When choosing between ARM and x86 architectures, power consumption is an important consideration. ARM processors are generally more power efficient than x86, making them better suited for mobile and embedded devices where battery life is a key factor. However, x86 processors offer greater performance, especially for desktops and servers where raw computing power is more critical.
In a nutshell, ARM uses less power but offers lower performance, while x86 uses more power but provides greater performance. The tradeoff depends on the specific application and priorities around power, performance, and cost.
Key Differences in Power Consumption
There are several key differences between ARM and x86 that impact power consumption:
- Instruction set architecture (ISA) – ARM has a RISC-based ISA optimized for power efficiency, while x86 uses a more complex CISC ISA.
- Manufacturing process – ARM processors are often built on smaller, more power-efficient manufacturing processes.
- Processor architecture – ARM utilizes in-order execution pipelines which use less power than x86’s out-of-order pipelines.
- Workloads – ARM targets mobile and embedded workloads which require less overall compute power.
- Power management – ARM has extensive power-saving features for idle and low-power states.
RISC vs CISC
ARM uses a Reduced Instruction Set Computing (RISC) architecture which has simpler, more power-efficient instructions compared to x86’s Complex Instruction Set Computing (CISC) design. RISC uses fewer transistors per instruction, reducing switching power consumption. The simplified instructions also allow for an easier pipeline without complex decoding logic.
ARM processors are typically manufactured on smaller, more advanced process nodes than desktop/server x86 chips. For example, flagship ARM chips are made on 5nm or 7nm nodes while high-end x86 CPUs are still on 10nm or 14nm nodes. The smaller transistors switch faster and leak less power.
ARM utilizes in-order execution pipelines that prevent instructions from being reordered. This simplifies the logic and reduces power consumption. In contrast, x86 uses out-of-order pipelines that enable reordering instructions for higher performance, but require more complex logic and power draw.
ARM processors primarily target mobile devices and microcontrollers running simple OSes and applications. These workloads don’t require the raw CPU horsepower needed for desktops/servers, so ARM chips can get away with lower clock speeds and overall TDPs. x86 must support more intensive workloads and higher performance at the cost of greater power consumption.
ARM implements extensive power management capabilities including clock gating, power gating, dynamic voltage/frequency scaling, idle states, and sleep modes. This allows ARM chips to conserve power in periods of low utilization. While x86 has power-saving features, the general-purpose nature of the architecture prevents the same extent of optimizations.
Benchmarking ARM vs x86 Power
When comparing the actual power consumption between ARM and x86 processors, ARM generally has a significant advantage – often consuming just 10-30% as much power as an x86 chip for the same workload. However, there are some caveats:
- Testing conditions – Factors like manufacturing process, clock speeds, and TDP settings can affect results.
- Idle power – ARM has major advantages in low utilization idle states.
- Load power – The power gap narrows under sustained heavy workloads.
- Performance – x86 completes work faster despite using more power overall.
While the above generalizations hold true, real-world power consumption depends heavily on the specific ARM and x86 processors being compared. For example, the M1 ARM-based chip has performance and efficiency rivaling entry-level x86 laptop chips despite being fanless and low power. But comparing a smartphone ARM chip to a 140W desktop Intel CPU is not apples-to-apples.
ARM vs x86 Power Examples
Some example comparisons that demonstrate the power efficiency advantages of ARM include:
- The Cortex-A53 ARM CPU core uses 75% less power than Intel’s Atom x86 core at the same performance level.
- The Exynos 5422 ARM SoC uses 70% less power than the Intel Core i5 Haswell x86 while benchmarking similarly.
- The Snapdragon 835 ARM mobile SoC provides 27% better energy efficiency than the Intel Skylake Core i5 desktop CPU.
- The Apple M1 ARM SoC delivers over 2x better performance per watt than comparable entry-level x86 laptop chips.
These examples reaffirm that ARM requires substantially less power for light workloads. However, maximum throughput performance and sustained loads still favor high-end x86 like Intel Core i7/i9 and AMD Ryzen. It depends on the priorities around power, performance, and workload types.
ARM Power Efficiency Technologies
In addition to the inherent architecture advantages, ARM employs multiple technologies explicitly aimed at improving power efficiency including:
Combines low-power LITTLE cores with high-performance big cores. Workloads can be scheduled intelligently to maximize either performance or battery life.
Unused parts of the die are power gated off to reduce leakage. Especially beneficial for mobile SoCs.
Adaptive Voltage Scaling
Dynamically adjusts CPU voltage and clock speed based on workload. Improves energy proportionality.
Reduces power wasted synchronizing multiple clock domains. Components only consume active power when needed.
TLB Power Optimizations
Reduce Translation Lookaside Buffer (TLB) power through partitioning, filtering, and looking up only active entries.
Compresses unused memory to optimize power. Similar concept as dark silicon for memory subsystems.
These technologies allow ARM processors to scale down to extremely low power and idle states. However, peak throughput still favors x86 for unconstrained workloads.
ARM in Data Centers
The superior power efficiency of ARM is driving adoption in data centers where electricity costs are a major operational expense. For example:
- Amazon EC2 now offers ARM-based Graviton instances claiming up to 40% better price/performance over x86.
- Fujitsu began deploying ARM-based A64FX chips in supercomputers like the Fugaku system ranked #1 in TOP500 claiming 2.8x better power efficiency than x86.
- Microsoft Azure offers ARM-based virtual machines for running scale-out workloads more efficiently.
While ARM is still a tiny fraction in data centers today, its lower TCO is gaining appeal for modern scale-out workloads. However, barriers around software ecosystem maturity remain.
x86 Power Optimization Technologies
Intel and AMD have also incorporated power-saving enhancements in x86 including:
- Enhanced SpeedStep – Dynamically adjusts voltage and frequency.
- Intel Turbo Boost – Increases clock speed on active cores.
- AMD Cool’n’Quiet – Cuts power during low utilization.
- Power gating – Turns off idle CPU sections.
These help reduce x86 power consumption, but inherent architecture advantages allow ARM to push efficiencies even further. But again, unconstrained peak throughput still favors high-end x86.
Ultimately, choosing ARM vs x86 depends heavily on workload types and performance requirements. Some guidelines:
- Low power mobile – Favors ARM for best battery life.
- Desktop/gaming – x86 preferred for maximum performance.
- Web serving – Light workload well-suited for ARM efficiency.
- High-performance computing – x86 has bandwidth advantage for number crunching.
- Machine learning inference – On-device ARM inference gaining traction.
Of course, there are exceptions like Apple’s ARM-based M1 successfully entering the laptop market. And server-class ThunderX ARM chips rivaling x86 on throughput. But in general, picking ARM or x86 should depend on assessing power, performance, and workload needs.
ARM processors clearly consume less power than x86 – often an order of magnitude difference for the same workload. This makes ARM ideal for mobile and embedded use cases focused on energy efficiency. However, x86 maintains advantages in raw throughput that make it suitable for performance-intensive desktop and server workloads. The choice depends heavily on properly assessing power consumption needs versus performance requirements for the target application and environment.