An Irrational Rant on ISAs

I want to take some time to address something that has bothered me about online discussion of CPUs for a while now: the widespread conflation of ISA (Instruction Set Architecture) and microarchitecture. It’s not uncommon to hear statements like “ARM is so power-efficient” or “x86 still has the highest performance.” These claims misattribute characteristics like power efficiency and performance to the ISA when, in reality, these traits are primarily determined by microarchitectural decisions.

To understand why microarchitecture and process node dominate performance characteristics, let’s break it down.

What is an ISA, and What Does It Do?

The ISA is the processor's hardware-software interface. It defines how software communicates with the processor, specifying the instructions the CPU can execute, their encoding, and the behavior of the hardware when those instructions are executed. Popular ISAs like ARM, x86, and RISC-V are essentially rulebooks for this interaction.

The ISA does influence design in some ways—x86’s small number of temporary registers, for instance, poses unique challenges for compilers and hardware designers. However, these quirks are relatively minor compared to the broader impact of microarchitectural decisions.

Microarchitecture: The Real Performance Driver

The microarchitecture is how the processor executes the ISA. It encompasses the design of the CPU’s pipeline, caches, branch predictors, execution units, and more. Here’s why this matters:

1. Instruction-Level Parallelism (ILP)
   Performance is heavily influenced by how well the microarchitecture can execute multiple instructions simultaneously. Features like superscalar execution, out-of-order scheduling, and speculative execution enable ILP. For example:

   • Apple’s M-series cores achieve high performance by incorporating sophisticated ILP features like wide pipelines and large reorder buffers.

   • ARM V1 and V2 cores also prioritize performance with features supporting parallelism, at the expense of power and die area.

2. Power Efficiency
   Power efficiency is dictated by how a design balances compute capability, clock speed, and voltage. The perception that ARM is “inherently efficient” often stems from ARM cores being used in designs optimized for low power consumption, not the ISA itself. Contrast this with Intel’s Lunar Lake, an x86-based processor whose “little-core” delivers ARM-like power efficiency through careful microarchitectural trade-offs.

3. Process Node
   The semiconductor process technology—measured in nanometers (e.g., 5nm, 3nm)—plays a critical role in performance and efficiency. Smaller nodes allow for higher transistor density, reduced power consumption, and increased clock speeds. Apple’s M1 and M2 processors benefit from TSMC’s advanced process technology, which amplifies the impact of their microarchitectural decisions.

The ISA-Microarchitecture Mismatch

To illustrate the divergence between ISA and microarchitecture, consider this thought experiment:

• Take an ARM ISA and implement it on a very basic microarchitecture without superscalar execution, minimal cache, and no out-of-order capabilities. The result? A processor far slower than any high-end x86 chip.

• Conversely, implement x86 on a state-of-the-art microarchitecture using advanced process technology and you’ll achieve high performance despite the supposed inefficiencies of the ISA.

The reality is that the ISA is more of a means to an end—a contract for software compatibility—than a defining factor in performance or efficiency.

Why the Confusion?

The conflation of ISA and microarchitecture likely arises from branding and historical context:

• ARM processors are frequently associated with mobile and embedded systems, where efficiency is paramount, creating a perception of inherent power efficiency.

• x86 dominates high-performance desktops and servers, leading to its association with peak performance.
  These are product and design choices, not intrinsic properties of the ISAs.

Conclusion

While the ISA provides the foundation for hardware-software interaction, the performance characteristics of a processor—whether it’s power-efficient, high-performance, or somewhere in between—are overwhelmingly dictated by microarchitecture and process node. Recognizing this distinction is key to understanding modern CPU design and avoiding oversimplified comparisons like “ARM is better than x86.”

In short, the ISA is the starting point, but it’s the microarchitecture and the implementation that defines where the journey ends.