There are many histories of the Mac, and no doubt as many futures too. Here I’d like to consider how it has changed at the heart, the hardware in the middle of that case, and where it’s going in the M1.
The Motorola 68000 processor inside the original Macs is just that: a processor. It didn’t even contain memory management sufficient to support the use of virtual memory. In the first 128K Mac of 1984, it ran at a speed of a mere 8 MHz. Three years later, the Mac II featured a second-generation Motorola 68020 running at twice the speed and with a separate co-processor, the 68881 FPU, for performing floating-point calculations. If you wanted to run A/UX, Apple’s implementation of Unix, then you also needed its optional memory management unit, in effect a third co-processor, to support virtual memory. It required a graphics card to drive a colour display.
One of the most powerful of the Classic Macs was the Quadra 660AV from late 1993, almost a decade after the 128K. This featured a Motorola 68040 processor running at a dizzy 25 MHz, with integral floating point and memory management units, and a digital signal co-processor to handle its AV enhancements and display output.
In March 1994, Apple started the transition to PowerPC processors, which it had been jointly developing with IBM and Motorola. Top of the range then was the Power Mac 8100, with a PowerPC 601 processor running at up to 110 MHz, complete with a built-in floating point unit, and a built-in graphics card. Eleven years later, the Power Mac G5 offered a top specification of two two-core PowerPC 970MP (G5) processors giving a total of four cores, with a processor speed of 2.5 GHz. It required a separate graphics card.
In January 2006, Apple switched processors again, to Intel Core Duo models. Its first desktop, an iMac, offered a single two-core processor running at 1.83 GHz, and a built-in graphics card. Later that year, Apple introduced its first Mac Pro, with up to two two-core Xeon 5100 processors running at 3.0 GHz. This required a separate graphics card.
Apple’s last high-end Mac to be based on Intel processors is likely to be its current 2019 Mac Pro. This uses Intel Xeon W processors with a total of 8 to 28 cores running at up to 4.4 GHz. It too requires a graphics card.
The first and low-end M1 Macs released in late 2020 have four Firestorm high-performance cores, and four Icestorm high-efficiency cores, running at a speed of up to 3.2 GHz. Integrated into the same chip is an 8-core GPU, a Neural Engine, an image signal processor, a Secure Enclave, and Apple Fabric, which is mainly concerned with I/O, including an SSD controller and Thunderbolt 4 controllers. Contained within the M1 package is 8 or 16 GB of LPDDR4X SDRAM memory.
Processor speeds increased steadily in the 21 years from 1984 to 2005, but have increased more slowly over the last 14 years, during which the number of processor cores has risen from two to eight or more. Following early system integration between 1984 and 1988, when external maths co-processors and memory management were brought into the processor chip, increase in the range of supporting systems and their integration didn’t really start until 2020, with the M1.
The popular myth that all that has happened with processors is that they’ve got faster and packed more transistors into a smaller space is an oversimplication which readily leads to the conclusion that modern computers are reaching the end of this evolutionary process. Yes, the 68000 had only 68,000 transistors against the M1’s 16 billion, but it’s also about what you do with those transistors.
One clear break with the past is the M1’s assimilation of features which have previously been relegated to other chips, often designed and made by a different vendor, and cobbled together with software drivers. GPUs, now used for much more than just processing graphics, have traditionally had their own limited high-speed memory, and in some cases (eGPUs) are external to the Mac itself. One sign of this integration in the M1 is its use of unified memory, which eliminates the need to transfer data from main (CPU) memory to GPU memory, or the reverse. Unification includes the Fabric supporting interfaces for internal storage and external devices connected via Thunderbolt.
There’s an interesting biological parallel (not analogy) here in the evolution of organs within animals. Around 750-800 million years ago, the first Metazoans emerged, with cells specialised into different roles. These came to form organs with specific functions, such as the heart and kidneys, which have enabled animals to function very differently and adapt to a much wider range of environments and roles.
Apple Silicon firmware and macOS no longer have to support the vagaries of different chipsets, or to use a wide range of different interfaces to peripherals. Hardware, firmware and operating system can be designed together at last, in a unique unification. Apple’s Neural Engine is one relatively recent example of what can be achieved. This apparently has 16 cores primarily for accelerating tensor and matrix learning operations at a speed of up to 11 trillion operations per second, and is supported through Core ML software. The end result can comfortably outperform the M1’s CPU and GPU, and simply doesn’t exist in the Intel architecture.
In many disciplines, the safest way to predict the future is to go for more of the same. That hasn’t happened for processor speed, and there are clear limits to the density of transistors. For Apple, the future is very different, and in Macs starts with the M1.