The M1 is a revolutionary achievement and the man responsible for it is a Senior VP at Apple whom you have probably never heard of
Few tech leaders are as unassuming as Johny Srouji. In fact, when his name cropped up last year alongside the launch of Apple’s first M1 powered machines, there was very little to go on about him. Yes, we knew that he and his team played a major role in Apple’s iPhone chips, but the announcement of the hardware revolution that is the M1 caught everyone off-guard.
Now, the man himself is as reticent as can be and has chosen to avoid the limelight and media glare that come with being part of the senior leadership team at Apple. Maybe it is the nature of the work that Srouji did that kept most tech enthusiasts almost—for lack of a better word— uninterested. Yes, Apple has been steadily improving its smartphone processors, but the incremental updates every year can only be called evolutionary. The fact that even the current flagship Android devices are no match to last gen iPhones made going into much detail about the processors or the people behind them seem almost redundant. But then, 2020 rolled around…
In the midst of a global pandemic, Apple made an announcement that would easily go down as one of the most monumental moments in its history. The first M1-powered machines, including a new Macbook Air, a 13-inch Macbook Pro, and a Mac mini, were announced. The devices, which looked exactly like models before them, were claimed to be significantly more powerful and efficient. The 13-inch Macbook Pro, for instance, claims a staggering 20-hour battery life on a video playback loop. Most in the tech media felt that the claims that Apple made were just that—claims—and wouldn’t hold up in real world testing, with reputed reviewers like Linus Tech Tips casting serious doubts over the announcement. But they were proven wrong when the machines arrived and were put through performance benchmark and battery cycles.
But what is actually at the root of this seemingly incredible leap? A leap of the magnitude that hasn’t been witnessed in mobile computing for decades. To understand that, we need to look at the brand-new 5nm process-derived M1 built on an ARM architecture—the same architecture that has been powering smartphones for a very long time. For more than a decade or so, an imaginary line was drawn, demarcating the devices that chip architectures would power—for quite long, that demarcation meant that x86 processors powered laptops and desktops and ARM processors powered smartphones and tablets. The reasons were fairly simple to comprehend: ARM processors couldn’t deliver the sort of performance that laptops/desktops required, while the lower power available to smartphones and tablets (due to size and portability constraints) meant that ARM processors were the go-to. Now, it wasn’t that other manufacturers hadn’t tried their hands at ARM powered computers, but just that the resultant devices weren’t quite successful.
In 2012, Microsoft introduced to the world the Surface RT—an ARM based device that ran Windows RT; a stripped-down version of Windows 8. It struggled to run a number of popular apps that other Windows systems could. It was a big bet that backfired for Microsoft, who eventually had to write-off $900 million when the plug was pulled on the project. That wasn’t Microsoft’s last attempt at an ARM powered machine, though. The Surface Pro X had a processor co-developed by Microsoft and Qualcomm, but it still struggled with app compatibility and performance issues. Microsoft isn’t the only company that has tried this: Lenovo and even Samsung have made attempts but with little success.
Apple, however, has done something that is markedly different, and Srouji is to be credited for that. An Israel-born engineer by training, Srouji joined Apple in 2008 after stints at Intel and IBM. He led the development of the A4 in 2010—Apple’s first SoC, designed completely in-house. First brought in by Bob Mansfield, Steve Jobs’ trusted lieutenant, Srouji was met with a 40-member team that put together components from other manufacturers. It soon grew to 150 with Apple’s acquisition of P.A. Semi. Srouji would then go on to transform Apple’s chip division.
The A4 powered the iPhone 4 and was essentially the first step in the development of Apple Silicon. Multiple iterations followed and Apple consistently worked on improving the integration and, by extension, performance and efficiency of its SoCs. The improvements helped add additional features that worked seamlessly. In its early years, the iPhone was mocked for being behind the curve when it came to features—the argument being that Android simply had more to offer. But Apple was restrained and focused on building a cohesive experience that just worked flawlessly. The tight integration of hardware and software meant that seemingly underpowered specifications on paper delivered peerless performance in the real world. While Apple Silicon was already helping iPhones leapfrog the competition, the arrival of the A7 took things up a notch: It was the first smartphone chip with 64 bits. The rest of the competition with 32-bit processors was caught napping.
