demonstrates the relevance for modern mobile devices. That Apple discontinued the iPad 3 so swiftly is nothing less than a tacit acknowledgement that high resolution displays were adopted early and at considerable cost.
For most companies, waiting for the next process node would have been a far better choice. However, Apple is in a relatively unique position. The company dominates the tablet market and has incredibly high margins, but is facing an increased threat from the Android and Microsoft ecosystems. The Retina display was an excellent way to extend product leadership through innovation, and pre-emptively ward off competitors (e.g., the Nexus 10). The downside is that Apple had to be willing to accept lower margins on the iPad 3. Since Apple’s margins are already quite hefty, that is an uncomfortable proposition, but certainly not impossible.
In contrast, other companies with lower margins and volume might not be able to sustain the increased costs associated with a more powerful GPU, not to mention the higher quality and more expensive display. While high PPI displays are not essential to succeed in the tablet market, the omission is certainly a competitive disadvantage for premium products. In some sense, Apple’s competitive position (and superb marketing) allows them to more aggressively pursue innovative technologies for mobile devices that may not be economically feasible for other companies.
Whether Apple’s decision to move to adopt a high resolution display for the iPad 3 was premature is open to debate. The results are self-evident though; the iPad 3 was the first tablet to market with a Retina display by over 6 months. More importantly, this is an excellent example that highlights the critical importance of selecting the right technology and ensuring a balanced system. Had Apple delayed, it is quite possible that the Nexus 10 would have been released first and chipped away at the perceived leadership of the tablet market. On the other hand, the early adoption meant that the graphical quality of some applications was compromised by an under-powered GPU and overall battery life suffered. One of the advantages of leading companies like Apple, IBM, Intel, and Microsoft is that they can afford to be early adopters and take such risks, but the downside is that not all risks ultimately payoff. This sort of trade-off is rife within the industry and crucial to keep in mind when contemplating new products and new technologies.
Discuss (132 comments)