After years of slowly closing the gap between itself and Intel, Apple can fairly claim to have slashed the difference between its ARM CPUs and Intel’s x86 chips to ribbons — at least in one test. Geekbench 4 reports an iPad Pro with a single-core score of 5030 and a multi-core score of 17995. This compares well against the Apple MacBook Pro — but not necessarily as well as you think.
When you actually look at specific test results, there are areas where the Intel CPU is dramatically faster than the ARM equivalent. I’m not trying to imply that John Poole, the author of Geekbench, weights his tests incorrectly or that there’s anything untoward about the benchmark. Complex tests that return a single score always perform weighting under the hood to make certain no one test skews the results. Geekbench 4 is designed to capture an overall picture of SoC performance rather than highlighting just one metric.
But these subtleties of testing are going to become more important as ARM and Intel approach parity in some power envelopes. The general Geekbench score predicts just an 8 percent single-threaded gap between the eight-core A12X Bionic (four high-efficiency cores, four low-efficiency cores) and the four-core, eight-thread Core i7-8559U. This elides the fact that the 8559U is more than 2x faster in certain tests like SGEMM that clearly aren’t weighted very much and 1.72x faster in SFFT. Intel remains 1.2x faster in LZMA and 1.27x faster in JPEG decoding, and 1.46x faster in PDF rendering.
I’m not trying to take away from Apple’s accomplishments here — there is simply no arguing that Apple’s Axx SoC family has been on a tear these past five years. I’m not sure any other company in the world has done as much to boost its single-threaded CPU performance over the same time period as Apple has. Intel, in contrast, hasn’t released a new consumer CPU architecture since 2015. The changes made to Kaby and Coffee Lake have increased clocks and added cores, but they didn’t make any major architectural changes.
Results like this draw attention because they feed a narrative that Intel is losing ground to Apple. There are persistent rumors that Apple will start swapping Intel CPUs for its own silicon in 2020. From there, it’s easy to connect the dots and think that this is evidence of Intel’s own performance collapse, the end of x86, etc. Digging deeper into results often gives a more nuanced picture of what’s going on and where the limits and problems are. For example: One potential reason these results favor Apple is that Apple is still building its laptops with DDR3-2133, while its iPads use LPDDR4 at higher clocks. In theory, a laptop with DDR4-2400 instead of DDR3-2133 would perform a bit better in these tests.
If Apple wants to truly take the general-purpose CPU performance crown away from Intel by 2020 and replace x86 silicon with its own ARM chips, it’s going to have to either improve those areas of performance where it still lags far behind its competitor or say goodbye to the community of Mac users that rely on superior performance in those types of mathematical operations. That’s going to cost the company power and die area at some level. This is by no means an insurmountable problem — it’s more-or-less exactly what Intel did when it transformed its Pentium M Dothan core (2003) into Nehalem (2008). Dothan was a great CPU with some multimedia processing weak spots compared with its predecessors. Over time, Intel fixed those weaknesses and added new capabilities, setting the stage for a brand-new architecture to debut a decade ago.
The other major issue Apple will have to continue to work on is the suitability of iOS as a serious work platform. iPad Pro reviews have always praised the tablet for its build quality and performance. The question of whether you can use it as a replacement for a traditional laptop (including a Mac laptop) has always come down to software support and ease-of-use. One can make a pretty good argument that Microsoft’s Surface tablets are a more effective way of combining a work and entertainment device into a single tablet than the iPad Pro has historically been, and to the extent that this is true (I think there’s a pretty good argument for it), it’s because Windows 10 has done a better job of planting a foot in both camps. The flip side to that, of course, is that iOS did a much better job of being a tablet OS than Windows 8 ever was.
My point here is not to diminish the success of the A12X or Apple’s design teams, but to explore the larger context around the achievement. There are still areas of noted performance difference that Apple will have to address if it wants to continue to appeal to certain types of users while swapping to its own silicon. iOS is going to need to evolve if Apple is serious about shifting to a new paradigm in which that OS sits as an equal multi-tasking and work-capable partner next to OS X. I don’t doubt that the A12X Bionic is one hell of a fast chip, but speed alone isn’t going to create the kind of paradigm shift Apple is apparently gunning for.
VIA Technologies, Zhaoxin Strengthen x86 CPU Development Ties
VIA and Zhaoxin are deepening their strategic partnership with additional IP transfers, intended to accelerate long-term product development.
How L1 and L2 CPU Caches Work, and Why They’re an Essential Part of Modern Chips
Ever been curious how L1 and L2 cache work? We're glad you asked. Here, we deep dive into the structure and nature of one of computing's most fundamental designs and innovations.
Every CPU, GPU, and Console Debut This Fall Was Effectively a Paper Launch
Every CPU, GPU, and console launch since midsummer has effectively (if not technically) been a paper launch for the majority of consumers who wanted the hardware.