While the iPhone was bulldozing competition, the iPad wasn’t too far behind. Starting with the third gen, iPads would employ custom-designed chips built by Srouji’s team. When Apple first announced its intention to use ARM chips for their Macs, it was widely believed that Apple would do something similar to what it has done before and employ a modified A14. They were partly right. Yes, the A14 proved to be a major piece in the M1’s development, but, as Srouji has now clarified, it is actually a superset of the A14, rather than a mere modification.
But how did Apple make the jump from increasingly superior iPhone processors to the M1? More than a decade of tweaking and improving on the processors used in the iPhone meant that Apple already had an extremely powerful and efficient subset to expand on. So, when the decision to move forward with making their own silicon was made, Apple was actually better positioned than most would think. It has been pointed out that Intel’s slow growth rate in performance and functionality in their newest chips has possibly led to Apple making the switch to the M1. But that can hardly be the case.
Apple, as has been revealed now, has been working on the M1 for quite a few years. It all started when a handful of Apple engineers, including Srouji, got together to rip out the innards of old Macs and used custom fabricated chips based on processors in the iPhones instead. It is believed that this happened sometime around Srouji’s appointment as Senior VP, Hardware Technologies, in 2015. For a company with a multi-billion dollar R&D budget, this was reminiscent of the early days of Apple with Steve Wozniak designing the circuitry of the first Apple machines in a garage. But that’s where the similarities end: it might have started off as kind of an experiment, but billions were sunk into the project; after all, Apple Silicon was the only way that Macs could have the tight integration that the iPhones are so renowned for—something Jobs tried with PowerPC but couldn’t accomplish, eventually ditching it for Intel’s then far superior processors. Apple Silicon has given the company control over the last missing piece of the puzzle.
Apple’s recent announcement of M1-powered iMacs and iPad Pro means that instead of pure cross platform software interplay, there is now a much deeper hardware-level connection. Srouji has revealed that analyses of Mac application workloads, graphic/GPU capabilities that were required to run a typical Mac workload, even the number of cores plus the ability to drive Mac-sized displays, all reportedly played a part in the M1’s design. He claims that engineering an SoC for such functionality meant that the M1 had to be a superset of the A14, and all of these functionalities were leveraged in components of the SoC that Apple has been perfecting for over a decade in iPhones and iPads. The CPU, the GPU, the Neural Engine, and the ISP are now refined because of Srouji and his team’s efforts over the last decade.
One key aspect of the M1’s design that makes it possible to attain such performance and efficiency is the use of Unified Memory Architecture (UMA). Engineering all of the components around the UMA means that the CPU, the GPU, the Neural Engine and Image Signal Processor have access to a pool of very fast memory positioned very close to them on the same chip instead of discrete memory being assigned to different blocks. This makes for an extremely efficient, tightly controlled SoC where very little time or efficiency is sacrificed in copying or moving around data. Another reason why this can be more efficient as opposed to traditional x86 processors is the fact that a number of applications these days use computational/graphic rendering pipelines that employ a hybrid of compute, GPU rendering, ISP, and much more. UMA proves to be extremely versatile and efficient in such use cases—an example is the camera application and the post-processing involved. It goes without saying that there are a ton of other processes that benefit from UMA as well. This ARM SoC with UMA also means that the power consumption is measly, as there is very little energy expended in moving stuff around.
The benchmarks have confirmed that for single-core workloads, the M1 obliterates the competition and is only challenged by high-end machines that often cost far more. In multi-core tasks too, the M1 holds its own. Sure, Intel has now questioned the performance of the M1 with an assortment of cherry-picked benchmarks that showcase the x86’ supremacy while dealing with demanding workflows. Intel, however, conveniently fails to mention that while the company has a bevy of processors suited to an assortment of applications, the M1 is only the first (and possibly least powerful) in a line of new processors from Apple; the news of an imminent M2 has certainly spurred interest among enthusiasts. Intel also fails to mention that a high-performing Mac can now be had at a price point that’s much lower than a similarly powered Windows device with an Intel processor— something that was hard to believe before the M1 was launched.
The presence of Rosetta 2 on macOS solves a problem that was endemic to previous non-Apple attempts at ARM based laptops—app compatibility. Rosetta 2 essentially translates all apps made to run on Intel systems without a significant loss in performance. This will hold the Macs in good stead till a library of ARM based applications is created. Additionally, iPhone and iPad apps can now run on Macs. All this points towards a future where all Apple devices have a uniform architecture and apps made for one device could seamlessly run on other devices as well. Evidently, that level of integration and cohesion is the stuff of dreams, but Apple, more than any other company, seems better placed to turn it into reality, and Srouji and his team will have a massive role to play in that.