Why Modern Processors Are Increasing Cores Rather Than Clock Speeds

The evolution of processor design has seen a shift from increasing clock speeds to increasing the number of cores. This change is driven by both hardware limitations and architectural challenges. Understanding the reasons behind this shift is crucial for anyone interested in computer architecture and performance optimization.

Limitations of Increasing Clock Speed

One of the most significant limitations of increasing clock speeds lies in the substantial power requirements and the accompanying cooling issues. Processors require a high amount of energy, which leads to increased heat generation. For many modern computers, especially laptops, the form factor restricts the space available for large heat sinks and cooling fans, which are essential for high-performance desktop towers. This limitation becomes even more critical as clock speeds increase, making it difficult to sustain high performance without compromising the user experience.

From Clock Frequency to Core Count: A Scalable Solution

Since the early 2000s, the trend has shifted towards increasing the number of cores rather than boosting the clock frequency. This change is more scalable and less power-intensive. However, simply adding cores does not automatically translate to a linear increase in performance. There are several architectural issues that need to be addressed to fully leverage the benefits of multiple cores.

Cache Coherency and Performance Impact

Each core on a processor has its own unique Level 2 (L2) cache, and sometimes a unique Level 3 (L3) cache, which can be shared. The primary purpose of these caches is to keep the core from having to access slower main memory for each instruction. However, with unique L2 and L3 caches, there is a risk that the same memory addresses may be loaded in multiple cores. When one core makes a change, it needs to signal other cores to invalidate their cached data and reload, causing a significant performance impact.

This phenomenon, known as cache coherency, adds additional circuitry that must be synchronized. For instance, if you have two cores, the circuitry required for cache coherency is essential. However, as the number of cores increases, the amount of required circuitry grows exponentially. For four cores, you need to manage the cache coherency for all combinations, duplicating the effort for each additional core.

Performance Calculation and Real-World Impact

Real-world performance gains can be quite different from theoretical expectations. According to empirical data, adding cores to a processor generally yields the following performance increases:

2 cores: Perform about 1.8x the work 4 cores: Perform about 2.6x to 2.8x the work 8 cores: Perform about 3.6x the work

These performance increases are fundamentally impacted by the need to maintain cache coherency and the overhead of the operating system (OS) in managing memory synchronization. As the number of cores increases, the OS needs to schedule and manage multiple cores, which can lead to additional delays and overhead.

Moore's Law and Core Count

The trend towards increasing core count is further reinforced by Moore's Law, which states that the number of transistors in a dense circuit doubles approximately every two years. This has led to a doubling in the number of cores whenever the transistor density doubles. Although Moore's Law has been a driving force in the evolution of processor architecture, it has slowed down since 2010 due to practical limitations in manufacturing and power consumption.

The Power Wall and Thermal Limitations

Another key factor is the Power Wall, which is the point where increasing clock speeds becomes impractical due to power density and thermal issues. Since 2005, the increase in clock frequency has plateaued due to these limitations. This means that while adding more cores can provide a meaningful performance boost, the benefits are tempered by the need to manage the additional circuitry required for cache coherency and the OS overhead.

Conclusion

In summary, the trend towards increasing the number of cores in processors rather than boosting clock speeds is a strategic choice driven by both hardware constraints and the need for more efficient and sustainable computing. The architecture and management of cache coherency play a significant role in determining the real-world performance of multi-core processors. While adding more cores can lead to substantial performance improvements, the ongoing challenges in cache coherency and OS management ensure that the benefits are not always linear.