Arm vs x86 Performance

The battle between Arm and x86 architectures has been going on for decades in the CPU and microprocessor industry. With Arm based chips slowly making their way into PCs and servers, while x86 chips continue dominating desktops and data centers, many wonder how the two compare when it comes to performance.

In terms of raw processing power, x86 CPUs generally outperform Arm CPUs in benchmarks and real-world workloads. x86 chips have higher clock speeds, more cores, larger caches and superior single-threaded performance compared to most Arm designs. However, Arm CPUs are catching up with each new generation.

That said, Arm’s performance per watt is far superior to x86. Arm’s RISC architecture and smaller transistor size gives it a significant advantage in power efficiency over x86’s CISC design. This allows Arm chips to provide decent performance using very little power, making them ideal for mobile devices.

CPU Architecture

The key architectural differences between Arm and x86 that impact performance are:

• Arm uses RISC (Reduced Instruction Set Computing) whereas x86 uses CISC (Complex Instruction Set Computing). RISC has simpler instructions that execute in single cycles while CISC has complex instructions that take multiple cycles.
• Arm has a 32-bit or 64-bit instruction set while x86 supports 16-bit to 64-bit instructions for backwards compatibility.
• Arm has fixed length instructions (32-bit) whereas x86 has variable length instructions.
• Arm uses load/store architecture and has fewer general purpose registers while x86 has memory-memory architecture and more registers.
• Arm implements the ARMv8-A 64-bit instruction set while x86 uses x86-64 (also called AMD64).

Historically, RISC chips like Arm focused on simplicity, lower power consumption and higher clock speeds while CISC chips like x86 focused on doing more work per cycle. But modern implementations blur the lines between RISC and CISC.

Benchmarks

Benchmarks provide standardized ways to measure CPU performance. Some popular CPU benchmarks include:

• Geekbench: Tests integer and floating point performance using common algorithms and functions.
• SPECint: Tests integer compute performance with usage simulations.
• SPECfp: Evaluates floating point performance through simulations.
• JavaScript: Runs JavaScript benchmarks to test real-world web browsing.
• SYSmark: Uses tasks like video editing, office work, data analysis to gauge performance.

As per Geekbench 5 results, top-end x86 chips like Intel Core i9-12900KS (single core score: 2191, multi core score: 20000) clearly outperform the fastest Arm chips like Apple M1 Max (single core: 1879, multi core: 15353). But Arm is catching up.

JavaScript benchmarks used to test web browsing paint a similar picture. Chips like Apple M1 are faster than many x86 laptop chips but still trail behind desktop processors like AMD Ryzen 7 5800X.

In SPECint, x86 processors like AMD EPYC 7763 (score: 654) beat Arm server chips like AWS Graviton2 (score: 166). SPECfp shows comparable results. So in raw compute, x86 has a performance lead over Arm.

Design and Manufacturing Process

The performance differences between Arm and x86 arise from design choices and manufacturing processes including:

• Microarchitecture: x86 cores have more complex designs optimized for high single thread performance. Arm cores are simpler and more power efficient.
• Instructions Per Cycle (IPC): x86 can achieve higher IPC than Arm even at lower clock speeds due to pipelining and hyper-threading.
• Clock Speed: x86 CPUs often clock over 5 GHz while Arm chips clock under 3 GHz on average. Higher clocks translate to higher throughput.
• Core Count: x86 chips meant for servers and desktops have up to 64 cores while Arm chips max out at 32 cores for mobile usage.
• Process Node: x86 uses smaller transistors (5-7 nm) vs Arm (5-12 nm) allowing higher frequency and core count.
• Cache Memory: x86 has larger L1, L2 and L3 caches (over 100 MB) compared to Arm (32 MB). Bigger cache reduces memory latency.

So x86 has more performant cores, clocks higher, has more cores and larger caches – all contributing to superior multi-threaded and single-threaded performance over Arm.

Performance Per Watt

Although x86 CPUs are faster in terms of sheer processing power, Arm processors are much more power efficient providing more performance per watt. Some key reasons are:

• Arm’s RISC architecture reduces power consumption vs x86 CISC.
• Arm has smaller transistor sizes (5nm to 12nm vs x86 at 5nm to 7nm).
• Arm cores are simpler and smaller than complex x86 cores.
• Arm chips clock much lower than x86 allowing lower voltage operation.
• Idle power consumption is significantly lower on Arm vs x86.

For example, the Apple M1 chip offers similar performance to entry level x86 chips while consuming only 2-3 Watts compared to 15+ Watts on x86. This ability to provide adequate performance at minimal power is Arm’s strength.

Performance Segmentation

Here is a broad performance comparison between Arm and x86 chips segmented by use case and power consumption:

• For low power mobile devices (phones, tablets) – Arm dominates while x86 lags far behind.
• For laptops and ultraportables – Entry level Arm chips match basic x86 but high end x86 still leads.
• For mid-range desktops – x86 moderately outperforms Arm.
• For high end desktops and workstations – Top x86 chips crush Arm chips.
• For cloud servers – High core count x86 CPUs outmatch current Arm server chips considerably.
• For supercomputers – x86 rules this space though Arm is entering it.

So in mobile and extremely low power segments, Arm beats x86 hands down. In high performance computing, x86 still dominates over Arm. For mainstream laptops and desktops is where they closely compete.

The Road Ahead

Arm is rapidly improving its CPU performance and catching up to x86 with new architectural advancements. Here are some upcoming changes that can impact the Arm vs x86 performance equation:

• New Arm v9 architecture in 2022 with gains across the board.
• Higher core counts and SMT support in Arm server chips.
• Introduction of big.LITTLE technology in laptop processors.
• Higher power Arm chips optimized for desktops and HPC.
• Better software optimization for Arm including Windows 11 support.
• Migration to advanced 3nm or 2nm process nodes for Arm.
• Adoption of Arm cores in high performance computing initiatives.

While x86 dominates in raw compute power today, Arm is projected to greatly boost its performance across all computing segments in the next 5 years. This can make the Arm vs x86 battle even more competitive, especially in client computing devices and cloud servers where both architectures have a lot at stake.

Conclusion

The x86 architecture currently has superior per-core and multi-core performance compared to Arm CPUs. But Arm has an enormous lead in power efficiency and performance per watt which makes it ideal for mobile devices. As Arm catches up in higher power computing through architectural improvements, the rivalry between x86 and Arm is likely to heat up even more.